Journal of Economic Literature 2014, 52(4), 1075–1118 http://dx.doi.org/10.1257/jel.52.4.1075

Behavioral Theory †

Botond Ko˝ szegi*

This review provides a critical survey of psychology-and- (“behavioral- economics”) research in contract theory. First, I introduce the theories of individual decision making most frequently used in behavioral contract theory, and formally illustrate some of their implications in contracting settings. Second, I provide a more comprehensive (but informal) survey of the psychology-and-economics work on classical contract-theoretic topics: , , , and incomplete . I also summarize research on a new topic spawned by psychology and economics, exploitative contracting, that studies contracts designed primarily to take advantage of agent mistakes. ( JEL A12, D03, D82, D86)

1. Introduction is in the process of full integration into eco- nomic analysis: researchers are using the sychology and economics—also known new psychologically based models to study Pas —is a mind- how individual behavior plays out in organi- set for doing economics that espouses the zations and markets, what the welfare conse- importance of thinking about the psycho- quences are, and how policy should respond logical accuracy of models. After a his- to outcomes—questions in which tory of identifying deviations from classical have always been interested. approaches, modeling these deviations for- This review summarizes and organizes mally, and empirically establishing their one important part of the above develop- importance in economic decisions, the field ment, the rapidly growing literature on behavioral contract theory. In a well-known textbook less than a decade old, Bolton and * Central European University. I am grateful to Paul Heidhues and Klaus Schmidt for insightful discussions Dewatripont (2005, p. 10) wrote that “even and detailed comments on drafts of this review. I am also though there is by now a large literature hugely indebted to Matthew Rabin for countless conver- exploring a wide range of alternative models sations over the years that helped clarify my thinking on many of the issues covered in this paper. I thank Mark of individual choice . . . there have been rela- Armstrong, Petra Bischofberger, Andrea Canidio, Takeshi tively few explorations of the implications for Murooka, Yuval Salant, Patrick Schmitz, Ferdinand von optimal contracting.” By the time of Bolton Siemens, Rani Spiegler, and four anonymous referees for comments, and the European Research Council for sup- and Dewatripont’s book, contract theory had port under Starting Grant #313341. Kinga Marczell pro- evolved to a curious point: while research- vided excellent research assistance, including drafting the ers had explored optimal contracting in first version of the auction-theory parts of section 5. † Go to http://dx.doi.org/10.1257/jel.52.4.1075 to visit many complex economic environments, they the article page and view author disclosure statement(s). largely followed a simplistic approach for

1075 1076 Journal of Economic Literature, Vol. LII (December 2014) modeling agent behavior. In a moral-hazard to contracts, mechanisms, or other extended context, for instance, an agent has an outside interactions. This excludes purely game- option, off the monetary benefits and theoretic issues, such as the winner’s curse the costs of effort, and optimally chooses and overbidding in auctions (e.g., Eyster and whether to participate and what to do. Rabin 2005), that are studied in behavioral In contrast, recent research has fruitfully ; and pure pricing issues, such incorporated behavioral-economics ideas into as oligopolistic pricing with consumer loss most of the classical contract-theoretic top- aversion (e.g., Heidhues and Ko˝szegi 2008) ics, including moral hazard, screening, auction that are studied in behavioral industrial theory, and . In addition, organization.2 the recognition that agents often fail to act in I begin in section 2 by presenting the the- their best —and that principals might ories of individual decision making in psy- know this better than agents do—has inspired chology and economics that are most used in a novel literature on exploitative contracts, behavioral contract theory, and by working contracts whose exclusive or primary aim is to out some of their implications in simple set- take advantage of agent mistakes. For each of tings. These theories are (i) loss aversion— these topics, I attempt to organize the main whereby an individual evaluates economic insights researchers have uncovered, as well as outcomes relative to reference points, and point out some of the many additional ques- weights losses more heavily than similar- tions raised by existing work. sized gains; (ii) present bias—whereby an Like in any review, it is necessary to draw individual weights an earlier relative to a some, partly arbitrary, lines regarding which delayed outcome more heavily when the papers are included. I chose papers that earlier outcome is in the present; (iii) ineq- are “behavioral” in the sense that a formal, uity aversion—whereby people in a social relatively general model with psychologi- situation dislike advantageous and especially cal foundations or compelling psychologi- disadvantageous inequality in material out- cal interpretation is involved, and for which comes; and (iv) overconfidence—whereby some supports the specific decision- a person displays unrealistically positive making process. This excludes, for instance, beliefs regarding her ability or prospects. In the very interesting literature on bounded almost all applications, researchers assume rationality predicated on optimization with that the agent (she) behaves according to cognitive, effort, or information-acquisi- one psychologically based model, while the tion costs (e.g., Tirole 2009 and Bolton and principal (he) is fully rational and has a clas- Faure-Grimaud 2010), which does not yet sical goal (usually maximization). seem to be based solidly on psychology inter- The main sections of the paper cover work pretation and evidence.1 Further, I focus on from psychology and economics on the main papers that address design questions related contract-theoretic topics. Deviating from reviews by Rabin (1998) and DellaVigna (2009), I organize the discussion primarily­ 1 Many papers in this literature, including those cited above, uncover mechanisms that intuitions suggest would hold and continue to be important in more psychologically 2 See Spiegler (2011) for an excellent textbook cover- based models. In other cases, however, the main results age of many industrial-organization models. A substantial are based on essentially neoclassical optimization princi- amount of research is either at the intersection of behav- ples with novel types of presumed costs. Garicano and Prat ioral and behavioral contract theory, (2013) review some of this work. How the insights in this or is not clearly categorizable, and is hence covered in both literature connect to those based more closely on psychol- this survey and Spiegler’s textbook; see especially chapters ogy evidence is a promising topic for future research. 2–5 and 11 of the book. Ko˝szegi: Behavioral Contract Theory 1077 not around behavioral-economics phenom- Section 5 turns to a less extensive literature ena—whether the underlying model of deci- in behavioral contract theory, mechanism sion making involves loss aversion, hyperbolic design—where a principal is interacting with discounting, etc.—but rather according to multiple agents whose private information contract-theoretic principles—whether the he does not know—and its particular appli- interaction qualifies as moral hazard, screen- cation, auction theory. Many of the papers in ing, etc. I do so for two main reasons. First, this area study the implications of reference- this organization reflects the fact that the dependent and loss aversion for auc- literature has been addressing some of the tions. The main results can be categorized classical contract-theoretic topics that inter- around two forces. First, because loss-averse est other economists. Second, for some top- bidders dislike risk in how much they will ics, the precise psychological source of the pay, an auction designer has an incentive phenomena in question is not important to choose auctions that insure them from (although for other questions it is). this risk. Second, a designer may be able to In section 3, I discuss work on moral haz- manipulate bidders’ reference points to his ard, where the principal’s main goal in con- advantage. tracting is to provide incentives for the agent Section 6 considers findings in exploitative to exert costly effort. Much of this literature contracting, the study of contract designs focuses on sources of nonfinancial motiva- whose central consideration is exploiting an tion that affect the agent’s willingness to agent’s mistakes. In most of these models, work hard. As the most important example, the agent is a consumer who either does not researchers recognize—following original fully understand all features (e.g., all discussions by Akerlof (1982) and Akerlof and fees) of a product or mispredicts her own and Yellen (1988, 1990)—that organizations behavior with respect to the product, and the are often social in nature, so that agents’ principal is a profit-maximizing firm aware of attitudes toward others have an important the consumer’s tendencies. This literature has impact on behavior. Researchers study the reinvigorated the long-recognized but under- nature of these and other nonpecuniary studied topic of contracting with noncommon­ motivations, as well as their interaction with priors, but because of its conceptualization financial incentives. of noncommon priors as resulting from con- In section 4, I summarize existing research sumer mistakes, its focus on specific forms on asymmetric information and screening, of mistakes, and its resulting ability to make where the principal is attempting to inter- welfare statements, it is usefully categorized act with an agent whose private information as a new topic in contract theory. he does not know. Psychology and econom- In section 7, I discuss two nascent areas ics introduces a number of important novel of research: incomplete contracts and envi- themes, including a new reason for screen- ronment design. In the existing research ing contracts: present-biased agents prefer on incomplete contracts, authors assume to write contracts to constrain their own that contracts change preferences by affect- future behavior, while knowing that infor- ing feelings of entitlement or the reference mation pertaining to optimal behavior will point to which parties later compare out- arrive in the future. In addition, the psycho- comes, and study the implications for opti- logical phenomena emphasized in behavioral mal contracting and the optimal allocation of economics also lead to the natural screening rights. Finally, section 8 provides a issue of how to deal with agents who exhibit few general thoughts on the state of the lit- these phenomena to different extents. erature and the challenges ahead for it. 1078 Journal of Economic Literature, Vol. LII (December 2014)

2. Key Theories in Psychology where 0, 1. The first term in the η > λ > and Economics utility function is consumption utility (lin- ear in this example), which corresponds to This section summarizes the main theories the conventional notion of outcome-based of individual decision making in psychology utility. The second term is gain–loss utility, and economics that are used in the applica- which captures the agent’s sensations of gain tions below. Rabin (1998) and DellaVigna or loss relative to the reference point. The (2009) provide excellent reviews of the evi- parameter is the degree of the agent’s loss λ dence for these models.3 To simultaneously aversion—with 1 capturing that losses λ > introduce the model of human behavior, give are more painful than gains are pleasant— a sense of how to work with it, as well as to and is the weight on gain–loss utility. If η illustrate some of its key economic implica- there are other outcomes for which the agent tions, I present each theory together with an evaluates gains and losses separately—for application covered in this review. For brev- instance, the object in an auction—a similar ity, I do not carefully introduce all necessary additively separable utility function can be assumptions in each example, nor present used to capture reference-dependent utility formal proofs. in those dimensions. A crucial issue in using loss-averse prefer- 2.1 Loss Aversion ences is the determination of the reference Model. As has been demonstrated in a pow- point r. The most common approach in the erful body of work starting with Kahneman and applications below is to set the reference Tversky (1979) and Tversky and Kahneman point equal to the agent’s rational expec- (1991), individuals evaluate economic out- tations as defined by her full probabilistic comes not just according to an absolute valu- beliefs (Ko˝szegi and Rabin 2006, 2007). It ation attached to the outcomes in question, is then necessary to extend the above utility but also relative to subjective reference points. function to allow for the reference point to The most important property of such refer- be a F(r): ence-dependent preferences is loss aversion: losses relative to a reference point are more (2) U(w F) ​ ​ ​u(c r) dF(r). | = | painful than equal-sized gains are pleasant. ∫ I present a formal model based on Ko˝szegi This formulation captures the notion that the and Rabin (2006, 2007), assuming here that sense of gain or loss from a given consump- the decision concerns only monetary out- tion outcome derives from comparing it to all comes such as a . If the agent’s wage outcomes possible under the reference lot- is w and her reference point is r, then her tery. For example, if the reference lottery is reference-dependent utility is a gamble between $0 and $100, an outcome of $50 feels like a gain relative to $0 and like (1) u(w r) a loss relative to $100, and the overall sensa- | w (w r) if w r, and tion is a mixture of these two feelings. + η − ​ ​ ≥ ​ = ​ w​ (w r) if w r, { + ηλ − < Application. I illustrate some implications of the model in the context of wage setting, 3 Note, however, that the ability of these models to beginning with a property of risk attitudes explain important economic phenomena as explained in that reappears in multiple applications. this review can—to the extent that other models cannot Suppose ​w​ ​ w​ ​ ​w​ ​, and let F denote as convincingly account for the same phenomena—be L ≤ ​ M ≤ ​ H thought of as additional evidence for the models. the lottery that pays ​w​ L​ with probability Ko˝szegi: Behavioral Contract Theory 1079

​p​ L​, ​w​ M​ with probability ​p​ M​, and ​w​ H​ with In particular, the agent greatly values con- probability ​p​ H​. What is the agent’s expected tracts that insure her from risk on any part of utility from F? An agent with rational expec- the distribution of outcomes. These proper- tations correctly anticipates the distribution ties of the agent’s aversion to risk are distinct of outcomes, so her reference point will be from the generated by classical F. Equations (1) and (2) then imply: expected utility over wealth, leading to many qualitatively different predictions in eco- U(​w​ ​ F) w​ ​ ​p​ ​ (​w​ ​ ​w​ ​) nomic environments. L | = ​ L − M ηλ M − L As an example of the implications for con- ​p​ ​ (​w​ ​ ​w​ ​ tracting, I illustrate a loss-aversion-based − ​ H ηλ H − L) explanation for “bonus” contracts—binary U(​w​ ​ F) w​ ​ ​p​ ​ (​w​ ​ ​w​ ​) payment schemes in which the employee M | = ​ M + L η M − L has a base salary, and, depending on whether ​p​ ​ (​w​ ​ ​w​ ​ her performance exceeds a threshold, may − ​ H ηλ H − M) in addition receive a bonus—by Herweg, U(​w​ ​ F) w​ ​ ​p​ ​ (​w​ ​ ​w​ ​) Müller, and Weinschenk (2010). Suppose H | = ​ H + L η H − L an agent chooses between low effort (​e​ L​) ​p​ ​ (​w​ ​ ​w​ ​ and high effort (​e​ ​), where the cost of effort + ​ M η H − M) H ​e​ ​, i L, H , is ​c​ ​. The agent’s output takes i ∈ { } i Hence, expected utility is one of three values; low, medium, or high. The following table gives the probabilities (3) U(F F) for the three levels of output occurring as a | function of effort: p​ ​ U(​w​ ​ F) p​ ​ U(​w​ ​ F) p​ ​ U(​w​ ​ F) = ​ L L | + ​ M M | + ​ H H | low medium high p​ ​ ​w​ ​ p​ ​ ​w​ ​ p​ ​ ​w​ ​ output output output = ​ L L + ​ M M + ​ H H ​e​ L​ 2/3 1/3 0 ( 1) ​p​ ​ ​p​ ​(​w​ ​ ​w​ ​)​ − η λ − [ L M M − L ​e​ H​ 1/3 1/3 1/3 p​ ​ ​p​ ​(​w​ ​ ​w​ ​) + ​ L H H − L p​ ​ ​p​ ​(​w​ ​ ​w​ ​)]. + ​ M H H − M I assume that parameters are such that the 8 principal would like the agent to exert the 0 unless ​w​ ​ w​ ​ w​ ​ < L = ​ M = ​ H high level of effort. Then, if the agent is averse to risk, the informativeness principle The agent’s expected gain–loss utility is in the classical model of moral hazard pre- always negative for risky outcomes, making dicts that the principal pays different her averse to risk. Intuitively, for instance, for different levels of output (Holmstrom when ​w​ ​ w​ ​, the agent experiences ​w​ ​ 1979). Intuitively, the three levels of out- H > ​ M M as a loss relative to ​w​ H​, and ​w​ H​ as a gain rela- put are differently informative regarding tive to ​w​ M​; but since the sensation of loss whether the agent exerted high effort, and outweighs the sensation of gain, her average hence should be rewarded differently. gain–loss sensation is negative. Furthermore, In contrast, the same is not optimal for the agent is first-order averse to risk: her util- a loss-averse agent (even though she is ity decreases linearly with the dispersion in averse to risk). Let the wage paid for low, outcomes, for instance linearly in ​w​ ​ ​w​ ​. medium, and high output be ​w​ ​, ​w​ ​,​ and ​ H − M L M 1080 Journal of Economic Literature, Vol. LII (December 2014) w​ ​, respectively. Clearly, the principal wants (IC) __​ 1 ​ (​w​ ​ w​ ​) H 3 H − ​ L ​w​ ​ ​w​ ​ ​w​ ​. Using equation (3), the H ≥ M ≥ L principal’s problem is4 __​ 2 ​ ​ ( 1)(​w​ ​ w​ ​) ​c​ ​ c​ ​ − 9 η λ − ​ H − ​ M ≥ H − ​ L min __​ 1 ​ (​w​ ​ w​ ​ w​ ​) 3 L + ​ M + ​ H Now I show that ​w​ ​ w​ ​ w​ ​ is not opti- H > ​ M > ​ L (PC) s.t. __​ 1 ​ (​w​ ​ w​ ​ w​ ​) mal by identifying a better contract if this is 3 L + ​ M + ​ H the case. First, increase ​w​ ​ by and decrease​ M ϵ ​ __1 ​ ​ ( 1)(​w​ ​ w​ ​ w​ ​ w​ ​ and ​w​ ​ by 2. As a result of this, profits − 9 η λ − H − ​ M + ​ M H L ϵ/ are unchanged, PC is still satisfied, and IC ​w​ ​ w​ ​ ​w​ ​) becomes slack. Then, decrease ​w​ ​ by and − L + ​ H − L H ϵ′ increase ​w​ ​ by . This again leaves profits L ϵ′ ​c​ ​ u unchanged, and for a sufficiently small , IC − H ≥ ϵ′ remains slack, while now PC becomes slack (IC) __​ 1 ​ (​w​ ​ w​ ​ w​ ​) as well. With both constraints slack, the prin- 3 L + ​ M + ​ H cipal can decrease all three wage levels by __​ 1 ​ ​ ( 1)(​w​ ​ w​ ​ w​ ​ a sufficiently small , increasing profits and − 9 η λ − H − ​ M + ​ M ϵ″ still satisfying both constraints. ​w​ ​ w​ ​ ​w​ ​) ​c​ ​ The above argument implies that the − L + ​ H − L − H principal chooses at most two different ​ __2 ​ ​w​ ​ __1 ​ ​w​ ​ wage ­levels. Clearly, one wage level (i.e., ≥ ​3 L + ​ 3 M a constant wage) is not incentive compat- __​ 2 ​ ​ ( 1)(​w​ ​ w​ ​) ​c​ ​ ible, so in the optimal contract the principal − 9 η λ − ​ M − ​ L − L uses exactly two wage levels. Hence, either ​w​ ​ w​ ​ w​ ​ or ​w​ ​ w​ ​ w​ ​. Notice L = ​ M < ​ H L < ​ M = ​ H The principal minimizes the expected wage that the above improvement applies when​ paid to the agent when she exerts high effort, w​ ​ w​ ​ w​ ​, so this is not optimal either. L = ​ M < ​ H subject to two constraints. The participa- Hence, in the optimal contract ​w​ ​ w​ ​ L < ​ M tion constraint (PC) captures that the agent w​ ​. = ​ H must be willing to take the contract over Intuitively, because a loss-averse agent her best alternative, which has utility u. The strongly dislikes random variation in the incentive-compatibility constraint (IC) cap- wage due to in the environment, tures that the agent must be willing to exert the principal has an incentive not to vary high effort over low effort. Rewriting the wages too finely with output. At the same constraints: time, he also needs to provide incentives, so the agent’s wage cannot be completely flat. (PC) __​ 1 ​ (​w​ ​ w​ ​ w​ ​) As a result, the principal chooses the mini- 3 L + ​ M + ​ H mal amount of wage variation that can still __​ 2 ​ ​ ( 1)(​w​ ​ w​ ​) ​c​ ​ u provide incentives: two wage levels, that is, a − 9 η λ − ​ H − ​ L ≥ H + bonus contract.

