AEA Papers and Proceedings 2018, 108: 1–21 https://doi.org/10.1257/pandp.20181124

RICHARD T. ELY LECTURE

Private Paternalism, the Commitment Puzzle, and Model-Free Equilibrium†

By David Laibson*

College instructors adopt course policies that marketing materials would boast: “Choose our force students to focus on their coursework: pop University because we have classrooms enabled quizzes, classroom attendance requirements, with Wi-Fi blocking.” Colleges are more likely cold calling, graded problem sets,1 deadlines,2 to discuss national rankings, famous graduates, classroom Wi-Fi blocking, and classroom lap- student endorsements, recreation centers, climb- top bans.3,4 In my experience, most students ing walls, saunas, juice bars, golf simulators, don’t welcome these paternalistic restrictions. driving ranges, ropes courses, water parks, and The unpopularity of such teaching policies is lazy rivers.5 Even colleges that market their implicitly revealed by the way that colleges academic strength talk about research prowess market themselves: public relations campaigns and Nobel Laureates instead of explaining how don’t mention their paternalistic policies. No effort and focus is coaxed from students. Despite the lack of student enthusiasm, most colleges have embraced programs of private * Harvard University. This paper describes the content paternalism. I define paternalism as a policy that of the Richard T. Ely Lecture, presented at the ASSA meet- advances an individual’s interests by restrict- ing on January 5, 2018. Omeed Maghzian provided excel- lent research assistance. I also received helpful advice from ing his or her freedom. Private paternalism is Dan Benjamin, John Beshears, Olivier Blanchard, James paternalism implemented by private institu- Choi, Xavier Gabaix, Matthew Gentzkow, Brigitte Madrian, tions.6 Universities aren’t unique in their use of Matthew Rabin, Ran Spiegler, and colleagues at the Hebrew paternalistic policies. Many other private orga- University, the University of Pennsylvania CHIBE , and the ASSA annual meeting. I gratefully acknowledge( )financial nizations deploy paternalistic policies, includ- support from the National Institute on Aging P01AG005842, ing firms in thefor-profit ­ sector. Such private P30AG034532, R01AG021650 and the Pershing( Square paternalism is often shrouded, which suggests Fund for Research on the Foundations) of Human Behavior. † that these organizations’ workers customers Go to https://doi.org/10.1257/pandp.20181124 to visit students members do not explicitly/ demand/ the article page for additional materials and author disclo- / sure statement. paternalism. A firm doesn’t recruit workers by 1 In my undergraduate classes, I now give two small prob- telling them how it’s going to limit their free- lem sets each week. dom. But firms do limit their workers’ freedom 2 For example, a term paper might have more than one with intermediate deadlines, progress reports, binding deadline: e.g., a deadline for a topic choice, a deadline for an initial draft, and a deadline for a final draft. production targets, “attendance” requirements Likewise, doctoral students are required to meet many inter- during working hours, frequent evaluations, and mediate milestones each with a deadline before receiving their degree: general( exams, field exams,) dissertation pro- spectus, ­semester-frequency seminar presentations, etc. 3 Dan Rockmore, “The Case for Banning Laptops 5 Courtney Rubin, “Making a Splash on Campus: College in the Classroom.” New Yorker, June 6, 2014 Recreation Now Includes Pool Parties and River Rides.” https://www.newyorker.com/tech/elements/the-case-for- New York Times, September 19, 2014 https://www.nytimes. banning-laptops-in-the-classroom( . com/2014/09/21/fashion/college-recreation-now-includes-( 4 If externalities were the sole) problem with laptops, pool-parties-and-river-rides.html . then those who wish to use laptops could be allowed to 6 I contrast private paternalism) with libertarian paternal- ­self-segregate into some area of the classroom, where their ism Thaler and Sunstein 2003 in Section VIA. In a nut- screens wouldn’t be visible to those who prefer not to be shell,( private paternalism is ­choice-restricting,) unlike the near laptop users. pure form of libertarian paternalism. 1 2 AEA PAPERS AND PROCEEDINGS MAY 2018 mandatory health retirement benefits. Firms modern work arrangements have continued in could move toward/ more of a piece rate system this vein: “When we look at the organization of i.e., the worker is able to produce as much or as work from the perspective of the twentieth cen- little( as they want on a linear payment schedule , tury, the prevailing system, factory discipline, but, by and large, formal sector firms don’t use) seems the natural and timeless way of organiz- this compensation system. Firms look more like ing work. Under factory discipline workers face classrooms in which bosses teachers success- a very constrained choice. In return for their fully coax productivity out of/ workers students wage, they surrender to the employer complete and then reward them for their high output./ 