4 Technically, in this example there are two dimensions 2.2 Present Bias and Time Inconsistency of utility, and effort, so one must also account for gain–loss utility in effort. In this case, however, gain–loss Model. In intertemporal decisions, many utility in effort is zero, so that total utility in effort is equal individuals the future significantly less to consumption utility: when the agent chooses effort i, with both her reference point and her than the present, yet place similar weights on outcome in the effort–cost dimension are ​c​ i​. different points in time in the future. This Ko˝szegi: Behavioral Contract Theory 1081 implies that a person would not Application. I present a model of credit like to make into the future contracts based on Heidhues and Ko˝szegi today, but would like to make similar invest- (2010) and Eliaz and Spiegler (2006). ments in the future. Time inconsistency—a Suppose a consumer with 1/2 and 1 β = δ = conflict between the person’s preferences is contracting in period 0 with competitive at different points in time—arises because suppliers of credit, who have funds available once the future comes, she would again pre- to them at an rate of zero. The con- fer not to make immediate investments. sumer borrows for future consumption and Laibson (1997) and O’Donoghue and repays her loan in periods 1 and 2. A credit Rabin (1999a) formalize the above “pres- contract consists of a borrowed amount c and ent bias” using the model. They a menu of installment plans ​r​ ​ 0, ​r​ ​ 0 β − δ 1 ≥ 2 ≥ assume that self t (the decisionmaker’s according to which the consumer can repay period-t incarnation) evaluates the stream her loan in periods 1 and 2. Once the con- ​u​ t​, ​u​ t 1​, ​u​ t 2​, ​u​ t 3​, … of instantaneous utili- sumer signs a contract with a firm, she can- + + + ties at times t, t 1, t 2, t 3, … as not borrow from other firms, but she can + + + decide in period 1 which of the installment 2 3 ​u​ t​ u​ t 1​ ​ ​ ​u​ t 2​ ​ ​ ​u​ t 3​ , plans designated in her contract to follow. + βδ​ + + β​δ + + β​δ + + ⋯ The consumer’s instantaneous utility from where , 1 are the discount factors that consumption is u(c), whereas her instanta- β δ ≤ parameterize present bias. In this formula- neous disutility from making a payment is tion, the discount factor between periods t r. Hence, since borrowing is for future con- and t 1 is , whereas the discount factor sumption, the consumer’s utility in period 0 + βδ between any two consecutive future periods is u(c) ​r​ ​ ​r​ ​; but in period 1, she chooses − 1 − 2 is . For 1, therefore, self t discounts the installment plan to follow minimizing ​ δ β < between the present and the future more r​ ​ r​ ​ 2. 1 + ​ 2/ heavily than between different periods in To find the optimal contract, consider first the future, capturing the above pattern of fully sophisticated consumers, for whom ​ intertemporal preferences. For 1, the ̂​ . Since a sophisticated consumer β = β = β model reduces to the exponential discount- knows how she will behave, without loss of ing model commonly used in economics. generality we can suppose that the menu of A key consideration in applications of the installment plans specifies only the one plan​ model is whether the agent correctly r​ ​, ​r​ ​ she will choose. Hence, a firm’s prob- β − δ 1 2 anticipates the present bias that she will lem is (but would prefer not to) have in the future. Following O’Donoghue and Rabin (2001), max ​r​ ​ r​ ​ c 1 + ​ 2 − I suppose that the agent believes that her future will be ​ ̂​. The assumption ̂ PC s.t. u(c) ​r​ ​ ​r​ ​ u, β β β = ​β​ ( ) − 1 − 2 ≥ corresponds to full sophistication regarding time inconsistency—whereby the agent per- where u is the consumer’s utility from her fectly understands that she will behave dif- outside option, and the constraint (the “par- ferently than she would now like—whereas ​ ticipation constraint” or PC) captures that ̂​ 1 corresponds to full naïveté—whereby the consumer must be willing to accept the β = the agent completely ignores her time principal’s contract over her best alterna- inconsistency in preferences. Intermediate tive. Substituting the binding PC into the values of ​ ̂​ capture intermediate values of maximand produces the first-order condition β naïvete. u (c) 1, which means that the consumer’s­ ′ = 1082 Journal of Economic Literature, Vol. LII (December 2014)

consumption is efficient. This simple analy- The solution to this problem is (​r ​̂1​, ​r ​̂2​) sis leads to an important point made by (u u(c), 0) and (​r​ ​, ​r​ ​) (0, 2(u u(c))).5 = − 1 2 = − DellaVigna and Malmendier (2004). Even Given that in a competitive equilibrium though the consumer suffers from a time- firms make zero profits, the equilibrium inconsistency problem, the competitive- payments are (​r ​​, ​r ​​) (c 2, 0) and (​r​ ​, ​r​ ​) ̂1 ̂2 = / 1 2 equilibrium contract maximizes her utility, (0, c). In practice, this type of contract = as it leads her to borrow the optimal amount. would correspond to a credit arrangement in Hence, with sophisticated consumers, mar- which, as for instance with many credit cards kets can solve time-inconsistency problems. and ­nontraditional mortgages, the consumer The situation is entirely different for a is asked to repay her loan fast, and delaying ­nonsophisticated borrower, for whom ​ ̂​ . repayment carries large penalties. β > β To see this, suppose first that c is fixed, so The optimal contract, therefore, induces that the contract concerns only the repay- false beliefs in the consumer regarding how ment terms. For any contract the consumer she will repay, making the contract look more is offered, there is a “baseline” installment attractive than it really is. In fact, the contract plan (​r ​̂1 ​ , ​r ​̂2 ​ ) she believes in period 0 she will maximizes the consumer’s mistake: by asking follow, and a possibly different, “alterna- her to repay only in period 1, the lender max- tive” installment plan (​r​ 1​, ​r​ 2​) she will actu- imizes the effect of the consumer’s mistaken ally follow. Without loss of generality, we can beliefs regarding her future willingness to assume that the contract contains no other pay to delay repayment. This first implica- repayment options. The installment plans tion of the example, whereby consumers are

(​r​ 1​, ​r​ 2​) and (​r ​̂1 ​ , ​r ​̂2 ​ ) must solve led to buy seemingly cheap products, is a general implication of exploitative contract- max ​r​ ​ r​ ​ c ing that appears in many papers discussed in 1 + ​ 2 − section 6. The fact that consumers are sold seem- (PC) s.t. u(c) ​r​ ̂1​ ​r​ ̂2​ u − − ≥ ingly cheap products can have many types of adverse welfare consequences. In this (PCC) ​r​ ̂1​ ​ ̂​ ​r​ ̂2​ ​r​ 1​ ̂r​ 2​ + ​β ≤ + ​β​ ​ particular case, a welfare cost arises if the borrowed amount c is endogenous. Since (IC) ​r​ 1​ r​ 2​ 2 ​r​ ̂1​ r​ ̂2​ 2 + ​ / ≤ + ​ / the consumer believes that borrowing c costs only c 2, the competitive equilibrium with / As explained by Heidhues and Ko˝szegi endogenous c has u (c) 1 2—the con- ′ = / (2010), the firm faces the following con- sumer overborrows relative to the optimum. straints in designing its contract. First, for Note that since the borrowing is for future the borrower to be willing to accept the consumption, the overborrowing is not due firm’s offer, self 0’s utility given how she to the borrower’s time inconsistency per se, believes she will behave must be at least u but due to her false beliefs about her future (PC). Second, if self 0 is to think that she will repayment behavior. choose the baseline option, then given her beliefs ​ ̂​ she must think she will prefer it to β the alternative option (perceived-choice con- 5 For any (​r ​​, ​r ​​), (​r​ ​, ​r​ ​) (0, 2 ​r ​​, ​r ​​) satisfies both IC ̂1 ̂2 1 2 = ̂1 ̂2 straint or PCC). Third, if self 1 is to actually and PCC (the latter constraint is slack), and no other (​r​ 1​, ​r​ 2​) choose the alternative repayment schedule, that satisfies IC achieves higher profits, so it is the optimal she has to prefer it to the baseline repayment (​r​ 1​, ​r​ 2​). Then, among (​r ​̂1​, ​r ​̂2​) that satisfy PC, the option that maximizes the principal’s resulting profits, 2 ​r ​̂1​, ​r ​̂2​, schedule (IC). has ​r ​​ 0. ̂2 = Ko˝szegi: Behavioral Contract Theory 1083

All of the above holds for any ​ ̂​ — where ​ ​ ​ satisfying 0 ​ ​ ​ 1 parameter- β > β βi ≤ βi < that is, for even arbitrarily small degrees of izes the agent’s aversion to being ahead and​ naïveté regarding time inconsistency. This ​ ​ 0 parameterizes her aversion to being αi ≥ leads to the second important implication behind. of the example: with time inconsistency, even small amounts of naïveté often have Application. I illustrate the implications of large welfare implications. Heidhues and inequity aversion for incentive contracts. Ko˝szegi (2010) show that simple restrictions First, consider a “gift-exchange” situation, on the contract form can drastically increase in which the principal pays the agent a fixed welfare. As an illustration, suppose that we output-independent wage w 0, and then ≥ impose a sufficiently low constraint p on how the agent chooses effort level e 0. The ≥ much the consumer can be penalized for fail- agent’s material cost of effort is c(e), and this ing to choose the lowest-cost installment plan produces output e for the principal. Hence, in her contract. Then, in one competitive- the agent’s material payoff is ​x​ ​ w c(e), 2 = − equilibrium contract for a fixed consump- whereas the principal’s is ​x​ ​ e w, and 1 = − tion level c, the consumer chooses between the effort level that maximizes social welfare installment plans (c p, 0) and (0, c).6 Since satisfies c (e) 1. Note that if the principal − ′ = the consumer thinks the cost of borrowing c and the agent were selfish, the agent would is c p, the competitive-equilibrium level of exert zero effort for any wage, and hence the − borrowing becomes u (c) 1. Hence, with principal would pay a wage of zero. ′ = this the consumer borrows the As has long been recognized in the litera- optimal amount, and—although she falsely ture, however, inequity aversion can lead to believes she will get a better deal—ulti- a higher wage and a higher level of effort, mately repays exactly that amount. potentially increasing material welfare for both the principal and the agent. Suppose 2.3 Inequity Aversion that the principal and the agent are inequity Model. To explain a number of disparate averse as defined above,N 2, the princi- = facts regarding prosocial behavior, Fehr and pal is player 1 and the agent is player 2, and Schmidt (1999) present a model in which ​ ​ ​ ​ ​ ​ 1 2. We first solve for how the αi ≥ βi > / individuals dislike both advantageous and agent reacts to a wage w 0, assuming that > disadvantageous inequality in material out- the optimal e has c (e) 1 (the analysis below ′ ≤ comes, but dislike being behind more than makes clear that this is the case for the prin- they dislike being ahead. Suppose that there cipal’s optimal contract). Clearly, the agent are N individuals in agent i’s “reference never chooses an effort level that leads to group”—the group of individuals (including ​x​ ​ x​ ​; relative to ​x​ ​ x​ ​, this would lead 1 > ​ 2 1 = ​ 2 herself) whose material outcomes she cares to a lower material payoff for her and gen- about. Then, her utility from material pay- erate further disutility from disadvantageous offs ​x​ 1​, … , ​x​ N​ is inequality. Hence, the agent maximizes ​x​ ​ ​ ​ ​(​x​ ​ ​x​ ​) w(1 2​ ​ ​) ​ ​ e 2 − β 2 2 − 1 = − β 2 + ​β 2 − (4) (1 ​ ​ ​)c(e) subject to the constraint ​x​ ​ ​x​ ​, − β2 2 ≥ 1 1 or w c(e) e w. Now it is easy to see x​ ​ ​ ​ ______​ ​ ​ ​ ​ ​ (​x​ ​ ​x​ ​) ​ ​ ​ ​ (​x​ ​ ​x​ ​) ​, i − αi ∑ j − i + βi ∑ i − j − ≥ − N 1 j;​x​ ​ ​x​ ​ ​ j;​x​ j​ ​x​ i​ that w c(e) e w cannot be optimal: if − ( j≥ i < ) − > − this were the case, then (using that ​ ​ 2​ 1 2 6 In this case, there are other competitive-equilibrium β > / and therefore ​ ​ 2​ 1 ​ ​ 2​, and c (e) 1) contracts: for instance, if ​r ​ ​ p and ​r ​ ​ r ​ ​ c p, then β > − β ′ ≤ ̂1 ≥ ̂1 + ​ ̂2 = − (​r ​​, ​r ​​) and (​r​ ​, ​r​ ​) (​r ​ ​ p, ​r ​ ​ 2p) is a competitive the agent would want to increase e. As a ̂1 ̂2 1 2 = ̂1 − ̂2 + equilibrium. result, the agent’s inequity aversion leads 1084 Journal of Economic Literature, Vol. LII (December 2014) her to choose her level of effort to equalize inequity aversion to reward the agent’s material payoffs, setting w c(e) e w, of a high effort. − = − or w (e c(e)) 2. As the principal’s bonus-setting = + / Given that the agent’s behavior eliminates behavior results in no inequality, the inequality, the principal maximizes material agent maximizes material payoffs payoffs, and her problem can therefore be b c(e) (e c(e)) 2 c(e), which again − = + / − written as yields c (e) 1. Hence, the voluntary-bonus ′ = arrangement also maximizes social surplus. (5) max e w The intuition is simple: knowing that the − principal will split the social surplus e c(e) − e c(e) equally, the agent has an incentive to maxi- s.t. w ______+ ​. = ​ 2 mize social surplus. The above analysis, however, provides Substituting the constraint into the maxi- a simplistic view of the power of inequity mand yields that the principal is maximizing aversion to generate efficient outcomes. (e c(e)) 2, leading to c (e) 1. This means Motivated by Fehr et al. (2007), suppose − / ′ = that inequity aversion generates an efficient that a proportion q of individuals in soci- outcome: it induces an effort level that maxi- ety are inequity averse as above, but a pro- mizes material efficiency, and since this goes portion 1 q are selfish. In that case, the − along with the lack of inequality, it is also gift-exchange and bonus arrangements are efficient in terms of inequity-averse prefer- neither optimal nor equivalent. To illustrate, ences. To understand the intuition, note that I suppose for simplicity that ​ ​ ​ , so that αi = ∞ the agent’s inequity aversion acts as an incen- inequity-averse individuals do not tolerate tive device: since a higher wage means that a disadvantageous inequality. higher effort is necessary to split the surplus Consider first a gift-exchange arrange- equally, the agent responds to a higher wage ment. An inequity-averse agent responds by increasing effort. Knowing this and know- to a positive wage as above, but a selfish ing that the agent will give him exactly half of agent chooses zero effort. Since a positive the social surplus, the principal has an incen- wage followed by zero effort generates dis- tive to induce the effort level that maximizes advantageous inequality for the principal, social surplus. an inequity-averse­ principal—afraid to run An alternative to gift exchange is the “vol- into a selfish agent—never chooses a positive untary-bonus” system, in which the agent first wage. Denoting an inequity-averse agent’s chooses her level of effort, and then the prin- effort as a function of the wage by e(w), cipal chooses a bonus b 0 to reward her. the selfish principal chooses w to maximize ≥ Similarly to how the agent chooses her level qe(w) w, yielding the first-order condition − of effort in the case of gift exchange above, qe (w) 1. Totally differentiating equation ′ = the principal chooses the bonus to equalize (5) with respect to w, solving for e (w), and ′ material payoffs, setting b c(e) e b, or plugging into the first-order condition gives − = − b (e c(e)) 1. Again, therefore, inequity = + / aversion creates an incentive device, but it c (e) 2q 1. ′ = − does so in a different way: while the gift- exchange arrangement relies on the inequity Hence, effort is lower than optimal for any aversion of the agent to reward the princi- q 1. < pal’s investment of a high wage, the volun- Now I return to the voluntary-bonus tary-bonus system relies on the ­principal’s arrangement. An inequity-averse agent Ko˝szegi: Behavioral Contract Theory 1085 would find it intolerable to exert positive in either case the investing party gets half effort and then not be rewarded by a selfish of the increase in the social surplus. When principal, so she exerts zero effort. Recalling there is a chance that the investment will not that an inequity-averse principal chooses be rewarded, however, investment is lower b (e c(e)) 2, a selfish agent maximizes when its is higher. = + / q(e c(e)) 2 c(e), which gives the first- + / − 2.4 Quasi-Bayesian Models order condition q Framework. The literature has docu- c (e) _____ ​. mented a number of systematic mistakes ′ = ​ 2 q − individuals commit when thinking about Again, effort is lower than optimal for any statistical properties of random phenom- q 1. The intuition for why effort is lower ena. A commonly used reduced-form way to < than optimal in the presence of selfish indi- capture such mistakes is the quasi-Bayesian viduals under either arrangement is simple: approach: one posits that the agent updates since a party’s attempt to increase the sur- her beliefs fundamentally in a Bayesian way, plus might go unrewarded, she has less of an but commits a particular error that is incon- incentive to increase the social surplus. sistent with rational inference. These models But while both arrangements generate typically also assume that the agent does not lower-than-optimal effort, they do not gen- learn about her error from her observations. erate the same effort. Simple arithmetic The quasi-Bayesian approach is not a theory, shows that q (2 q) 2q 1, so effort is but merely a framework for thinking about / − > − higher under the bonus arrangement than statistical mistakes. A fully-fledged theory under gift exchange. Furthermore, while specifies the particular mistake the agent the bonus system generates effort whenever is committing. In the models reviewed in the agent is selfish, the gift-exchange system this article, there are two general mistakes generates effort only if the principal is self- authors assume: systematically incorrect pri- ish and the agent is inequity-averse. Hence, ors, and mistakes in updating beliefs based under heterogeneity in inequity aversion, on information. The most common version the bonus system is superior. This difference of an incorrect prior is overconfidence: the is due to a difference in how much the first agent has overly positive views about herself mover has to invest in an attempt to increase or some of her prospects. the social surplus and be rewarded for it by an inequity-averse second mover. For the Application. I consider the implications of agent to increase the social surplus under overconfidence for contracting. First, sup- the bonus system, she must pay her mar- pose that an agent produces one of two levels ginal cost of effort. But for the principal to of output, high or low. The true probability increase the social surplus by increasing the of high output is q, but the agent has an wage in the gift-exchange system, he must incorrect prior in that she believes the prob- compensate the agent for her marginal cost ability of high output is ​q ​ q. The princi- ̃ > of effort plus half of the marginal increase in pal offers an output-contingent wage, paying ​ the social surplus (since when splitting the w​ H​ for high output and ​w​ L​ for low output. social surplus, the agent will keep half of the The principal is risk neutral, while the agent increase for herself). When this investment has mean-variance preferences of the form is certain to be rewarded—when there are ​q ​w​ ​ (1 ​q ​)​w​ ​ ​q ​(1 ​q ​)(​w​ ​ ​w​ ​)2. ̃ H + − ̃ L − ̃ − ̃ H − L no selfish individuals around—it does not For now, suppose that there is no asymmet- matter which investment is more costly, as ric information or moral hazard. 1086 Journal of Economic Literature, Vol. LII (December 2014)

In this case, the principal’s problem is welfare effects—suggesting that the conse- quences of false beliefs can be less serious (PC) min q​w​ ​ (1 q)​w​ ​ under time consistency than under time H + − L inconsistency. As a first example of how s.t. ​q ​w​ ​ (1 ​q ​)​w​ ​ agents with small overconfidence can fare ̃ H + − ̃ L well, I consider a version of de la Rosa’s ​q ​(1 ​q ​)(​w​ ​ ​w​ ​)2 u. (2011) moral-hazard model. Suppose that the − ̃ − ̃ H − L ≥ agent chooses between two levels of effort, Noting that q​w​ ​ (1 q)​w​ ​ ​q ​w​ ​ high and low. Her output with high effort is H + − L = ̃ H + (1 ​q ​)​w​ ​ (​q ​ q)(​w​ ​ ​w​ ​), we can plug as described above, but her output with low − ̃ L − ̃ − H − L PC into the maximand to get effort is low with probability 1. The cost of low effort is zero, and the cost of high effort min u q ​(1 ​q ​)(​w​ ​ ​w​ ​)2 is e. The agent’s incentive-compatibility­ con- + ​̃ − ̃ H − L straint is then (​q​ q)(​w​ ​ ​w​ ​). − ̃ − H − L ​q ​w​ ​ (1 ​q ​)​w​ ​ ̃ H + − ̃ L This gives the optimal wage spread 2 ​q ​(1 ​q ​)(​w​ H​ ​w​ L​) e ​w​ L​. q ​ q − ̃ − ̃ − − ≥ (6) ​w​ ​ ​w​ ​ ______​̃ − ​ * H − L = ​ 2​q ​(1 ​q ​) ≡ Δ ̃ − ̃ or Given ​w​ ​ ​w​ ​, the principal sets the levels H − L (7) ​q ​(​w​ ​ ​w​ ​) ​q ​(1 ​q ​)(​w​ ​ ​w​ ​)2 e​. of the two wages so that PC binds. ̃ H − L − ̃ − ̃ H − L ≥ Equation (6) identifies a basic exploitation motive with overconfident agents: if the agent Let the lowest ​w​ ​ ​w​ ​ satisfying this H − L is too optimistic that high output will occur, inequality be **. The comparison of * Δ Δ she overvalues the wage paid for high out- and ** highlights an important distinction Δ put, so the principal can decrease expected regarding different kinds of contracts in this wages by paying her a little more for high review. If * **, then ​w​ ​ ​w​ ​ * Δ > Δ H − L = Δ output and much less for low output. This is optimal, and the agent’s IC constraint is a special case of the well-known observa- is not binding. In the sense that the prin- tion that individuals with different priors cipal’s primary consideration is exploiting would like to speculate against each other the agent’s mistaken beliefs—and possible (Harrison and Kreps 1978, for instance). It is constraints derive from this attempt, while also an example of a situation where a type of other constraints are either nonexistent or contract that may appear to have a classical not binding—the contract is exploitative purpose (an output-contingent wage often by the (somewhat informal) standards of serves to overcome moral hazard) instead Section 6. In contrast, if * **, then Δ < Δ has an exploitative purpose. ​w​ ​ ​w​ ​ ** is optimal. In this case, H − L = Δ The exploitation motive, however, is lim- therefore, the consideration determining​ ited for agents with only slightly overoptimis- w​ ​ ​w​ ​ remains to make sure that the H − L tic beliefs, so that in more realistic settings agent’s incentive-compatibility constraint these agents do not receive exploitative is satisfied, not to take bets against her. In contracts. This finding contrasts with that this sense, the contract is not primarily about for time-inconsistent agents above—where exploitation, although the agent’s expected even small degrees of naïveté can have large wage is still affected by her overconfidence. Ko˝szegi: Behavioral Contract Theory 1087

It is interesting to note that—as is easy to which for small implies ​w​ ​ u ​q ​ , ϵ L ≈ − ̃ ϵ check using equation (7)—in this latter case yielding q​w​ ​ (1 q)​w​ ​ w​ ​ q u H + − L = ​ L + ϵ ≈ − the optimal wage spread is decreasing in (​q ​ q) . Hence, relative to a contract with​ ̃ − ϵ overconfidence. Intuitively, an overconfident w​ ​ ​w​ ​ 0, the principal makes a profit of H − L = agent overestimates how easily she can pro- approximately (​q ​ q) on agent ​q ​. But how ̃ − ϵ ̃ duce high output, and hence a less output- much does she lose on agent ​Q ​? To see this, ̃ sensitive incentive contract is sufficient to note that agent ​Q ​’s perceived value from tak- ̃ induce her to choose high effort. This insight ing the above contract is has some interesting welfare implications. First, because a lower-powered incentive ​w​ ​ Q ​ ​Q ​(1 ​Q ​) 2 L + ​ ̃ ϵ − ̃ − ̃ ϵ contract allows the agent to bear less risk, mild overconfidence increases efficiency. ​w​ ​ Q ​ u (​Q ​ ​q ​) Second, when principals compete for the ≈ L + ​ ̃ ϵ ≈ + ̃ − ̃ ϵ agent, the agent enjoys the full increase in social efficiency, so that having slightly wrong This means that, when contracting with agent​ beliefs actually benefits her.7 Q ​, the principal loses a profit of (​Q ​ ​q ​) ̃ ̃ − ̃ ϵ As a second limitation on exploiting small because having the less speculative con- degrees of overconfidence, consider screen- tract around improves agent ​Q ​’s perceived ̃ ing agents with different degrees of opti- ­alternative option. If ​q ​ q is sufficiently ̃ − mism, based on Eliaz and Spiegler (2008). lower than ​Q ​ q, therefore, exploiting agent​ ̃ − Suppose that there is no moral hazard, but q ​ is not worth it for the principal. Intuitively, ̃ in addition to the agent above, there is an while the less optimistic type values a con- agent with beliefs ​Q ​ q ​, and the principal tract with ​w​ ​ w​ ​ and therefore allows the ̃ > ​̃ H > ​ L would like to contract with both agents with- principal to decrease expected wages, the out directly observing their beliefs. Based on more optimistic type values the same con- equation (6), the principal would like to offer tract much more, so that—to avoid having to a slightly exploitative contract to the less opti- pay a kind of information rent—the principal mistic agent ​q ​ and a more exploitative con- would rather not exploit the less optimistic ̃ tract to the more optimistic agent ​Q ​. Given type. This result implies that the principal ̃ the latter incentive, however, the principal either does not exploit the agent or exploits might not exploit agent ​q ​. I demonstrate her substantially, generating the empiri- ̃ this heuristically by showing that the princi- cal implication that if we see exploitation, it pal prefers to offer agent ​q ​ a contract with should be large. ̃ ​w​ ​ ​w​ ​ 0 rather than ​w​ ​ ​w​ ​ , H − L = H − L = ϵ where 0 is small. With ​w​ ​ ​w​ ​ , ϵ > H − L = ϵ 3. Moral Hazard the less optimistic agent’s binding PC becomes For the rest of this review, I turn to a more comprehensive, but informal, survey ​w​ ​ q ​ ​q ​(1 ​q ​) 2 u, of efforts to incorporate psychology-and- L + ​̃ ϵ − ̃ − ̃ ϵ = economics ideas into contract theory. I orga- nize the topics according to classical con- tract-theoretic principles, and begin with discussing moral hazard. 7 To see this, note that under , principals Moral hazard arises in contracting situ- pay the same expected wage to any agent—the agent’s expected output conditional on high effort—so a decrease ations when a decision taken by the agent in the wage spread increases the agent’s utility. after the contract is signed affects the ­utility 1088 Journal of Economic Literature, Vol. LII (December 2014) of the principal, and the action cannot be higher ­payoff; but if the principal pays a directly contracted upon. In these settings, high fixed wage, the agent responds with there are typically outcomes related to the high effort so that the principal receives a agent’s action that can be contracted upon fair share of the pie. Hence, a high wage is (e.g., output), and the key issue is how often profit maximizing because it puts the the principal optimally designs a contract agent ahead and thereby creates intrinsic that induces the agent to take an action he motivation to help the principal. Englmaier prefers (captured in the incentive-com- and Leider (2012) also investigate con- patibility constraint), and that the agent ditions under which it is optimal to use will accept (captured in the participation ­inequity-aversion-based incentives versus constraint). This section summarizes the performance-based pay, and show that existing ­behavioral-economics literature on inequity aversion is more likely to be used moral hazard.8 if output is a poor signal of effort or the agent is highly inequity averse or produc- 3.1 Intrinsic Motivation: Reliance on tive. This is one formalization of Akerlof’s Voluntary Actions distinction between “primary” labor mar- One important theme that has emerged kets (governed by gift exchange and above- from the literature is that optimal incen- market-clearing wages) and “secondary” tive schemes may rely crucially on “intrin- labor markets (governed by explicit incen- sic motivation”—actions that do not carry a tives and market-clearing wages). financial reward. This contrasts with classi- cal models of moral hazard, where the only Voluntary bonuses. Fehr, Klein, source of motivation is explicit or implicit and Schmidt (2007) establish that if financial incentives. heterogeneity in inequity aversion is present in the population, a contract Social preferences as a source of intrinsic based on voluntary bonuses—by which the motivation. Akerlof (1982) was perhaps the principal can reward the agent for high first (within economics) to argue that firms effort ex post—often dominates the above might pay high wages to induce high effort high-wage contract based on gift exchange. as part of a gift exchange. Englmaier and A bonus contract engages inequity aversion Leider (2012) formalize this idea in a model much like a high-wage contract above, in which the agent has inequity-averse except with the order of moves reversed: preferences vis-à-vis the principal. As in much like an inequity-averse­ worker previous models, the principal can induce responds to a high wage (and only a high high effort with performance-based pay. wage) with high effort, an inequity-averse But he might also induce high effort with principal is willing to pay the bonus in performance-independent pay. Specifically, response to (and only in response to) high if the principal pays a low (near-market- effort. Hence, if there are sufficiently many clearing) fixed wage, the agent does not fair-minded principals, both fair-minded work hard, lest the principal enjoy a much and selfish agents are willing to exert effort. The superiority of the bonus scheme arises from the presence of selfish individuals 8 Kamenica (2012) reviews evidence of many systematic in the population: a party that first takes psychological forces on how individuals respond to incen- a costly action (exerting effort or paying a tives, and more specifically, Fehr, Goette, and Zehnder (2009) review evidence on how reference-dependent fair- high wage, respectively) faces the risk that ness concerns affect interactions in a labor-market setting. a selfish partner will not reward it, and this Ko˝szegi: Behavioral Contract Theory 1089 risk is lower for the bonus contract because Despite the general tendency of inequity the action has lower cost in that case.9 aversion to generate wage compression, Itoh (2004), Rey-Biel (2008), and Bartling and 3.2 Reduced Wage Sensitivity von Siemens (2010) identify some condi- Another finding in the literature is that— tions under which increased pay inequality for multiple reasons—an agent’s wage may can result. Specifically, if limited liability pre- be less responsive to her output than a classi- vents the principal from punishing the agent cal moral-hazard model predicts. by reducing her pay, he can instead punish her by paying another agent more. In Rey- Wage compression. Complementing the Biel’s (2008) two-agent deterministic set- research on moral hazard with social prefer- ting, for instance, an agent who works hard ences reviewed in the previous subsection, when the other agent shirks is more than it seems natural to assume that employ- compensated for her cost of effort. In other ees compare themselves not just to their words, the principal may prefer tournament- employer, but also to fellow employees. A type incentives even when classical reasons very intuitive implication of inequity aver- to prefer such an incentive structure—e.g., sion is then that wages tend to be more com- common shocks—are absent. pressed than what one would expect based on classical models. Because agents do not Simple schemes due to loss aversion. In the like the ex post inequality that can occur by canonical model of moral hazard in a risky chance as a result of high-powered (i.e., very environment, increasing the power of incen- performance-sensitive) individual incen- tives is beneficial to the principal because tives, to relax their participation constraint it aligns the agent’s motives with the prin- individual incentive pay is reduced. When cipal’s goals, but it is also costly because it agents are sufficiently averse to ex post requires exposing the agent to more risk and inequality, the principal might institute hence paying her a higher risk premium. team-based incentives not because he can- Because loss aversion affects a person’s atti- not observe individual performance or wants tudes toward risk, it affects this fundamental to encourage cooperation (as in many clas- tradeoff. A finding that reappears in mul- sical models), but to insure agents against tiple papers is that—because it generates painful inequality (Englmaier and Wambach first-order —loss aversion leads 2010; Bartling 2011). Furthermore, Bartling to the unresponsiveness of transfers to out- and von Siemens (2010) argue that because comes over some regions of outcomes. As comparisons are less pronounced across an extreme example in the context of con- firms than within firms, consistent with the tracting when the consumer is loss averse evidence inequity aversion predicts no wage and her demand for the product is uncertain compression across firms. (e.g., mobile phone contracts), Herweg and Mierendorff (2013) show that—consistent 9 Similarly, Fehr and Schmidt (2004) show that a with many real-world examples—the seller ­voluntary-bonus scheme outperforms an explicit incen- might prefer a flat-rate contract. The flat-rate tive contract in a multitasking environment in which per- contract leads the consumer to overuse the formance on both tasks is observable to the principal, but performance on only one is contractible. As is well-known product, but it is still profit maximizing if this since Holmström and Milgrom (1991), providing incen- moral-hazard problem is sufficiently weak. tives on the contractible task leads an agent to ignore the In many moral-hazard settings, especially unincentivized task, leading to inefficiency. A scheme in which the principal can reward high effort on both tasks employment contracts, however, incentives with a voluntary bonus increases efficiency. must also be provided, so completely flat 1090 Journal of Economic Literature, Vol. LII (December 2014) contracts are not optimal. In such situations, ­expectations of the wage. In each period, wages must be responsive to performance firms face a productivity shock and make a somewhere, but nevertheless, loss aversion one-period fixed-wage take-it-or-leave offer predicts simpler reward schemes than do to workers. The subgame-perfect equilib- typical classical settings. Herweg, Müller, rium displays a number of interesting and and Weinschenk (2010) establish that if the economically important features. First, agent’s reference point is her rational expec- because firms are reluctant to cut the wage tations about the wage (as in Ko˝szegi and below the reference wage, workers already Rabin 2006, 2007), then the optimal contract inside the firm experience downward wage often has a “bonus structure,” with two pos- rigidity: in response to moderate decreases in sible wage levels. In a dynamic extension of productivity, the wage remains unchanged. the expectations-based model due to Ko˝szegi Second, if productivity falls too much to and Rabin (2009), Macera (2012) proves that pay the reference wage, firms do not induce under some circumstances the current wage discretionary effort. Then, depending on is completely unresponsive to performance, the importance of discretionary effort, they with incentives provided by future opportu- either retain the worker with a low wage nities.10 And de Meza and Webb (2007) show knowing that this will result in low effort, or that if the reference point is the median lay the worker off, both of which are inef- wage, then the agent’s wage is unresponsive ficient. Third, since newly hired workers had to performance up to the median perfor- expected to be employed with lower prob- mance level, but responsive to performance ability, they come in with lower expectations above that level.11 regarding how much they will make, so their wage is more flexible than that of existing Wage stickiness. Eliaz and Spiegler (2014) workers. develop a model of labor-market dynam- 3.3 The Effect of Extrinsic Motivation on ics that combines the two major features Intrinsic Motivation of worker behavior studied separately by the papers above: (i) that workers can be As mentioned in section 3.1, a number induced to exert discretionary effort by pay- of theories in behavioral economics imply ing them a fair wage; and (ii) that workers’ that an agent might be willing to exert costly notion of a fair wage is based on their lagged effort for reasons other than explicit mate- rial incentives. At the same time, all theories 10 In Daido and Murooka (2012), team incentives assume that agents are also responsive to emerge under some circumstances as an optimal way to explicit incentives. The recognition that an manage agents’ loss aversion, even when individual incen- agent may be motivated by different types of tive pay is feasible. In particular, suppose that a hard-work- ing agent’s probability of success in her project is low, and incentives raises the natural question of how consider first individual incentive pay that is based on the different incentives interact. A number of performance of only her own project. Then, working hard papers in the literature suggest that extrin- creates an expectation of high pay and thereby exposes the agent to a likely sense of loss, dampening the incentive to sic incentives can crowd out intrinsic moti- work. By paying the agent also for other agents’ successes, vation, so that it can be optimal to employ the principal reduces the probability of loss, increasing the weaker explicit incentives than classical incentive to work. From a single agent’s point of view, get- ting paid for another agent’s success is equivalent to the models would suggest. principal sometimes “turning a blind eye”—i.e., forgiving In addition to the stylized fact that explicit failure—but Daido and Murooka argue that team incen- performance incentives are often quite tives are a credible way to implement such a blind-eye policy. weak, different strands of research in this 11 For related work, see also Jofre et al. (2012). area are motivated by (and consistent with) Ko˝szegi: Behavioral Contract Theory 1091 somewhat different kinds of empirical obser- knows that the cost is high, he is worried vations. Some researchers have found cir- that the agent might also have received a sig- cumstances under which explicit incentives nal that the cost is high, so that he chooses have a contemporaneous negative effect on a higher bonus to motivate the agent. As a effort—that is, effort is lower when greater result, a large bonus for completing a task explicit incentives are in place. In particu- becomes a signal of task difficulty for the lar, Gneezy and Rustichini (2000) document agent, which reduces her motivation to work that imposing a modest fine on parents for in the future. In a closely related paper with picking up their kids late at a day-care center multiple agents, Fang and Moscarini (2005) increases the prevalence of late pickups. And suppose that the principal receives a private Falk and Kosfeld (2006) find in an experi- signal about each agent’s ability, and can ment that a “controlling contract”—which decide whether to make different contract prohibits the lowest contribution levels offers for agents with different signals. If by the agent—also lowers contributions.12 agents are overconfident, the principal might This evidence is slightly different from that prefer not to differentiate contracts, as the studied extensively since Deci (1975) and information from differentiated contracts reviewed for instance by Deci, Koestner, and would (given their overconfident prior) pro- Ryan (1999), whereby providing extrinsic vide information to agents that is on average incentives to perform a task leads individu- bad news and, hence, would lower intrinsic als to engage in the same or similar tasks less motivation. often once the incentives are removed. Other papers have a similar theoretical Given the topic of this subsection, I orga- structure to that of Bénabou and Tirole, nize existing research according to the mech- but the information the principal has is of anism through which explicit incentives act a different nature. In Sliwka (2007), there on intrinsic motivation. are three types of agents in the population: selfish, fair, and conformist. Selfish agents Informed principals. A number of theories care only about their material well-being, employ an informed-principal framework in whereas fair agents care also about the prin- which the principal has superior informa- cipal’s payoff. Conformists behave like the tion regarding a variable that affects both other type (selfish or fair) that they believe his incentive-design problem and the agent’s is in the majority in the rest of the popula- willingness to exert effort, so that a contract tion. The principal knows the share of fair with strong explicit incentives can undermine types, which affects both his optimal incen- the agent’s motivation through its effect on tive scheme and conformists’ willingness to her beliefs. In Bénabou and Tirole (2003), exert effort. And in Herold (2010), the agent the principal chooses a bonus for task com- may be “trustworthy”—motivated to exert pletion knowing the agent’s ability or cost of effort even without explicit incentives—or completing the task, while the agent receives “untrustworthy”—willing to exert effort only only a signal of this variable.­ If the principal­ if explicitly motivated to do so. The principal has privately held beliefs about the agent’s probability of being trustworthy (his “trust” 12 For the above negative effect to occur, it seems cru- cial that the control is imposed by the principal rather than in the agent). This always affects the profit- exogenously. For instance, List (2007) finds that when allo- ability of different incentive schemes. It can cating money between themselves and a recipient, many also affect the agent if she takes nontrusting fewer dictators give a positive amount if they have the option of taking away money than if they do not have this behavior as a hostile act and reciprocates it, option. or if the principal also takes an action that is 1092 Journal of Economic Literature, Vol. LII (December 2014) either a complement or a substitute to the for ­prosocial behavior. This crowding out agent’s action. reduces the effectiveness of extrinsic incen- The above informed-principal models can tives, and under some conditions can induce naturally explain evidence on the future neg- a negative effect of extrinsic incentives. ative effect of explicit incentives mentioned Bénabou and Tirole’s model explains the at the beginning of the subsection: once the evidence on the contemporaneous nega- agent draws the respective negative conclu- tive effect of explicit incentives on prosocial sion from a high-powered incentive contract behavior mentioned above. Since removing in each case, she will be less motivated in the explicit incentives reinstates the signal- future tasks. But these models do not pro- ing motive, however, the model does not vide full explanations for the contempora- seem to predict a future effect. neous negative effect of explicit incentives. While the models explain why explicit incen- Reciprocal behavior. Another potential tives might be weakened by the crowd-out explanation for the positive contemporaneous of intrinsic motivation, they remain positive impact of weak explicit incentives on effort reinforcers in the sense that, if a given worker is reciprocal behavior. The original model of receives strong rather than weak incentives intentions-based reciprocity, Rabin (1993), (among equilibrium contract offers), she defines a kind act as one that sacrifices the exerts greater effort on average. The intu- person’s own material well-being to increase ition derives simply from the negative future the material well-being of the other player, effect of explicit incentives: since strong and assumes that a reciprocal player prefers explicit incentives have negative future con- to respond to a kind act with a kind act. While sequences, the principal uses them only if in this model weak incentives cannot generate they increase current effort.13 an increase in effort if the agent is known to be reciprocal—if they did, they would benefit Social signaling. Bénabou and Tirole the principal and hence could not be consid- (2006) develop a model in which agents have ered kind—von Siemens (2011b) shows that heterogeneous degrees of altruism, and all they can do so under some circumstances if types prefer to be perceived by others as both selfish and reciprocal agents are pres- altruistic. If there are no explicit incentives, ent. Since selfish agents take advantage of prosocial behavior signals a high degree of weak explicit incentives at the expense of the altruism, so that many agents are willing to principal, choosing such incentives is kind act prosocially for image reasons. But with vis a vis selfish agents. If there are many self- explicit incentives for prosocial behavior, ish agents, therefore, choosing weak explicit such behavior could be due to material- incentives is on average a kind act, to which ism and is thus a less convincing signal of reciprocal agents respond with high effort. altruism, reducing the image motivation Furthermore, this is profitable for the prin- cipal if reciprocal agents respond sufficiently strongly to the principal’s kindness. Ellingsen 13 Nevertheless, Bénabou and Tirole (2003) point out and Johannesson (2008) demonstrate a simi- that the unconditional correlation between explicit incen- lar result in a closely related model that dif- tives and effort could be negative, which could lead a naïve observer to incorrectly conclude that extrinsic incentives fers in why exactly the agent responds to a crowd out contemporaneous intrinsic motivation. This can kind act kindly: the principal’s choice of happen because explicit incentives are more likely to be incentive contract reveals how nice the prin- used when the principal’s information points to low agent motivation, which is also when the agent is less likely to cipal is, and some agents prefer to be nice to work. nice principals. Ko˝szegi: Behavioral Contract Theory 1093

Anticipatory utility. Immordino et al. her bias benefits her in ­equilibrium.15 In con- (2011) consider the problem of a risk-neu- trast, because baseline optimism implies that tral principal offering an output-contingent the agent is overly willing to accept high- wage to a risk-neutral agent with unlimited powered contracts with a low base wage, the liability, who receives a private signal about principal responds to large overconfidence her return to effort before choosing her with very high-powered­ contracts to exploit effort level. The agent derives utility from the agent’s mistake, and in equilibrium the anticipating high wages, so that she has an agent’s incentive constraint may not bind. incentive to suppress signals that indicate a In this range, the contract is more like an low return to effort. If anticipatory utility is exploitative contract as defined in section 6, sufficiently important relative to actual out- and the large overconfidence always hurts comes, then (in contrast to the prediction the agent.16 of a classical model) implementing the first- In the context of insurance best level of effort is impossible. Intuitively, and job-search behavior, Spinnewijn (2012) the agent exerts high effort only if she is suf- assumes that agents are baseline optimistic ficiently rewarded for a good result, but this and may be control pessimistic or control makes a good result very desirable, increas- optimistic, and contrasts the response of a ing the motive to suppress bad news and social planner (who maximizes agents’ utility hence distorting information use. subject to a budget constraint) and a com- petitive profit-maximizing firm. When choos- 3.4 Moral Hazard and Overconfidence ing the level of benefits for a biased agent, Individuals often overestimate their abil- the social planner accounts for the effect of ity to do well in a task, either in the sense a change in search behavior on the agent’s of baseline optimism—overestimating one’s welfare, so that he responds to control opti- performance given an effort level—or in the mism by increasing benefits to lower search sense of control optimism—overestimating effort. A profit-maximizing firm, in contrast, the return to effort—or both.14 Researchers does not care about the agent’s welfare, so have explored a few implications of these it does not face the same consideration. But phenomena for moral hazard. because the agent’s baseline optimism lowers De la Rosa (2011) and Gervais, Heaton, and her demand for insurance, the firm responds Odean (2011) study moral-hazard models in by lowering benefits. which the risk-averse agent is both baseline­ optimistic and control optimistic. Since a control-optimistic agent requires lower- powered incentives than a rational agent to 15 This result is related to the point made by Mullainathan et al. (2012) that underutilization of medical implement a given level of effort, the princi- care due to present bias can lower the moral-hazard cost pal responds to moderate overconfidence by of providing insurance, potentially benefitting the insured. lowering the power of incentives. This allows 16 See also Santos-Pinto (2008) for a closely related analysis that focuses on identifying conditions under which the agent to bear less risk, increasing social agent overconfidence benefits the principal in a single- welfare. Interestingly, if there is competi- principal setting. If effort is observable—so that only the tion between principals, the agent receives participation constraint is relevant—the principal ben- efits from agent overconfidence both because the agent is all of this increase in social welfare, so that overly willing to accept a given contract and because he can exploit her as above. If effort is unobservable, the principal benefits if the agent is control optimistic (and other, more technical, conditions hold) as control optimism makes it 14 The above distinction is taken from Spinnewijn easier to satisfy the agent’s incentive-compatibility condi- (2012). tions as well. 1094 Journal of Economic Literature, Vol. LII (December 2014)

3.5 Potential Future Directions player never sacrifices his own payoff for that of the other player, so the principal’s choice In this section, I discuss some outstanding of wage will never be considered kind by the issues in the literature in the same order in agent. As a result, the agent never exceeds which the related research appears above. the materially cost-minimizing level of per- Further progress in the literature exploring formance.17 This means that gift exchange optimal moral-hazard contracts with social is either not about reciprocity, or a different preferences seems somewhat hindered by or more complex reciprocity mechanism is the lack of a theory of reference-group for- involved. A model with agent heterogeneity mation—the determination of which other similar to that of von Siemens (2011b), dis- individuals’ outcomes an agent cares about. cussed above, might provide a starting point, This not only makes it more difficult to test but whether such a model will prove to be a or the economic importance of the compelling general explanation requires fur- above results, but also prevents researchers ther research. from asking natural questions, such as how The literature on loss aversion does not reorganizations affect the reference group seem to face similarly central puzzles, but and hence behavior, in a theoretically disci- there are also plenty of unanswered ques- plined manner. tions. For instance, note that the contract While existing models with inequity-averse terms for loss-averse agents above emerge decisionmakers can explain gift exchange in because the firm offers partial insurance to the laboratory and in some organizations, an agent who is first-order averse to risk. they do not predict the phenomenon when— From this perspective, it is interesting to note as in many real-life settings—the employer is that (as far as I am aware) no paper has care- significantly wealthier than most employees fully incorporated loss aversion into an insur- and cares mostly about his own material out- ance model with moral hazard and derived comes. Because in this case a generous wage implications for optimal contracts. The logic does not put the worker ahead, gift exchange of Herweg, Müller, and Weinschenk (2010) should not result. And because a selfish prin- seems to imply that in many settings, an cipal does not pay a bonus she does not have insurance contract with a simple deduct- to, the voluntary-bonus system should not ible—whereby the agent pays a deductible if work. losses exceed a threshold—is optimal. Deepening the above puzzle, Netzer and While the literature on the interaction Schmutzler (2012) show that another major between extrinsic and intrinsic motivation is model of social preferences, Rabin’s (1993) one of the most exciting and productive in model of intentions-based reciprocity, can- behavioral contract theory, it also calls for not convincingly account for gift-exchange substantial future research on when differ- behavior, either, so that to date there is no ent sources of motivation are likely to domi- compelling explanation of this simple phe- nate behavior. As explained above, all of the nomenon. At first glance, Rabin’s model appears to predict gift exchange: offering a high wage seems to be a kind act, to which 17 Nevertheless, if the agent could choose lower lev- els of performance—that is, exert costly effort to punish the agent should naturally respond with her the principal—her wage can be much higher than that of own kind act of a high effort. In contrast, a selfish agent in the same situation, materially benefit- Netzer and Schmutzler show that if the prin- ing her and hurting the principal. In this case, the princi- pal chooses a high wage not to induce discretionary high cipal is selfish, this mechanism cannot work effort, but merely to avoid the punishment that lower to increase effort. By definition, a selfish wages would trigger. Ko˝szegi: Behavioral Contract Theory 1095 existing models explain only a part of the has, to my knowledge, not been studied in evidence and make orthogonal or opposing the classical literature. predictions on the other evidence—with no sound guidelines as to where exactly each 4. Asymmetric Information and Screening model applies. One possible reason for the confusing state of the literature may be that, In classical screening models, a principal unlike extrinsic motivation, intrinsic moti- (e.g., an airline) faces agents whose prefer- vation is a complex multifaceted phenom- ences (e.g., willingness to pay for a ticket) he enon that is poorly understood. Individuals does not know. Ideally, the principal would may derive intrinsic motivation from many like to contract with multiple agent types, sources, including those in the models differentiating the contract (e.g., the above, and perhaps additional motives such and conditions of the airline ticket) accord- as the sense of being in a good organization ing to the agent’s preferences. The principal’s or the sense of doing good work. Both more main constraint—captured in the incentive- evidence and more theory seems necessary compatibility constraint that a more profit- to sort out what situations evoke the differ- able type not take a less profitable option—is ent kinds of intrinsic motivation identified by that the agent may not reveal that she is a researchers. profitable type (e.g., that she has a high value The observation that a bias—here, slight for a ticket). The central issue in contract overconfidence—can benefit the agent raises design is how to off minimizing this the natural question of when this is likely to informational advantage with the objective happen. Clearly, fixing the contract the agent of achieving gains from trade. This section signs, a bias always (weakly) reduces her wel- summarizes the bulk of the literature incor- fare. Hence, for a bias to benefit the agent, it porating ideas from psychology and econom- must change contract terms in her favor. This ics into situations of hidden information and can happen if the agent’s bias leads her to screening.18 either overvalue her outside option or under- 4.1 Commitment versus Flexibility value interacting with the principal—both forcing the principal to offer better terms. It has been understood that individu- The former is the case above, where com- als with self-control problems benefit from peting principals offer terms the agent over- commitment, and hence might be willing to values. The latter seems less likely to happen sign contracts that restrict their choices in in many settings, as the principal—in an some way. At the same time, uncertainty can attempt to induce the agent to accept—will make it inefficient to remove all choice, thus try to write contracts the agent overvalues generating a possible tradeoff between com- rather than undervalues. But the question mitment and flexibility. A number of papers of when biases can benefit an agent has not explore aspects of this contracting question. received sufficient attention in the litera- In an early contribution, DellaVigna and ture. Research on this question might draw Malmendier (2004) study the optimal two- insights from the large classical literatures on part tariff of a firm facing a consumer who commitment and reputation, which also rec- has known present-biased preferences and ognizes that appearing nonrational can bene- fit a player. For instance, the beneficial effect of overconfidence above is similar to that of a 18 Section 6 discusses a different set of screening issues related to exploitative contracts, which arise because dif- commitment to work hard, although the par- ferent agents might exhibit the mistake the principal is ticular type of cost ­overconfidence implied looking to exploit to different extents. 1096 Journal of Economic Literature, Vol. LII (December 2014) can make an investment such as or levels than slightly lower types, as this would exercising. Investment carries an immediate mean that the highest types are overconsum- cost that is unknown to the firm, and gen- ing from an ex ante point of view. In addition, erates a fixed future benefit that is known. Amador et al. show that it is often optimal to If the consumer is sophisticated regarding allow for complete flexibility above the mini- her present bias, the optimal two-part tariff mum requirement. Roughly, what- implements the first-best outcome by impos- ever flexibility in savings the principal allows ing a subsidy that equals the amount by above the minimum, the agent tends to over- which the consumer undervalues the future consume in that range from an ex ante point benefit. Hence, the contract induces more of view. Since the principal cannot do much future-oriented behavior without compro- about the overconsumption, he might as well mising the consumer’s flexibility to respond allow the agent to adjust consumption to the to cost shocks. While this contract achieves taste shock.20 first-best despite asymmetric information Galperti (2012) studies a setting where, about the cost, it is clear that the optimality much like in DellaVigna and Malmendier is specific to the problem and form of uncer- (2004), the principal can implement first- tainty: for instance, if the consumer’s benefit best if she knows the agent’s degree of time was also uncertain, the first-best could not be inconsistency, but a screening issue arises achieved with a two-part tariff. because there are both time-inconsistent and Amador, Werning, and Angeletos (2006) time-consistent agents in the population. A study optimal contracting in a setting where flexible contract that solves the time-incon- the first-best is typically not achievable. sistent agent’s commitment problem must They consider a consumption–savings prob- offer rewards for saving. But because a time- lem in which a present-biased agent faces a consistent agent can expect to receive the privately observed taste or needs shock that rewards more often, she derives rent from affects her marginal instantaneous utility of such a flexible contract. To lower the time- consumption and that she does not initially consistent agent’s information rent, then, the know.19 Thinking of the consumer’s late self principal curtails the flexibility of the time- as the agent and her early self (or a social inconsistent agent’s contract, restricting planner maximizing the expected utility of both high and low levels of savings. Galperti her early self) as the principal, they char- argues that this feature is consistent with acterize the optimal commitment, defined restrictions on retirement savings devices in as the optimal subset of the agent’s budget the US. set from which she will be allowed to choose Esteban, Miyagawa, and Shum (2007) her savings level. Amador et al. show that the study a monopolist’s nonlinear pricing optimal commitment always features a mini- problem in which the heterogeneity in mum savings rule akin to those observed in many retirement systems. Intuitively, it is not 20 This insight does not always hold due to a subtle optimal to allow the highest types—those consideration in the principal’s problem. By eliminating a with the greatest taste for immediate con- range of possible consumption levels, the principal forces sumption—to choose higher consumption those types who would otherwise have preferred to con- sume in that range to consume either less or more. The former response is welfare-increasing, while the latter response is welfare-decreasing. The former force domi- 19 Technically, this kind of situation is called a “hidden nates the combined harmful effects of the latter force knowledge” problem, because—as distinct from non-linear and lowers flexibility if sufficiently more types respond by pricing or screening problems—exogenous asymmetric consuming less, which is the case if the density of types information emerges after the contract is signed. decreases sufficiently quickly. Ko˝szegi: Behavioral Contract Theory 1097 agent ­preferences is due to heterogene- to ex post preferences. The logic of these ity in temptation disutility in the sense of results (especially those for upward tempta- Gul and Pesendorfer (2001).21 In Gul and tion) relies crucially on the assumption that Pesendorfer’s model, the agent has prefer- individuals differ only in temptation disu- ences consisting of commitment utility—the tility, and as the authors point out, it would utility of an option if the agent can perfectly be interesting to study the same problem commit to it in advance—and temptation if commitment utility also differs between disutility, where she suffers the latter if she types. does not choose the most tempting option 4.2 Overconfidence and Screening Issues from her choice set. Whereas with hyper- Related to False Beliefs bolic discounting the agent would like to commit her future behavior because she The observation that some individuals will have different preferences in the future have overly positive views about themselves than she does now, with temptation disutil- or their prospects raises the question of how ity the same demand for commitment arises to optimally screen agents with different because she wishes to lower temptations. The levels of overconfidence. Before discussing impact of temptation disutility on the opti- research on this question, it is worth noting mal menu turns out to depend on the kind a general limit to screening beliefs according of temptation agents experience. If all agents to their accuracy: because two individuals experience upward temptation—whereby with the same beliefs about future outcomes high consumption is more tempting than and same ex ante preferences choose from a low consumption (e.g., cigarettes)—then the menu in the same way, separating them at optimal menu is a singleton, so that the seller the contracting stage by means of self-selec- does not separate consumers with different tion is impossible. In particular, this is the preferences at all. Intuitively, because trying case even if one has correct and the other to take advantage of an agent’s temptation incorrect beliefs, and hence the two agents to sell her more would make her less will- will behave differently given the contract. ing to participate, the seller just chooses to As a result, a contract signed by agents with maximize agents’ commitment utility, which given beliefs are often priced according to is possible with a singleton since (by assump- some average of the actual outcomes of these tion) agents have homogenous commitment agents. In these situations, a biased agent utility. If all agents experience downward exerts an on rational agents. The temptation—whereby lower consumption externality is negative if the biased agent is is more tempting than higher consumption less profitable than the rational agent (e.g., (e.g., exercise)—the optimal menu is identi- in insurance, where an overconfident agent cal to that with standard preferences equal underestimates her risk), and positive if the to ex post preferences. Intuitively, since not biased agent is more profitable (e.g., in many buying is the most tempting alternative and exploitative contracts discussed below). consumers can always choose this alterna- tive, increasing the menu does not increase Overconfidence in the insurance market. temptation disutility, so that it does not make Sandroni and Squintani (2010) study insur- consumers less willing to participate. Hence, ance contracts when some high-risk agents the seller screens consumers according­ believe that they are low-risk, and show that the presence of overconfident agents has 21 Esteban and Miyagawa (2006) analyze the competi- qualitative observable implications in a com- tive case. petitive (but not in a monopolistic) insurance 1098 Journal of Economic Literature, Vol. LII (December 2014) market. To understand their results, recall rejected by high-risk consumers. As a result, that in the seminal competitive insurance low-risk and overconfident consumers pre- model of Rothschild and Stiglitz (1976) with fer not to buy more insurance at prevailing rational agents, high-risk and low-risk agents prices. Hence, contrary to common intuition separate in equilibrium and the price of insur- that biases increase the scope for govern- ance is actuarially fair for both types, so that ment intervention, in this case mandatory the price of insurance in each class does not insurance is not Pareto-improving.22 depend on the proportion of types in the pop- Contrary to the prediction of classical ulation. In addition, because high-risk agents insurance models that higher-risk consum- able to fully insure at actuarially fair prices ers buy more insurance, recent empirical do not want small amounts of insurance even research finds a zero or negative correla- at low prices, low-risk agents receive at least tion between risk and insurance coverage in partial insurance. Overconfidence changes many circumstances (for instance Chiappori each of these predictions. If some agents and Salanie 2000; Finkelstein and McGarry who believe themselves to be low-risk are in 2006). To explain this puzzle, Spinnewijn fact high-risk, the group of low-risk and over- (2013) studies insurance markets with opti- confident—who by the above logic cannot be mistic agents and both moral hazard and distinguished because they choose contracts asymmetric information. If a single-crossing in the same way—receive more expensive property holds, the only way for a principal insurance that depends on the proportion of to separate agents is to offer them differ- overconfident in the population, generating ent amounts of insurance, and this separates heterogeneity in insurance prices within a agents (roughly) only according to baseline risk class. In addition, since the low-risk and optimism. But because control optimism overconfident believe that their insurance is affects how much precaution an agent takes overpriced, they may not buy insurance. when insured, the correlation between risk Sandroni and Squintani (2007) reconsider and insurance coverage depends on the cor- the scope for Pareto-improving interventions relation between baseline optimism and con- in the insurance market when overconfident trol optimism. agents are present. Although in a rational competitive insurance market low-risk agents Overconfidence and debt financing. receive cheap insurance, such insurance is Manove and Padilla (1999) consider a model partial to discourage high-risk agents from in which overoptimistic entrepreneurs taking it. In this equilibrium, low-risk agents ask for loans from banks to finance overly would prefer to buy more insurance at the large projects relative to the most produc- same price, so that a government policy of tive use of money, and banks cannot distin- mandatory insurance—which allows low-risk guish overoptimistic entrepreneurs from agents to get more insurance at reasonable realists. Manove and Padilla’s main result prices—can be Pareto-improving. But with overconfident agents, the group of low-risk and overconfident are offered more expen- 22 Schumacher (2012) studies a model with a related sive insurance, and since this group believes effect, where—similarly to the overconfident in Sandroni and Squintani (2007)—naïve present-biased agents exert a that they are receiving insurance above the negative pricing externality on sophisticated agents by not actuarially fair price, they may not want taking enough precautions, making government interven- much insurance. If this is the case, the group tion less desirable. Nevertheless, in Schumacher’s (2012) model, government intervention to induce taking more of low-risk and overconfident can choose a precaution is Pareto-improving if the share of naïve agents perceived-optimal insurance contract that is is substantial. Ko˝szegi: Behavioral Contract Theory 1099 is that banks are often willing to fund too model of von Siemens (2011a), employ- many projects, from a social point of view. ees might be selfish or inequity-averse with Intuitively, banks care only about recouping respect to coworkers. Since a worker can their investment in expectation, and do not affect only her own outcome and both self- internalize the fact that an optimistic entre- ish and inequity-averse workers prefer more preneur could have used the money better money to less (holding others’ outcomes elsewhere. Furthermore, because collateral constant), the firm cannot differentiate self- requirements help ensure that banks recoup ish and inequity-averse workers who have their loans, they exacerbate the excessive been hired. In order to employ low-ability lending problem. inequity-averse individuals, therefore, a firm needs to pay a premium to compensate 4.3 Other Topics in Screening these workers for falling behind coworkers. This section discusses various other To economize on this cost, the firm either research on screening. reduces the gap between high- and low- ability workers by distorting production, or Using framing for screening. Salant and excludes low-ability inequity-averse work- Siegel (2013) consider the screening prob- ers by paying no premium. In contrast, von lem of a seller who can temporarily manipu- Siemens (2012) shows that in a competi- late buyers’ preferences. The seller chooses tive market, the existence of multiple firms quality and price for two types of buyers, allows high- and low-ability workers to sort and in addition may employ a “frame”—such into different firms, reducing the impact of as a salient comparison price in a shop or a social comparisons, and hence often allow- glitzy environment in a casino—to change ing low-ability inequity-averse workers to the types’ utility functions. Crucially, how- be employed as well. In a related setting, ever, the effect of the frame is fleeting, and Kosfeld and von Siemens (2011) assume either (i) the buyer returns the item if she that some workers are selfish—i.e., exert later realizes that she does not value it above individual effort if compensated for it, but her outside option, or (ii) the buyer antici- never exert team effort—while other work- pates the manipulation and stays away from ers are conditional cooperators—i.e., exert the store if she does not like what she will team effort if coworkers do too, and prefer to buy. Salant and Siegel show, roughly, that work in a team environment. In equilibrium, such framing can help the seller if and only if firms offer two types of contracts: a high- it relaxes the high type’s incentive constraint powered incentive contract designed for the without increasing the low type’s incentive to selfish workers, and a low-powered incen- mimic the high type. In this case, the prin- tive contract designed for the conditional cipal can offer the low type a cheap product cooperators. To prevent selfish workers from while manipulating the high type into buying accepting the latter contract, its wage must the expensive product, eliminating the high be relatively low, so that it may be profitable type’s information rent and increasing effi- for the firm despite competition. This means ciency of provision to the low type. that if team effort is sufficiently important, firms with lower pay are more productive. Sorting in the labor market. Kosfeld and von Siemens (2011), von Siemens (2011a), Loss aversion and screening. Hahn et al. and von Siemens (2012) study labor-market (2010) analyze a monopolist’s optimal menu outcomes for employees with different types when consumers are loss averse and do not of social preferences. In the know their willingness to pay in advance. Just 1100 Journal of Economic Literature, Vol. LII (December 2014) as a loss-averse employee dislikes the risk in much of the research above, the contract inherent in a wage schedule that discrimi- does not have a classical purpose or the pri- nates finely between performance levels (see vate information concerns a parameter in section 3.2), a consumer who does not ini- psychology and economics. tially know her willingness to pay dislikes the Another important question is the extent risk inherent in a menu that discriminates to which a social planner can screen agents— finely between different willingness-to-pay some of whom might be behaving optimally, realizations. Hence, the seller often offers a and some suboptimally—in the context of small number of products relative to the het- noncoercive interventions aimed at improv- erogeneity in the population. ing individual consumer decisions.23 For Carbajal and Ely (2012) assume instead instance, O’Donoghue and Rabin (2003) illus- that consumers know their types in advance trate that a “sin license,” whereby consumers and have a type-dependent reference point would be required to buy a moderately priced relative to which they evaluate outcomes. license to smoke, often dominates “sin taxes” Carbajal and Ely study how the optimal because it separates time-consistent consum- menu depends on the reference-point ers from both naïve and sophisticated pres- function, and also derive properties of self- ent-biased consumers. Intuitively, neither confirming reference consumption plans— type of present-biased­ consumer obtains the where a type’s consumption in equilibrium license—naïve consumers because they do coincides with her reference point. In con- not believe they will smoke in the future, and trast to an individual-decision-making set- sophisticated consumers because they want ting—where an increase in the reference to prevent their future selves from smoking— point always hurts the agent—a higher self- but time-consistent consumers who are mak- confirming reference consumption plan can ing an optimal decision to smoke do. benefit both the seller and some agents. Intuitively, a higher reference point leads the 5. Auction Theory and Mechanism Design seller to exclude fewer low types from the market (who, due to their higher reference Mechanism design is the study of what point, value the product more highly), and outcomes can be achieved, and how, when as a result of this market expansion, higher a principal is interacting with multiple types receive higher information rents. agents with private information. The central consideration in the principal’s problem is 4.4 Potential Future Directions how to set up the game form so that agents Screening seems to be an understudied reveal their private information. This section topic in behavioral contract theory. Notably, reviews applications of behavioral-econom- behavioral research has not been incorpo- ics ideas to mechanism design. The bulk of rated into optimal income taxation, although existing research concerns the theory of auc- some psychological phenomena (such as tions, a prominent class of mechanisms for hyperbolic discounting, anticipatory util- selling to buyers with unknown valuations ity, and overconfidence) seem important for the product. The section also describes in labor, leisure, and consumption choices. the limited amount of work on other issues More generally, there seems to be little in mechanism design. research on how insights from psychology and economics affect classical screening 23 See Thaler and Sunstein (2008) and Camerer et al. problems in which the private information (2003) for arguments in favor of such interventions, as well concerns a standard preference parameter; as some examples. Ko˝szegi: Behavioral Contract Theory 1101

5.1 Loss Aversion in Auctions generates a revenue ranking between the three most commonly analyzed auction for- When applying loss aversion to auction mats—first-price, second-price and all-pay theory, predictions depend on whether a bid- auctions—that depends on whether bidders der experiences gain–loss utility separately use separate or net evaluation. Under sepa- from money and from the product being sold rate evaluation, an all-pay auction is less risky (separate evaluation), or jointly from the net than a first-price auction because payment is utility of the transaction (net evaluation). In deterministic, and a first-price auction is less commodity auctions—where the payment risky than a second-price auction because is monetary and the product is nonmon- payment conditional on winning is deter- etary—the former assumption seems more ministic. In fact, Eisenhuth (2013) estab- appropriate, while in induced-value labora- lishes that under loss aversion with separate tory experiments—where both the payment evaluation, any optimal mechanism features and the “product” are monetary—the latter a riskless payment, and the optimal auction assumption seems to apply.24 As Lange and is an all-pay auction with a minimum bid. Ratan (2010) argue, this implies both that In contrast, with net evaluation the ranking one must use different theories to interpret between the all-pay and first-price auctions induced-value laboratory auctions and com- is reversed. Intuitively, the gap between the modity auctions, and that many experimental net utility from winning and not winning findings may not apply to real-world settings.25 equals the product’s value with an all-pay Lange and Ratan (2010), Eisenhuth auction but only the surplus with a first-price (2010), and Eisenhuth and Ewers (2012) auction, so the latter appears less risky for analyze the effect of expectations-based loss a net evaluator. In fact, Eisenhuth (2010) aversion as modeled in Ko˝szegi and Rabin proves that in this case, the optimal auction is (2007) on the revenue ranking of standard the first-price auction with a minimum bid.26 sealed-bid private-values auctions. Ko˝szegi Another possibility explored by research- and Rabin’s model implies that individuals ers is that the reference point of auction are first-order averse to local risk, and hence participants is partially determined by the bid more aggressively in auction formats reserve price. In particular, Rosenkranz and that are better at insuring them against the Schmitz (2007) specify the reference point in uncertainty from participation. This ­behavior money as a weighted average of the reserve price and an exogenous parameter, so that an 24 The perspective that individuals use separate evalu- increase in the reserve price directly raises ation when a real commodity is traded off with money bidders’ willingness to pay, and hence their is consistent with a large body of evidence, such as the bids. When either the number of bidders endowment effect, that is commonly interpreted in terms of loss aversion. Nevertheless, Eisenhuth and Ewers or the exogenous component of the refer- (2012) present some suggestive experimental evidence ence point is high, the bid-raising benefit of that a model with net evaluation better describes a com- increasing the reserve price dwarfs the cost modity auction, as well. 25 Lange and Ratan (2010) find that with net evaluation, from the increased risk of not being able to bidders “overbid” in a first-price private-values action, sell, and therefore even a small degree of loss while with separate evaluation they underbid. Intuitively, with net evaluation, loss-averse agents experience a sen- sation of loss compared to the expected payoff when 26 From a theoretical perspective, a weakness of the losing the auction, motivating them to bid higher. In con- above literature is that (as the authors note) it makes pre- trast, with separate evaluation, loss aversion in the money dictions qualitatively similar to those of appropriately spec- dimension kicks in when the agent wins the auction, and ified models with classical risk-averse bidders. Of course, thus results in agents underbidding in order to avoid hav- the implications of loss aversion could be quantitatively ing to pay more than expected. more important. 1102 Journal of Economic Literature, Vol. LII (December 2014) aversion can lead to a significant increase in engages a naïve sunk-cost fallacy, whereby the optimal reserve price. Shunda (2009) bidders are reluctant to stop bidding in an assumes that bidders’ reference points auction when they have spent money on it, depend on the seller’s reserve price as well and do not predict the full extent of this effect. as his “buy price”—a price at which a bid- He shows that the profit-maximizing supply of der may purchase the good from the seller auctions is constrained by the consideration before the auction starts. Increasing the buy that an auction attracting few bidders tends price raises the bidders’ willingness to pay, to end early and generate large losses for the and hence leads to a larger pool of partici- auctioneer. The same consideration also cre- pants and higher bids in the auction stage, ates an effective barrier to entry, as competi- but lowers the probability that a bidder buys tors who cannot initially attract many bidders the item in the first stage. The optimal buy must absorb large losses. price trades off these two effects. Crawford et al. (2009) begin studying optimal auction design with level-k bidders. 5.2 Other Topics in Auctions Following Camerer, Ho, and Chong (2004) Nonequilibrium thinking. Jehiel and Lamy and Crawford and Iriberri (2007), their (2012) ask why sellers often employ abso- model distinguishes between bidders of dif- lute auctions (auctions with a reserve price ferent levels of strategic reasoning. L0 types of zero) despite positive value for the object, follow some simple bidding strategy, such as and why sellers often set secret reserve prices. bidding a random number or always bidding In the spirit of Jehiel (2005) and Eyster and their true valuation. L1 types formulate their Rabin (2005), Jehiel and Lamy propose that strategy as the best response to the L0 strat- some bidders do not understand how the egy. For example, if L0 types bid randomly, potential for getting a good deal varies with then L1 types believe that winning does not the auction format. Absolute auctions can be reveal any information about the value of the used to attract bidders who underappreciate object, so that they—similarly to fully cursed how product quality and the participation players in Eyster and Rabin (2005)—overbid rate depend on the auction format, as these compared to the predictions of the classical bidders overestimate the quality of in model. L2 types best respond to L1 types. such auctions and may also underestimate Crawford et al. (2009) find that the optimal the participation rate. And auctions with a reserve price may either be higher or lower secret reserve price can be used to attract than in an equilibrium model, while seller rev- bidders who do not understand how secret enue—due to overbidding by many players— reserve prices differ in distribution from pub- is usually higher. They also give an example lic reserve prices, failing to anticipate that a of an exotic auction specifically designed to secret reserve price is likely to be higher than exploit bidders’ nonequilibrium thinking that public reserve prices. can yield arbitrarily large revenues. Augenblick (2011) studies the penny auc- tion, an auction format in which agents bid for Anticipated regret. Engelbrecht-Wiggans items in predefined increments (often 1 cent) (1989) and Filiz-Ozbay and Ozbay (2007) and have to pay a nonrefundable fixed price study a model in which bidders anticipate for each bid. Augenblick ­documents that feelings of regret in case their bid turns out to bidders severely overbid in these auctions, be suboptimal ex post. Losers of a ­first-price ­generating for instance an average profit mar- sealed-bid auction or a Dutch auction may gin of 104 percent in auctions for direct cash feel regret when confronted with the infor- payments. He argues that the penny auction mation that the winning bid was lower than Ko˝szegi: Behavioral Contract Theory 1103 their valuation, which means that they could agents with intentions-based social prefer- have walked away with a positive surplus had ences in the sense of Rabin (1993), where a they followed a more aggressive strategy. On player is willing to sacrifice her own material the other hand, winners of a first-price auction payoff for the sake of the other player if and may also feel regret when discovering that only if she believes the other player is simi- the second-highest bid was lower than their larly kind. The codependence of intentions own, so that they could have won at a lower makes kindness an equilibrium phenom- price. Anticipating “loser regret” induces par- enon. If in equilibrium agents have positive ticipants to make higher bids than otherwise, intentions toward each other, then the obser- while anticipating “winner regret” induces vation that they can achieve better outcomes them to make lower bids. These emotions is similar to that in Kucuksenel (2012). But are only triggered when the regret-inducing since intentions are endogenous, a key part information is revealed to participants, so a of the designer’s problem is to set up a mech- seller can influence bidder behavior through anism in which agents can develop positive information provision. In a setting that trig- intentions toward each other. Bierbrauer and gers only loser regret, participants overbid Netzer show that the designer can achieve relative to their consumption utility.27 this by adding unused options to enrich one- self to the mechanism, so that truth-telling 5.3 Mechanism Design will be considered a kind action. Even so, Beyond auction theory, there is little work there is always an equilibrium in which play- on the psychology and economics of mecha- ers are unkind to each other, so the designer nism design. One exception concerns the must make sure that the kind equilibrium is implications of social preferences in public- played—for which the theory provides no interest situations, those in which everyone guidance. Finally, Carmichael and MacLeod benefits if everyone gives up personal material (2003) make a related point in the context gain for the sake of others. In a number of dif- of holdup: they show that caring about the ferent models, authors find that social prefer- other party’s investment—a notion akin to ences tend to increase efficiency. Kucuksenel a preference for —can soften the (2012) considers mechanism design with holdup problem in contracting.28, 29 altruistic agents and shows that more altruis- tic agents—those who attach a larger weight to others’ material payoffs—are more likely 28 Desiraju and Sappington (2007) consider a nonlinear pricing problem with two agents who are averse to ineq- to produce a that is efficient to uity in the total ex post payoff they receive, a setting that produce, and are more likely to trade when does not neatly fit the public-interest environment above. it is efficient to do so. The intuition is simple: While with standard preferences a two-agent screening problem reduces to a single-agent screening problem, with more altruistic agents care more about social inequity aversion it is fundamentally multi-agent mecha- efficiency, so it is easier to implement socially nism design, as each agent cares about the other’s surplus. efficient outcomes with them. Bierbrauer and Desiraju and Sappington show that if the agents are ex ante identical, the principal can implement the standard out- Netzer (2012) establish a similar result for comes while eliminating all ex post inequality. In one such mechanism, a low-type agent gets paid more if the other agent is a high type than if the other agent is also a low 27 See also Cramton et al. (2012) for an application of type. If the agents are ex ante different, however, there is this concept to clock auctions such as those commonly no way to implement the standard outcomes while elimi- used for selling radio spectrums, electricity, or gas. And nating inequality for all type realizations. Roider and Schmitz (2012) study optimal reserve prices 29 There is a also a mini-literature on implementa- in a related model that assumes bidders get utility directly tion when agents have a preference for reporting their from winning or losing, independently of whether they types honestly, showing that even a weak such preference could have done better with another strategy. can greatly help the principal (Alger and Renault 2006; 1104 Journal of Economic Literature, Vol. LII (December 2014)

5.4 Potential Future Directions distribution of bids, but in an English auc- tion she receives some information on this Several topics in mechanism design are distribution. Such informational issues seem natural candidates to link to behavioral understudied in auction theory and mecha- ideas. A notable aspect of auctions is that nism design. they seem to trigger some unique emotions related to competition that are not trig- 6. Exploitative Contracting gered in many other settings. In particular, as noted by Malmendier and Lee (2011), This section considers contracts designed it has been understood since ancient times exclusively or primarily to exploit false beliefs that auctions invoke “bidding fever,” while by the agent. While there does not seem to Morgan, Steiglitz, and Reis (2003) and Cox, be a precise formal way to distinguish such Smith, and Walker (1988) argue that auc- contracts from nonexploitative contracts that tions engage a spite motive and the joy of are affected by agent mistakes, an informal winning, respectively. An interesting issue and subjective distinction seems useful to is how these emotions work and how they make. Namely, a contract is exploitative if affect optimal auction design. For instance, the economically central considerations driv- it is not immediately clear why these emo- ing it derive from trying to profit from the tions are triggered by auctions, while other agent’s mistake, and other considerations or situations (including many in this review) constraints are nonexistent, not binding, or are more likely to generate positive social not central. preferences. The literature has explored two broad While classical mechanism design recog- forms of false beliefs. First, an agent’s false nizes that public revelation of information beliefs may be about the contract itself; in can help the principal—e.g., in an auction if Gabaix and Laibson (2006), for instance, bidders’ valuations are positively correlated naïve consumers underappreciate that the (Milgrom and Weber 1982)—behavioral product they are purchasing will require economics introduces at least two novel rea- some add-on purchases. Alternatively, an sons for why information could matter. First, agent might believe that regulation prevents as in the case of regret above and in the case certain charges, when in fact it does not of anticipatory utility discussed in other sec- (Armstrong and Vickers, 2012). Second, an tions of the review, information can directly agent’s false beliefs may be about her own affect agents’ preferences. Second, infor- behavior given the contract; in DellaVigna mation can lead strategically naïve agents and Malmendier (2004), for instance, naïve to change their views about the strategic hyperbolic discounters are aware of all con- situation, either directly through provid- tract features, but they mispredict how they ing information on how others play or indi- will behave given a contract. rectly through focusing their attention on specific aspects of the game. For instance, Methodological issues. The models dis- in a second-price auction a bidder must rely cussed in this section assume that the prin- exclusively on her predictions regarding the cipal knows the agent’s tendency to commit mistakes, so that in this domain, the princi- pal is more informed about the agent than Matsushima 2008). These papers, however, make specific she is herself. While unusual from a clas- assumptions about the structure of honesty preferences that do not seem to be based on psychology evidence, so I sical point of view—where it is commonly do not review them here. assumed that the agent’s tendencies are Ko˝szegi: Behavioral Contract Theory 1105 her private information—this­ assumption Nevertheless, the fact that firms distort often makes sense when the principal is a prices to take advantage of consumer mis- firm with the capacity and willingness to takes can have a number of efficiency impli- collect and analyze tremendous amounts cations, so that the above “safety in markets” of data about consumers, and the agent is very partial. Consumers might buy prod- is an individual consumer. The theories ucts whose value is below the true price but are also unlike informed-principal models above the misperceived low price, generat- with novel types of principal information ing inefficiency in participation decisions because the agent does not make infer- (Heidhues and Ko˝szegi 2013). Other ineffi- ences about her tendencies from the con- ciencies obtain due to the high prices naïve tract she is offered. Hence, as discussed consumers do not appreciate. In Gabaix and by Eliaz and Spiegler (2006), these models Laibson (2006) and Armstrong and Vickers can be treated formally as contracting with (2012), naïve consumers ignore add-on ­noncommon priors. While the classical lit- prices, firms respond by choosing high add- erature also recognizes the potential rele- on prices, and this leads sophisticated con- vance of noncommon priors for contracting sumers to take socially inefficient steps (such and other settings, the recent literature has as arranging for a substitute with higher pro- both reinvigorated research on such models duction costs) to avoid the add-on. In Grubb and changed its style and emphasis: the the- (2009), consumers believe they can predict ories posit a particular form of agent beliefs their consumption more precisely than they based on psychology evidence, take a stance actually can, firms respond by setting high on which beliefs are correct, and tend to marginal prices for high usage, and this can consider welfare issues. lead consumers to underutilize the ex post. In Gottlieb and Smetters (2012), life- 6.1 Nature and Implications of Exploitative insurance buyers do not properly account for Contracts the probability that they will lapse their poli- Seemingly cheap products. Exploitative cies, firms respond by front-loading premi- contracts often make products appear ums to make the policy look better, and this cheaper than they really are. The reason is distorts consumption smoothing. At the same simple: firms want to encourage consumers time, distorted prices could also have ben- to buy, so if they choose multiple prices (e.g., eficial effects if they correct another prob- a basic price and an add-on price) that con- lem. For instance, firms take advantage of sumers will pay, they aim to obtain revenues naïve present-biased agents by offering high more from the prices consumers underap- per-usage fees for pleasurable goods whose preciate. In a competitive market, this does consumption the agent underestimates, and not necessarily affect firm profits or con- these high prices have the beneficial effect sumer or social welfare: similar to the logic of lowering the present-biased agent’s con- of loss-leader pricing (Lal and Matutes 1994) sumption (DellaVigna and Malmendier, as well as that of many switching-cost mod- 2004). Further, as argued by -Gill (2006), els (e.g., Farrell and Klemperer 2007), firms if a consumer’s mistake leads her to consume compete aggressively for valuable consumers too little, the higher consumption induced ex ante and bid down the more noticeable by deceptively low prices can benefit her. component of the price until they eliminate net profits. In this sense, naïve consumers Cross-subsidies. Another recurring fea- can be protected from their mistakes by mar- ture of exploitative contracts, first discussed ket forces. in detail by Gabaix and Laibson (2006), is 1106 Journal of Economic Literature, Vol. LII (December 2014) that sophisticated consumers benefit from students would value such a contract much the presence of naïve consumers. Because more highly, forcing the school to pay a kind naïve consumers do not anticipate some of information rent. This implies that con- fees that they will pay, they tend to generate tracts should either involve no speculation or higher profits than sophisticated consumers substantial levels of speculation. buying the same contract. In a competitive Eliaz and Spiegler (2006) analyze a market—where firms make zero profits on closely related problem in which the prin- average—it must therefore be that firms cipal knows that the agent’s preferences make money on naïve consumers and lose will change between the time of signing the money on sophisticated consumers, in effect contract and the time of taking an action, acting as a tool for cross-subsidizing the lat- whereas the agent may assign positive proba- ter type. bility to her preferences remaining the same. Exacerbation of mistakes using high fees. The time inconsistency in the agent’s pref- DellaVigna and Malmendier (2004) were the erences introduces an additional value from first to point out that firms might fine-tune contracting similar to that in section 4.1: the contracts to exacerbate consumers’ mistakes. agent prefers the contract to restrain her A number of papers explore the specific fea- future behavior. Again, if the agent’s beliefs tures of such exploitative contracts. are sufficiently close to that of the principal, Eliaz and Spiegler (2008) study optimal she receives the ex ante optimal contract that dynamic contracting when an agent’s pref- fully commits her behavior, but if she is suf- erences are uncertain at the time of signing ficiently naïve regarding her time inconsis- the contract, and the agent may be more tency, she receives an exploitative contract optimistic than the principal about the bet- in which she is effectively rewarded if she ter state occurring. As an example, suppose does not change her mind and punished if that all -school students will end up she does change her mind. in the nonprofit sector or in the corporate Heidhues and Ko˝szegi (2010) study exploit- sector with the same fixed probabilities, ative credit contracts, and analyze the kinds but students have different, to the school of mistakes that are exacerbated by firms and unobservable, overoptimistic beliefs that lead to large welfare losses. In their model, they will end up in the nonprofit sector. The present-biased consumers who might be school’s optimal screening menu includes a partially naïve about their time inconsistency standard nonexploitative loan contract that can sign exclusive contracts with competitive is not contingent on subsequent outcomes, suppliers of credit, agreeing to a menu of as well as exploitative loan contracts that installment plans according to which credit amount to speculation on where a student can be repaid in the future. In the competi- will end up. With the latter type of loan, tive equilibrium firms offer seemingly cheap the student pays somewhat less if she ends credit to be repaid quickly, but introduce up in the nonprofit sector and much more large penalties for falling behind this front- if she ends up in the corporate sector. An loaded repayment schedule. The contracts interesting feature of the optimal menu is are designed so that borrowers with even an that students with beliefs close to those of arbitrarily small degree of naïveté both pay the school receive a nonexploitative con- the penalties and repay in an ex ante subop- tract. Intuitively, although the school could timal back-loaded manner more often than make a little money on slightly overoptimis- they predict. Intuitively, a lender chooses the tic students by taking a small bet on what the repayment options so that, when deciding student will do, much more overoptimistic how to repay in the future, the consumer will Ko˝szegi: Behavioral Contract Theory 1107 be indifferent between the front-loaded and should unshroud price components underap- back-loaded schedules. As a result, if she is preciated by consumers, and then compete naïve about her time inconsistency—no mat- on them. While it is unclear whether and ter by how little—she falsely believes that she how one can educate consumers, and this will repay quickly. To make matters worse, question requires further research, a num- the same misprediction leads the consumer ber of papers have investigated the incen- to underestimate the cost of credit and bor- tive to unshroud assuming that firms can row too much—despite borrowing being for costlessly do so. Gabaix and Laibson (2006), future consumption. Spiegler (2006), and Heidhues, Ko˝szegi, and Murooka (2012) show that unshrouding is 6.2 Helping Consumers often unprofitable because it turns profitable Many researchers have asked whether naïve consumers into unprofitable sophisti- there are ways of improving consumer wel- cated consumers. Hence, deceptive products fare in the types of situations described in or contracts can often survive in markets. the previous subsection. The theme that Furthermore, Heidhues, Ko˝szegi, and emerges from this literature so far is that both Murooka (2012) identify a perverse aspect of market-based and regulatory solutions face when this can happen: products that generate severe problems, but some forms of inter- lower social surplus than the best alternative vention may nevertheless increase welfare. facilitate deception precisely because they As has been pointed out by Armstrong and would not survive in the market if consumers Vickers (2012), for instance, one common understood hidden fees, and therefore firms feature of welfare-increasing interventions often make profits on exactly such products. is redistribution of wealth from sophisticated to naïve consumers through the reduction of Problems with financial advice. Armstrong cross-subsidies. While some researchers and and Zhou (2011), Stoughton, Wu, and Zechner observers use this as an argument against (2011) and Inderst and Ottaviani (2012) show intervention on the basis that we should not that if consumers are naïve, firms pay commis- hurt rational agents who are paying attention sions to financial advisors to steer consumers and making individually optimal decisions, to buy their high-priced products, and finan- from an economic perspective this seems no cial advisors do so even if this is suboptimal better a reason to refrain from intervention for a consumer. Murooka (2013) establishes than the fact that pollution taxes redistribute that even among financial wealth from potential polluters to the rest advisors does not push commissions down to of society. In fact, to the extent that sophis- a competitive level, as deceptive firms must ticated consumers are wealthier than naïve pay significantly higher commissions to stop consumers, redistribution might be an addi- advisors from explaining to consumers that tional argument for intervention. other products are superior. Since the high commissions are ultimately paid by consum- Lack of market incentives to educate con- ers, in this situation the presence of financial sumers. The fact that consumers underesti- advisors who can explain deceptive practices mate some costs associated with product use lowers consumer welfare. raises the question of whether market partic- ipants have incentives to educate consumers Interventions to improve consumer deci- or offer more transparent products or con- sion making. A number of researchers recog- tracts. First-pass logic suggests that in a suf- nize that when consumers might mispredict ficiently competitive industry, competitors their own behavior, classical disclosure, which 1108 Journal of Economic Literature, Vol. LII (December 2014) typically limits information to those related to Problems with government intervention. A the contract itself, is insufficient, and inform- number of authors point out potential prob- ing consumers about themselves is necessary lems with market intervention. Grubb (2012) (for example Bar-Gill and Ferrari 2010). An considers services, such as mobile phones approach by Thaler and Sunstein (2008) and and bank overdraft protection, for which Kamenica, Mullainathan, and Thaler (2011) consumers may not know the marginal price would require firms to provide contract details of the next unit of service, and asks whether and individualized usage information to all requiring firms to disclose this information consumers in a standard form, so that market- at the point of sale increases welfare. If con- based “choice engines” can emerge that help sumers correctly anticipate their probability consumers choose, making consumer choices of running into penalties, such regulation but not necessarily making consumers more can actually hurt because it interferes with sophisticated. efficient screening by firms. Intuitively, pen- alty fees for high usage prevent high-value Welfare-improving interventions that consumers from taking the contracts offered limit contract forms. A number of authors to low-value consumers; yet because con- have shown that limiting what firms can sumers do not know when they apply, these do—thereby targeting the tools firms use to fees do not distort the consumption of low- exploit consumers—can raise social welfare. value consumers. For instance, Heidhues and Ko˝szegi (2010) Kosfeld and Schüwer (2011) establish show that because the large welfare losses par- that, since the inefficiency in Gabaix and tially naïve consumers suffer in credit markets Laibson’s 2006 model of shrouded attributes are driven by large fees, prohibiting large fees is due to sophisticated consumers’ efforts to for small deviations from contract terms often avoid the add-on, in such a setting, educa- raises welfare for any combination of naïve tion—modeled as turning a portion of naïve and sophisticated consumers in the popula- consumers into sophisticated consumers— tion. Similarly, Inderst and Ottaviani (2012) can lower social welfare. And Warren and and Murooka (2013) show that because the Wood (2010) show that when naïve consum- distorted advice financial advisors provide is ers misperceive add-on prices, there is no driven by the large commissions firms pay, budget-balanced regulation that consumers capping or banning commissions, or requiring will perceive as improving their welfare, so commissions to be uniform across products, that there is no welfare-improving regulation can raise consumer welfare.30 that citizens will vote for. Intuitively, a com- petitive market redistributes income from naive to sophisticated consumers, and since 30 Korobkin (2003) provides an interesting legal per- spective on exploitative consumer contracts. Similarly to no consumer believes that she is naïve, she the economic research in this section, he argues that when believes she benefits from the redistribution. consumers do not understand all aspects of the contract, sellers have an incentive to include inefficient terms that 6.3 Potential Future Directions favor themselves. As a remedy, he proposes a modifica- tion of the unconscionability doctrine, a legal doctrine that While researchers have identified a num- renders contract features a party had no effective choice ber of features of exploitative contracts, over invalid. Although this is not the prevailing legal inter- pretation, Korobkin argues that a consumer who did not overall this literature seems to be in its understand part of the contract could not have had effec- infancy, with a variety of open questions. I tive choice over it. And in a completely different approach, highlight some major issues. First, while Bubb and Kaufman (2011) propose that different owner- ship structures, such as customer-owned firms, can lower many researchers employ a static framework firms’ incentive to exploit consumer mistakes. or assume that firms and consumers sign Ko˝szegi: Behavioral Contract Theory 1109 exclusive contracts, it seems important to to them—could be involved. Alternatively, understand the effect of ex post competition this pattern may be due to regulation or the and the possibility of switching for contract threat of regulation, making it more difficult structure. Gottlieb (2008) takes a first step in to conclusively establish that the feature in this research agenda, exploring the dynam- question is deceptive. ics of contracting with present-biased agents Another poorly understood issue is how to in the setting of DellaVigna and Malmendier improve consumer decision making, includ- (2004). He shows that ex post competition ing how to educate or inform consumers leaves contracts for products with immedi- about their own behavior. The likely impact ate costs and future benefits unaffected, but of proposed choice engines to improve con- renders contracts for products with immedi- sumer decision making, and their optimal ate benefits and future costs equivalent to regulation, also deserves closer theoretical spot markets. As an example in the context scrutiny. The concern arises that the platform of smoking, while a retailer and a present- for exploitation shifts from firms to choice biased consumer might prefer to sign an engines—which become like other inter- exclusive contract in which (to restrain her mediaries who deceive consumers—and it future smoking) the consumer is paid a lump is unclear what business model for choice sum ex ante and buys overpriced cigarettes engines will prevent this. In particular, if from the retailer ex post, this contract is choice engines are paid for finding a match ineffective if the consumer can go to other between buyers and sellers, then to attract retailers. Nevertheless, in some markets for consumers, they appear to have similar products with immediate benefits, such as incentives to mislead consumers as firms do. credit cards, consumers do not switch easily Furthermore, because choice engines can to competitors, and firms seem to take advan- analyze offers much more quickly, their pres- tage of this (Ausubel 1991). Procrastination ence can lead to a proliferation of products, on switching offers a plausible explanation, including products that take advantage of but this explanation and especially its impli- very specific or rare mistakes by consumers. cations for equilibrium contracts have not Finally, researchers have studied contract- been formally analyzed. ing to exploit agents who are naïve about A noticeable aspect of existing exploitative their suboptimal behavior much more than contracts seems to be that exploitative fea- they have studied contracting to help agents tures are often masked as having alternative who are sophisticated about their suboptimal uses. For instance, while the primary pur- behavior (e.g., commitment contracts)— pose of a high credit-card late fee is likely to although such sophisticated agents should exploit a consumer’s inattention to this fee or in principle demand help. Indeed, this her optimism that she will not pay late, one asymmetry in the literature seems to reflect can make the argument that the purpose is to an asymmetry in the real-life contracts we overcome the moral-hazard problem that she observe. Whether there are more exploit- might delay repayment. An interesting ques- ative than “helping” contracts, and, if so, tion for behavioral contract theory is why why, is an important area for future research. exploitative features that can in no plausible There are at least two main possibilities. environment be useful are so rare. Some First, the market mechanism might, for form of metasophistication on the part of some reason, favor the exploitation of biases consumers—whereby consumers might real- rather than their mitigation. Second, it might ize the possibility of exploitation, and pure be the case that most individuals are not suf- exploitative features might seem suspicious ficiently sophisticated to demand help, or are 1110 Journal of Economic Literature, Vol. LII (December 2014) averse to helping contracts for another rea- ­delivering a bad lecture. Hart and Moore son, such as being averse to committing their assume that a party is aggrieved and shades own behavior. if she gets a worse outcome than the best possible under the initial contract.31 The way to avoid aggrieving a party and shading, 7. Other Topics therefore, is to fix or restrict the terms in the This section covers two further small lit- ex ante contract; for instance, the parties eratures in behavioral contract theory. may agree to trading at a fixed price. This will result in a failure to trade in circumstances 7.1 Incomplete Contracts when the price does not fall between the A contract is incomplete if it leaves the buyer’s value and the seller’s cost, but econ- details of some transaction that the parties omizes on shading costs, and the optimal care about to be determined at a later time. contract trades off these two considerations. In the classical literature on incomplete con- Furthermore, among multiple terms the par- tracts, such as Grossman and Hart (1986) ties eventually decide on, it is most important and Hart and Moore (1990), parties make to fix the ones that create the most conflict of interim investments into the relationship, interest—and therefore the most potential and because the incomplete contract cannot for shading. Indeed, since the parties have provide direct incentives for these invest- a direct conflict regarding the price, an opti- ments, they are typically inefficient. The mal contract might fix the price but not the main contracting problem is to maximize the full description of the good to be delivered. efficiency of the interim decisions, primarily Finally, to minimize shading, the party who through the allocation of ownership and con- cares more about the good should have the trol rights. right to specify it ex post. In the extreme, if The psychology-and-economics literature the seller is almost indifferent between dif- on incomplete contracts is relatively small. ferent products but the buyer cares a lot— Papers in this literature study ways in which such as in a firm where the employee might individuals’ reactions to a contract generate perform a number of similarly difficult tasks, inefficiencies ex post (in contrast to the clas- only a few of which are useful—it is optimal sical literature, where inefficiencies arise at for the buyer to have control rights. If the the investment stage) and how this affects opposite is the case—such as when the seller optimal contracts and the optimal allocation is subject to large cost shocks—the seller of ownership rights. should be in control.32 In the most influential contribution of this Hart (2009) applies a similar model to a literature, Hart and Moore (2008) analyze a holdup problem, where holding up the other buyer–seller relationship when the contract party results in inefficient shading in the parties write initially cannot condition on the state of the world that determines the opti- 31 Psychologically, this means that a party is not mal terms of trade, and at the ex post stage aggrieved if the terms of the initial contract are bad, but is an aggrieved party can inefficiently “shade” aggrieved if the terms of the renegotiated transaction are on her performance in a way that is impos- bad (in her view). Hart and Moore (2008) explain this dis- tinction by arguing that there is a change in circumstances: sible to prevent contractually. As a simple the final agreement often occurs in situations of bilateral example, while a may be able to verify , whereas the initial contract is agreed to in a that a professor delivered an agreed-to lec- more competitive, “objective” setting. 32 See also Hart and Holmstrom (2010) for an applica- ture, it cannot verify that the lecture was tion of this approach to firm scope—whether and when good, and hence a professor can shade by units should integrate to a single firm or merely cooperate. Ko˝szegi: Behavioral Contract Theory 1111 renegotiated contract. A buyer and a seller a contract so as not to set a reference point, agree to a fixed price ex ante and, when the or—to optimize the insufficient adjustment state of the world is realized, decide whether ex post—might prefer to write a kind of to hold up the other party by refusing to “compromise” contract that they know they trade. Because holdup results in shading, a will trade away from. The paper also derives party only chooses it if the resulting price is two less immediate implications. First, it is sufficiently better than the price agreed to in optimal to write a specific contract rather the contract. Hart shows that an appropriate than rely on ownership rights to incentivize allocation of asset ownership can lower the relationship-specific investments if there is probability of inefficient holdup. In particu- little uncertainty or parties are not very loss lar, it is optimal to assign asset ownership to averse. Second, an employment contract the party who faces more uncertainty regard- dominates a fixed performance contract if ing her value from trade; since this increases the scope for inefficient abuse is small rela- the party’s outside option exactly when her tive to the renegotiation costs generated by value from trade is high, it makes holding her loss aversion due to uncertainty. up less profitable. This implies that—unlike 7.2 Environment Design in the classical model, where asset ownership is determined by the importance of interim A few papers investigate, under various investments—asset ownership is determined psychologically motivated assumptions, how by the importance of uncertainty in a party’s to design individuals’ decision-making envi- value from trade. ronment to achieve specific goals. Herweg and Schmidt (2012) take a some- what different approach to how contracts Optimal nudges. Psychology and econom- affect preferences over trades. They suppose ics has had a major practical impact in moti- that a loss-averse buyer and a loss-averse vating “soft paternalism”—designing policies seller write a contract specifying the nature such that individuals with a tendency to and price of the good to be delivered at a behave suboptimally make better decisions, later date, and the contract serves as a ref- but others are either free to choose as they erence point for outcomes ex post. Since a would otherwise (Thaler and Sunstein 2003; party evaluates a better-than-contracted Thaler and Sunstein 2008), or are not hurt term (e.g., an increase in price for the seller) much (Camerer et al., 2003). Some work as a gain and a worse-than-contracted term can be thought of as providing theoretical as a loss, and loss aversion implies that she is guidance as to how such “nudges” should be more sensitive to the loss, parties are reluc- designed. tant to compromise and trade away from the O’Donoghue and Rabin (1999b) ana- contract. In particular, if the ex post opti- lyze the problem of a principal employing mal terms are close to the specified terms, a naïve present-biased agent to complete a the parties stick to the contract, and even if single task with an uncertain cost of effort, the ex post optimal terms are far, the par- where he can pay the agent depending on ties adjust the terms only partially, in either when she completes the task. The principal case generating (material) inefficiency. faces a delay cost, yet because in any given Here, inefficiency arises when there is a spe- period the effort cost might be high, effi- cific contract, whereas in Hart and Moore ciency requires that the agent sometimes (2008), inefficiency arises when there is only wait. Consistent with the soft paternalist a vague contract. An immediate implication agenda but quite distinct from the litera- is that the parties might prefer not to write ture on exploitative contracts, O’Donoghue 1112 Journal of Economic Literature, Vol. LII (December 2014) and Rabin assume that the principal’s goal default is set such that it is good enough is to choose the most efficient (rather than for part of the population, while the rest of profit-maximizing) incentive scheme. Note the population is forced to pay the cost and that with time-consistent preferences, such switch to the optimal savings rate. an incentive scheme is obvious: the principal imposes a punishment for delay that is exactly Information optimization with emotions. equal to his delay cost, generating a first-best Caplin and Eliaz (2003) and Schweizer and outcome. When the agent is present-biased Szech (2013) characterize optimal medical and the principal does not know the task-cost tests when individuals derive anticipatory distribution, in contrast, the first-best out- utility from their beliefs about their health come is not achievable. Intuitively, since for , and find bad news more aversive high average costs the agent is more prone than good news pleasant. In both papers, to procrastinate, the principal needs severe the principal can commit to an information- punishment for delay to induce her to com- revelation policy that is conditional on an plete the task efficiently; but if he imposes individual’s status. Caplin and Eliaz establish such a punishment, an agent with low aver- that an optimal policy for stopping the spread age costs will tend to complete the task too of HIV gives only noisy bad news: it certifies soon. The optimal contract, then, resembles some, but not all, individuals who are unin- a deadline: punishment for delay is relatively fected, and gives the same signal to every- mild for a while to allow an agent with low one else. This policy protects individuals to delay if necessary, but severe from receiving very bad news, yet provides after a deadline to discourage an agent with sufficient information to help match unin- high average costs from procrastinating. fected individuals. Using methods similar Carroll et al. (2009) study optimal defaults to those in Kamenica and Gentzkow (2011), for retirement savings decisions when Schweizer and Szech (2013) show that such employees have time-inconsistent tastes for a policy also emerges as the optimal way to immediate gratification, and have heteroge- reveal information about infection status to neous optimal savings rates. Employees are an individual who uses the information to initially assigned the default savings rate, make better decisions. and in each period draw a stochastic cost at 7.3 Potential Future Directions which they can change their savings rate to the optimal one, suffering a flow utility loss Most of the literature on incomplete con- if they do not. Employees’ problem is that tracts is based on theories of how a contract (due to their time inconsistency) they may agreed to by the parties affects future pref- procrastinate paying the cost. Carroll et al. erences over outcomes. While this approach establish that if employees have a large time- has produced important insights, it seems inconsistency problem and are not too het- both too broad and too narrow for incomplete erogeneous, it is optimal to set the default contracts. It is too broad because the ques- at the optimal rate for the population distri- tion of how contracts affect preferences is an bution. If employees are very heterogenous important topic for research quite indepen- or the time inconsistency is not so serious, in dently of whether the contract is incomplete. contrast, the optimal policy involves “active Identifying the psychological underpinnings decisions”: the default is set at such an unat- of how contracts affect preferences—includ- tractive level that all employees are forced to ing the extent to which these can be derived pay the cost immediately. And in in-between from psychological phenomena, such as the cases, a compromise policy is optimal: the role of expectations or social preferences, that Ko˝szegi: Behavioral Contract Theory 1113 go beyond contracts, and the extent to which agenda—that is, when more heavy-handed they are specific to contracts—seems to be policies are desirable. an important agenda for future research. At In research on emotions and informa- the same time, the question of how contracts tion in contract theory, authors have largely affect preferences is also too narrow for confined themselves to Bayesian models, incomplete contracts because psychological although anticipatory utility might lead to phenomena that are not directly about this non-Bayesian information processing, and question are also likely to have implications beliefs formed in a non-Bayesian way can for incomplete contracts. change emotions. Oster, Shoulson, and A long-recognized issue at the foundations Dorsey (2013), for instance, argue that facts of incomplete contracts is whether and when regarding testing and economic behavior it is reasonable to assume that contracts will among patients at risk for Huntington’s dis- be incomplete. Since rational agents can ease are only consistent with non-Bayesian­ often write optimal contracts that render models of anticipatory utility, such as standard arguments for incomplete contracts Brunnermeier and Parker (2005). Thinking irrelevant (for example Moore 1992; Maskin about how to optimize information revela- and Tirole 1999), some researchers have tion with non-Bayesian beliefs seems like an conjectured that ideas from psychology and interesting agenda. economics might help provide more compel- ling foundations for the incompleteness of 8. Conclusion contracts. This is still an important problem for future research. This section discusses a few general issues Relative to the attention that has been paid regarding the state and direction of the lit- to the question of soft paternalism at least erature on behavioral contract theory, and since the success of Thaler and Benartzi’s applied behavioral-economics­ theory more (2004) Save More Tomorrow plan, the dearth generally. When applying a behavioral-eco- of theoretical research on such policies is nomics theory to an economic setting, the striking. One potential reason is that most all-too common tendency of much research real-life nudges take advantage of the default is to study how the parameters of the behav- effect—that individuals tend to stick with an ioral theory affect predictions. In studying option they have chosen or to which they have the implications of present bias for credit, been automatically assigned—and we do not for instance, a researcher might analyze how have good theories for why defaults influence less and more present-biased individuals behavior in such a powerful way.33 Yet provid- behave. While such analyses are useful for ing much better theoretical guidance on soft understanding the model and comparing it paternalism seems essential for the continued to a classical one, ultimately economists are practical success of the agenda, especially in more interested in comparative statics with figuring out what kinds of policies are likely respect to variables typically studied in eco- to work and which policies push people in the nomic analysis. To continue with the previ- right direction. Theoretical work is also nec- ous example, a researcher could study how essary for delineating the limits of the nudge the same present-biased individual responds to different kinds of credit or to changes in the . Such predictions are both 33 Most likely, there are multiple psychological reasons economically more relevant and easier to for people not to switch away from defaults, and the effect is robust and powerful because in most situations multiple test than those deriving from manipulating forces operate at the same time. the behavioral model itself. 1114 Journal of Economic Literature, Vol. LII (December 2014)

A central constraint on progress in behav- 70 (3): 489–520. Bénabou, Roland, and . 2006. “Incentives ioral economics is the of compelling and Prosocial Behavior.” American Economic Review portable models of individual decision mak- 96 (5): 1652–78. ing that researchers can apply to economic Bierbrauer, Felix, and Nick Netzer. 2012. “Mecha- nism Design and Intentions.” University of Cologne settings. Much of behavioral contract theory Department of Economics Working Paper 53. uses only the four models introduced in sec- Bolton, Patrick, and Mathias Dewatripont. 2005. Con- tion 2, highlighting both that useful models tract Theory. Cambridge, Mass. and London: MIT Press. can be taken up by many researchers, and Bolton, Patrick, and Antoine Faure-Grimaud. 2010. that there are too few portable models. As “Satisficing Contracts.”Review of Economic Studies one particular example, developing portable 77 (3): 937–71. Brunnermeier, Markus K., and Jonathan A. Parker. psychologically based models of limited or 2005. “Optimal Expectations.” American Economic distorted attention is a first-order question. Review 95 (4): 1092–1118. Bubb, Ryan, and Alex Kaufman. 2011. “Consumer References Biases and Mutual Ownership.” New York University School of Law Research Paper Akerlof, George A. 1982. “Labor Contracts as Partial 11-35. Gift Exchange.” Quarterly Journal of Economics 97 Camerer, Colin F., Teck-Hua Ho, and Juin-Kuan (4): 543–69. Chong. 2004. “A Cognitive Hierarchy Model of Akerlof, George A., and Janet L. Yellen. 1988. “Fairness Games.” Quarterly Journal of Economics 119 (3): and Unemployment.” American Economic Review 78 861–98. (2): 44–49. Camerer, Colin F., Samuel Issacharoff, George Loew- Akerlof, George A., and Janet L. Yellen. 1990. “The enstein, Ted O’Donoghue, and Matthew Rabin. Fair Wage-Effort Hypothesis and Unemployment.” 2003. “Regulation for Conservatives: Behavioral Quarterly Journal of Economics 105 (2): 255–83. Economics and the Case for ‘Asymmetric Paternal- Alger, Ingela, and Régis Renault. 2006. “Screening ism’.” University of Pennsylvania Law Review 151 Ethics When Honest Agents Care about Fairness.” (3): 1211–54. International Economic Review 47 (1): 59–85. Caplin, Andrew, and Kfir Eliaz. 2003. “AIDS Policy Amador, Manuel, Iván Werning, and George-Marios and Psychology: A Mechanism-Design Approach.” Angeletos. 2006. “Commitment vs. Flexibility.” RAND Journal of Economics 34 (4): 631–46. Econometrica 74 (2): 365–96. Carbajal, Juan Carlos, and Jeffrey C. Ely. 2012. Armstrong, Mark, and John Vickers. 2012. “Consumer “Optimal Contracts for Loss Averse Consumers.” Protection and Contingent Charges.” Journal of Eco- Unpublished. nomic Literature 50 (2): 477–93. Carmichael, Lorne, and W. Bentley MacLeod. 2003. Armstrong, Mark, and Jidong Zhou. 2011. “Paying for “Caring About Sunk Costs: A Behavioral Solution Prominence.” Economic Journal 121 (556): F368–95. to Holdup Problems with Small Stakes.” Journal of Augenblick, Ned. 2011. “Consumer and Producer Law, Economics and Organization 19 (1): 106–18. Behavior in the Market for Penny Auctions: A Theo- Carroll, Gabriel D., James J. Choi, David Laibson, retical and Empirical Analysis.” Unpublished. Brigitte C. Madrian, and Andrew Metrick. 2009. Ausubel, Lawrence M. 1991. “The Failure of Competi- “Optimal Defaults and Active Decisions.” Quarterly tion in the Credit Card Market.” American Economic Journal of Economics 124 (4): 1639–74. Review 81 (1): 50–81. Chiappori, Pierre-Andre, and Bernard Salanie. 2000. Bar-Gill, Oren. 2006. “Bundling and Consumer “Testing for Asymmetric Information in Insurance Misperception.” University of Chicago Law Review Markets.” Journal of 108 (1): 73 (1): 33–61. 56–78. Bar-Gill, Oren, and Franco Ferrari. 2010. “Inform- Cox, James C., Vernon L. Smith, and James M. Walker. ing Consumers about Themselves.” Erasmus Law 1988. “Theory and Individual Behavior of First-Price Review 3 (2): 93–119. Auctions.” Journal of Risk and Uncertainty 1 (1): Bartling, Björn. 2011. “Relative Performance or Team 61–99. Evaluation? Optimal Contracts for Other-Regarding Cramton, Peter, Emel Filiz-Ozbay, Erkut Y. Ozbay, and Agents.” Journal of Economic Behavior and Organi- Pacharasut Sujarittanonta. 2012. “Fear of Losing in zation 79 (3): 183–93. a Clock Auction.” Review of Economic Design 16 Bartling, Björn, and Ferdinand A. von Siemens. 2010. (2–3): 119–34. “The Intensity of Incentives in Firms and Markets: Crawford, Vincent P., and Nagore Iriberri. 2007. Moral Hazard with Envious Agents.” Labour Eco- “Level-k Auctions: Can a Nonequilibrium Model of nomics 17 (3): 598–607. Strategic Thinking Explain the Winner’s Curse and Bénabou, Roland, and Jean Tirole. 2003. “Intrinsic and Overbidding in Private-Value Auctions?” Economet- Extrinsic Motivation.” Review of Economic Studies rica 75 (6): 1721–70. Ko˝szegi: Behavioral Contract Theory 1115

Crawford, Vincent P., Tamar Kugler, Zvika Neeman, mal Incentive Contracts under Inequity Aversion.” and Ady Pauzner. 2009. “Behaviorally Optimal Auc- Games and Economic Behavior 69 (2): 312–28. tion Design: Examples and Observations.” Journal of Esteban, Susanna, and Eiichi Miyagawa. 2006. “Temp- the European Economic Association 7 (2–3): 377–87. tation, Self-Control, and Competitive Nonlinear Daido, Kohei, and Takeshi Murooka. 2012. “Team Pricing.” Economics Letters 90 (3): 348–55. Incentives and Reference-Dependent Pref- Esteban, Susanna, Eiichi Miyagawa, and Matthew erences.” http://papers.ssrn.com/sol3/papers. Shum. 2007. “Nonlinear Pricing with Self-Control cfm?abstract_id=1922366. Preferences.” Journal of Economic Theory 135 (1): Deci, Edward L. 1975. Intrinsic Motivation. New York: 306–38. Plenum Publishing Company. Eyster, Erik, and Matthew Rabin. 2005. “Cursed Equi- Deci, Edward L., Richard Koestner, and Richard M. librium.” Econometrica 73 (5): 1623–72. Ryan. 1999. “A Meta-analytic Review of Experi- Falk, Armin, and Michael Kosfeld. 2006. “The Hidden ments Examining the Effects of Extrinsic Rewards Costs of Control.” American Economic Review 96 on Intrinsic Motivation.” Psychological Bulletin 125 (5): 1611–30. (6): 627–68. Fang, Hanming, and Giuseppe Moscarini. 2005. de la Rosa, Leonidas Enrique. 2011. “Overconfidence “Morale Hazard.” Journal of 52 and Moral Hazard.” Games and Economic Behavior (4): 749–77. 73 (2): 429–51. Farrell, Joseph, and Paul Klemperer. 2007. “Coordina- DellaVigna, Stefano. 2009. “Psychology and Econom- tion and Lock-In: Competition with Switching Costs ics: Evidence from the Field.” Journal of Economic and Network Effects.” In Handbook of Industrial Literature 47 (2): 315–72. Organization, Volume 3, edited by Mark Armstrong DellaVigna, Stefano, and Ulrike Malmendier. 2004. and Robert H. Porter, 1967–2072. Amsterdam: Else- “Contract Design and Self-Control: Theory and vier, North-Holland. Evidence.” Quarterly Journal of Economics 119 (2): Fehr, Ernst, Lorenz Goette, and Christian Zehnder. 353–402. 2009. “A Behavioral Account of the Labor Market: de Meza, David, and David C. Webb. 2007. “Incentive The Role of Fairness Concerns.” Annual Review of Design under Loss Aversion.” Journal of the Euro- Economics 1: 355–84. pean Economic Association 5 (1): 66–92. Fehr, Ernst, Alexander Klein, and Klaus M. Schmidt. Desiraju, Ramarao, and David E. M. Sappington. 2007. 2007. “Fairness and Contract Design.” Econometrica “Equity and .” Journal of Econom- 75 (1): 121–54. ics and Management Strategy 16 (2): 285–318. Fehr, Ernst, and Klaus M. Schmidt. 1999. “A Theory of Eisenhuth, Roland. 2010. “Auction Design with Loss Fairness, Competition, and Cooperation.” Quarterly Averse Bidders: The Optimality of All Pay Mecha- Journal of Economics 114 (3): 817–68. nisms.” Munich Personal RePEc Archive Paper Fehr, Ernst, and Klaus M. Schmidt. 2004. “Fairness and 23357. Incentives in a Multi-task Principal–Agent Model.” Eisenhuth, Roland. 2013. “Reference Dependent Scandinavian Journal of Economics 106 (3): 453–74. Mechanism Design.” Unpublished. Filiz-Ozbay, Emel, and Erkut Y. Ozbay. 2007. “Auc- Eisenhuth, Roland, and Mara Ewers. 2012. “Auctions tions with Anticipated Regret: Theory and Experi- with Loss Averse Bidders.” Unpublished. ment.” American Economic Review 97 (4): 1407–18. Eliaz, Kfir, and Ran Spiegler. 2006. “Contracting with Finkelstein, Amy, and Kathleen McGarry. 2006. “Mul- Diversely Naïve Agents.” Review of Economic Stud- tiple Dimensions of Private Information: Evidence ies 73 (3): 689–714. from the Long-Term Care Insurance Market.” Amer- Eliaz, Kfir, and Ran Spiegler. 2008. “Consumer Opti- ican Economic Review 96 (4): 938–58. mism and Price Discrimination.” Theoretical Eco- Gabaix, Xavier, and David Laibson. 2006. “Shrouded nomics 3 (4): 459–97. Attributes, Consumer Myopia, and Information Sup- Eliaz, Kfir, and Ran Spiegler. 2014. “Reference-Depen- pression in Competitive Markets.” Quarterly Journal dence and Labor-Market Fluctuations.” In NBER of Economics 121 (2): 505–40. Annual 2013, edited by Jonathan Galperti, Simone. 2012. “Commitment, Flexibility, Parker and Michael Woodford, 159–200. Chicago and Optimal Screening of Time Inconsistency.” and London: University of Chicago Press. Unpublished.Garicano, Luis, and Andrea Prat. 2013. Ellingsen, Tore, and Magnus Johannesson. 2008. “Pride “ with Cognitive Costs.” and Prejudice: The Human Side of Incentive The- In Advances in Economics and : Tenth ory.” American Economic Review 98 (3): 990–1008. World Congress, Volume I, Economic Theory, edited Engelbrecht-Wiggans, Richard. 1989. “The Effect of by Daron Acemoglu, Manuel Arellano, and Eddie Regret on Optimal Bidding in Auctions.” Manage- Dekel, 342–88. Cambridge and New York: Cam- ment Science 35 (6): 685–92. bridge University Press. Englmaier, Florian, and Stephen Leider. 2012. “Con- Gervais, Simon, J. B. Heaton, and Terrance Odean. tractual and Organizational Structure with Reciprocal 2011. “Overconfidence, Compensation Contracts, Agents.” American Economic Journal: Microeconom- and Capital Budgeting.” Journal of Finance 66 (5): ics 4 (2): 146–83. 1735–77. Englmaier, Florian, and Achim Wambach. 2010. “Opti- Gneezy, Uri, and Aldo Rustichini. 2000. “A Fine Is a 1116 Journal of Economic Literature, Vol. LII (December 2014)

Price.” Journal of Legal Studies 29 (1): 1–17. Flat-Rate Tariffs.” Journal of the European Economic Gottlieb, Daniel. 2008. “Competition over Time- Association 11 (2): 399–432. Inconsistent Consumers.” Journal of Public Eco- Herweg, Fabian, Daniel Müller, and Philipp Wein- nomic Theory 10 (4): 673–84. schenk. 2010. “Binary Payment Schemes: Moral Gottlieb, Daniel, and Kent Smetters. 2012. “Narrow Hazard and Loss Aversion.” American Economic Framing and Life Insurance.” National Bureau of Review 100 (5): 2451–77. Economic Research Working Paper 18601. Holmstrom, Bengt. 1979. “Moral Hazard and Observ- Grossman, Sanford J., and Oliver D. Hart. 1986. “The ability.” Bell Journal of Economics 10 (1): 74–91. Costs and Benefits of Ownership: A Theory of Ver- Holmstrom, Bengt, and . 1991. “Multi- tical and Lateral Integration.” Journal of Political task Principal–Agent Analyses: Incentive Contracts, Economy 94 (4): 691–719. Asset Ownership, and Job Design.” Journal of Law, Grubb, Michael D. 2009. “Selling to Overconfident Economics, and Organization 7 (Special Issue): Consumers.” American Economic Review 99 (5): 24–52. 1770–1807. Immordino, Giovanni, Anna Maria C. Menichini, and Grubb, Michael D. 2012. “Consumer Inattention and Maria Grazia Romano. 2011. “A Simple Impossibil- Bill-Shock Regulation.” Massachusetts Institute of ity Result in Behavioral Contract Theory.” Economics Technology Sloan School Working Paper 4987-12. Letters 113 (3): 307–09. Gul, Faruk, and Wolfgang Pesendorfer. 2001. “Temp- Inderst, Roman, and Marco Ottaviani. 2012. “Finan- tation and Self-Control.” Econometrica 69 (6): cial Advice.” Journal of Economic Literature 50 (2): 1403–35. 494–512. Hahn, Jong-Hee, Jinwoo Kim, Sang-Hyun Kim, Itoh, Hideshi. 2004. “Moral Hazard and Other-Regard- and Jihong Lee. 2010. “Screening Loss Averse ing Preferences.” Japanese Economic Review 55 (1): Consumers.” http://papers.ssrn.com/sol3/papers. 18–45. cfm?abstract_id=1678825. Jehiel, Philippe. 2005. “Analogy-Based Expectation Harrison, J. Michael, and David M. Kreps. 1978. Equilibrium.” Journal of Economic Theory 123 (2): “Speculative Investor Behavior in a Stock Market 81–104. with Heterogeneous Expectations.” Quarterly Jour- Jehiel, Philippe, and Laurent Lamy. 2012. “Absolute nal of Economics 92 (2): 323–36. Auctions and Secret Reserve Prices: Why Are They Hart, Oliver. 2009. “Hold-Up, Asset Ownership, and Used?” Unpublished. Reference Points.” Quarterly Journal of Economics Jofre, Alejandro, Sofia Moroni, and Andrea Repetto. 124 (1): 267–300. 2012. “Dynamic Contracts under Loss Aversion.” Hart, Oliver, and Bengt Holmstrom. 2010. “A Theory Unpublished. of Firm Scope.” Quarterly Journal of Economics 125 Kahneman, Daniel, and Amos Tversky. 1979. “Prospect (2): 483–513. Theory: An Analysis of Decision under Risk.” Econo- Hart, Oliver, and John Moore. 1990. “Property Rights metrica 47 (2): 263–92. and .” Journal of Political Kamenica, Emir. 2012. “Behavioral Economics and Economy 98 (6): 1119–58. Psychology of Incentives.” Annual Review of Eco- Hart, Oliver, and John Moore. 2008. “Contracts as Ref- nomics 4: 427–52. erence Points.” Quarterly Journal of Economics 123 Kamenica, Emir, and Matthew Gentzkow. 2011. (1): 1–48. “Bayesian Persuasion.” American Economic Review Heidhues, Paul, and Botond Ko˝szegi. 2008. “Com- 101 (6): 2590–615. petition and Price Variation When Consumers Are Kamenica, Emir, Sendhil Mullainathan, and Richard Loss Averse.” American Economic Review 98 (4): Thaler. 2011. “Helping Consumers Know Them- 1245–68. selves.” American Economic Review 101 (3): 417–22. Heidhues, Paul, and Botond Ko˝szegi. 2010. “Exploiting Korobkin, Russell. 2003. “, Naivete about Self-Control in the Credit Market.” ­Standard Form Contracts, and Unconscionability.” American Economic Review 100 (5): 2279–303. University of Chicago Law Review 70 (4): 1203–95. Heidhues, Paul, and Botond Ko˝szegi. 2013. “Seller Kosfeld, Michael, and Ulrich Schüwer. 2011. “Add-on Information about Consumer Naivete Lowers Wel- Pricing, Naïve Consumers, and the Hidden Welfare fare.” Unpublished. Costs of Education.” Centre for Heidhues, Paul, Botond Ko˝szegi, and Takeshi Mur- Research Discussion Paper 8636. ooka. 2012. “Inferior Products and Profitable Decep- Kosfeld, Michael, and Ferdinand A. von Siemens. 2011. tion.” Unpublished. “Competition, Cooperation, and Corporate Culture.” Herold, Florian. 2010. “Contractual Incompleteness as RAND Journal of Economics 42 (1): 23–43. a Signal of Trust.” Games and Economic Behavior 68 Ko˝szegi, Botond, and Matthew Rabin. 2006. “A Model (1): 180–91. of Reference-Dependent Preferences.” Quarterly Herweg, Fabian, and Klaus M. Schmidt. 2012. “A Journal of Economics 121 (4): 1133–65. Theory of Ex Post Inefficient Renegotiation.” Ko˝szegi, Botond, and Matthew Rabin. 2007. “Refer- Unpublished. ence-Dependent Risk Attitudes.” American Eco- Herweg, Fabian, and Konrad Mierendorff. 2013. nomic Review 97 (4): 1047–73. “Uncertain Demand, Consumer Loss Aversion, and Ko˝szegi, Botond, and Matthew Rabin. 2009. “Refer- Ko˝szegi: Behavioral Contract Theory 1117

ence-Dependent Consumption Plans.” American ics 116 (1): 121–60. Economic Review 99 (3): 909–36. O’Donoghue, Ted, and Matthew Rabin. 2003. “Study- Kucuksenel, Serkan. 2012. “Behavioral Mechanism ing Optimal Paternalism, Illustrated by a Model Design.” Journal of Public Economic Theory 14 (5): of Sin Taxes.” American Economic Review 93 (2): 767–89. 186–91. Laibson, David. 1997. “Golden Eggs and Hyperbolic Oster, Emily, Ira Shoulson, and E. Ray Dorsey. 2013. Discounting.” Quarterly Journal of Economics 112 “Optimal Expectations and Limited Medical Testing: (2): 443–78. Evidence from Huntington Disease.” American Eco- Lal, Rajiv, and Carmen Matutes. 1994. “Retail Pricing nomic Review 103 (2): 804–30. and Advertising Strategies.” Journal of Business 67 Rabin, Matthew. 1993. “Incorporating Fairness into (3): 345–70. Game Theory and Economics.” American Economic Lange, Andreas, and Anmol Ratan. 2010. “Multi- Review 83 (5): 1281–1302. dimensional Reference-Dependent Preferences Rabin, Matthew. 1998. “Psychology and Economics.” in Sealed-Bid Auctions—How (Most) Laboratory Journal of Economic Literature 36 (1): 11–46. Experiments Differ From the Field.” Games and Rey-Biel, Pedro. 2008. “Inequity Aversion and Team Economic Behavior 68 (2): 634–45. Incentives.” Scandinavian Journal of Economics 110 List, John A. 2007. “On the Interpretation of Giving in (2): 297–320. Dictator Games.” Journal of Political Economy 115 Roider, Andreas, and Patrick W. Schmitz. 2012. “Auc- (3): 482–93. tions with Anticipated Emotions: Overbidding, Macera, Rosario. 2012. “Intertemporal Incentives Underbidding, and Optimal Reserve Prices.” Scandi- under Loss Aversion.” Unpublished. navian Journal of Economics 114 (3): 808–30. Malmendier, Ulrike, and Young Han Lee. 2011. “The Rosenkranz, Stephanie, and Patrick W. Schmitz. 2007. Bidder’s Curse.” American Economic Review 101 (2): “Reserve Prices in Auctions as Reference Points.” 749–87. Economic Journal 117 (520): 637–53. Manove, Michael, and A. Jorge Padilla. 1999. “Banking Rothschild, Michael, and . 1976. “Equi- (Conservatively) with Optimists.” RAND Journal of librium in Competitive Insurance Markets: An Essay Economics 30 (2): 324–50. on the Economics of Imperfect Information.” Quar- Maskin, Eric, and Jean Tirole. 1999. “Unforeseen Con- terly Journal of Economics 90 (4): 629–49. tingencies and Incomplete Contracts.” Review of Salant, Yuval, and Ron Siegel. 2013. “Framing in Mech- Economic Studies 66 (1): 83–114. anism Design.” Unpublished. Matsushima, Hitoshi. 2008. “Role of Honesty in Full Sandroni, Alvaro, and Francesco Squintani. 2007. Implementation.” Journal of Economic Theory 139 “Overconfidence, Insurance, and Paternalism.” (1): 353–59. American Economic Review 97 (5): 1994–2004. Milgrom, Paul R., and Robert J. Weber. 1982. “A The- Sandroni, Alvaro, and Francesco Squintani. 2010. ory of Auctions and Competitive Bidding.” Econo- “Overconfidence and Asymmetric Information: The metrica 50 (5): 1089–1122. Case of Insurance.” Unpublished. Moore, John. 1992. “Implementation in Environ- Santos-Pinto, Luís. 2008. “Positive Self-Image and ments with Complete Information.” In Advances in Incentives in Organisations.” Economic Journal 118 Economic Theory, edited by Jean-Jacques Laffont, (531): 1315–32. 181–282. Cambridge and New York: Cambridge Schumacher, Heiner. 2012. “Insurance, Self-Control University Press. and Commitment.” Unpublished. Morgan, John, Ken Steiglitz, and George Reis. 2003. Schweizer, Nikolaus, and Nora Szech. 2013. “Opti- “The Spite Motive and Equilibrium Behavior in Auc- mal Revelation of Life-Changing Information.” tions.” B. E. Journal of Economic Analysis and Policy Unpublished. 2 (1): 1–27. Shunda, Nicholas. 2009. “Auctions with a Buy Price: Mullainathan, Sendhil, Joshua Schwartzstein, and Wil- The Case of Reference-Dependent Preferences.” liam J. Congdon. 2012. “A Reduced-Form Approach Games and Economic Behavior 67 (2): 645–64. to Behavioral Public Finance.” Annual Review of Sliwka, Dirk. 2007. “Trust as a Signal of a Social Norm Economics 4: 511–40. and the Hidden Costs of Incentive Schemes.” Ameri- Murooka, Takeshi. 2013. “Deception under Competi- can Economic Review 97 (3): 999–1012. tive Intermediation.” Unpublished. Spiegler, Ran. 2006. “The Market for Quacks.” Review Netzer, Nick, and Armin Schmutzler. 2012. “Explain- of Economic Studies 73 (4): 1113–31. ing Gift-Exchange—The Limits of Good Intentions.” Spiegler, Ran. 2011. Bounded Rationality and Indus- Unpublished.O’Donoghue, Ted, and Matthew Organization. Oxford and New York: Oxford Rabin. 1999a. “Doing It Now or Later.” American University Press. Economic Review 89 (1): 103–24. Spinnewijn, Johannes. 2012. “Unemployed but Opti- O’Donoghue, Ted, and Matthew Rabin. 1999b. “Incen- mistic: Optimal Insurance Design with Biased tives for Procrastinators.” Quarterly Journal of Eco- Beliefs.” Unpublished. nomics 114 (3): 769–816. Spinnewijn, Johannes. 2013. “Insurance and Percep- O’Donoghue, Ted, and Matthew Rabin. 2001. “Choice tions: How to Screen Optimists and Pessimists.” Eco- and Procrastination.” Quarterly Journal of Econom- nomic Journal 123 (569): 606–33. 1118 Journal of Economic Literature, Vol. LII (December 2014)

Stoughton, Neal M., Youchang Wu, and Josef Zechner. Warren, Patrick L., and Daniel H. Wood. 2010. “Will 2011. “Intermediated Investment Management.” Governments Fix What Markets Cannot? The Posi- Journal of Finance 66 (3): 947–80. tive Political Economy of Regulation in Markets with Thaler, Richard H., and Shlomo Benartzi. 2004. “Save Overconfident Consumers.” Unpublished. More Tomorrow™: Using Behavioral Economics to Increase Employee Saving.” Journal of Political Economy 112 (Special Issue 1): S164–87. Thaler, Richard H., and Cass R. Sunstein. 2003. “Liber- tarian Paternalism.” American Economic Review 93 (2): 175–79. Thaler, Richard H., and Cass R. Sunstein. 2008. Nudge: Improving Decisions about Health, Wealth, and Happiness. New York: Penguin Books. Tirole, Jean. 2009. “Cognition and Incomplete Con- tracts.” American Economic Review 99 (1): 265–94. Tversky, Amos, and Daniel Kahneman. 1991. “Loss Aversion in Riskless Choice: A Reference-Depen- dent Model.” Quarterly Journal of Economics 106 (4): 1039–61. von Siemens, Ferdinand A. 2011a. “Heterogeneous Social Preferences, Screening, and Employ- ment Contracts.” Oxford Economic Papers 63 (3): 499–522. von Siemens, Ferdinand A. 2011b. “Intention-Based Reciprocity and the Hidden Costs of Control.” Unpublished. von Siemens, Ferdinand A. 2012. “Social Preferences, Sorting, and Competition.” Scandinavian Journal of Economics 114 (3): 780–807.