7 command of their labor for a fixed period each Private paternalism is the first half of a day. The employer sets the pace of work and phenomenon that I call the commitment puz- also dictates how workers will conduct them- zle: i private institutions help people solve selves on the job.”10 ­self-control( ) problems by using lots of forc- The specific approach that I take follows an ing mechanisms i.e., private paternalism and argument proposed by O’Donoghue and Rabin ii these institutions( don’t market advertise) 1999b . They characterize the socially effi- these( ) forcing mechanisms, so the mechanisms/ cient( incentives) that organizations should use if remain at least partially shrouded. These forcing those organizations were motivated to help naïve mechanisms are shrouded commitment tech- ­present-biased agents avoid procrastination. nologies that are indirectly “chosen” by agents They assume that organizations are interested who voluntarily join organizations that limit in maximizing their reputations as employ- their freedom. In this lecture, I explain why ers, arguing that “reputational pressures may such private paternalism is an equilibrium, even induce firms to offer incentive contracts that without explicit demand from its beneficiaries are ex post acceptable to agents, which would e.g., workers, customers, students . As a corol- imply that firms wish to induce efficient behav- lary,( I explain why private paternalism) is typi- ior.”11 O’Donoghue and Rabin’s mechanism cally shrouded. I describe the benefits of private will serve as the fulcrum of this lecture: when paternalism relative to public paternalism and prospective employees use ­backward-looking identify the domains where private paternalism “­model-free” data to compare employers, will fail to be socially efficient. firms( have socially) efficient incentives to cre- The ideas in this lecture are related to the ate shrouded forcing commitment technologies arguments made by Clark 1994 . Clark views that coax high effort /out of naïve present-biased­ the work arrangements that( arose) during the workers, which in turn engenders high levels of Industrial Revolution as an implicit solution to compensation in competitive equilibrium. I call ­self-control problems. “Whatever the workers this a ­model-free equilibrium because agents themselves thought, they effectively hired the make forecasts about their future payoffs by capitalists to discipline and coerce them. Even using historical payoff data and not by using in the factories of the Industrial Revolution a structural model of how they will respond to they were the ultimate masters of their fate, but future incentives. weakness of will meant they delegated that mas- The ­model-free equilibrium that I analyze is tery to the capitalists.”8,9 Clark also notes that also related to the literature on fictitious play Brown 1951; Fudenberg and Levine 1998 . The( fictitious play equilibrium concept assumes) 7 Cadena et al. 2011 report the results of a field exper- that agents playing a dynamic game use the iment, run collaboratively( ) with a financial firm, which pur- sued these goals. 8 Clark 1994, p. 131 . 9 This lecture( was named) by the AEA after Richard Legacy of Progressive Economics.” Democracy, March T. Ely, the organization’s founder and first Secretary, 15, 2016 https://democracyjournal.org/arguments/ who was also an early leader of the Progressive move- the-exclusionist-legacy-of-progressive-economics/( . ment. Relative to Clark, Ely took a far more critical view 10 Clark 1994, p. 131 . ) of nineteenth century factory work conditions. However, 11 O’Donoghue( and Rabin) 1999b, p. 783 . There is a Ely will not be remembered only as a champion of work- large economics literature on the( role of worker )and firm rep- ers’ rights. His writings are also fraught with eugenicism utations: e.g., Kanemoto and MacLeod 1992 , Holmström and racism. For analysis of his views, see: Bernard A. 1999 , Levin 2003 , Ely and Välimäki( 2003) , and Halac Weisberger and Marshall Steinbaum, “The Exclusionist and( Prat) 2016(. ) ( ) ( ) VOL. 108 RICHARD T. ELY LECTURE 3 past play of their opponents as a forecast of I. Present-Biased Discounting and an Effort Task those opponents’ future play. This experien- tial forecasting rule bears some resemblance In this lecture I focus on present bias, which to the model-free­ equilibrium that I will ana- is a specific bias that private paternalism can off- lyze. In both fictitious play and ­model-free set. In Section VIID I discuss the generalization equilibria, agents use historical data to pre- of private paternalism to a wider class of psy- dict future outcomes, without having a com- chological biases. plete structural model of strategic play. In the Time is indexed by t​​ , with t​ 0, 1, 2, . . . , T ​ , ­model-free equilibria that I study, historical and and that agents have intertemporal∈ { preferences} experiential data are used to predict one’s own characterized by present bias:12 future payoffs, rather than the actions of other 2 3 players. 1 ​​Ut​​ ​ut​​ ​u​t 1​ ​ ​​ ​ ​ut​ 2​ ​ ​​ ​ ​ut​ 3​ Section I introduces an illustrative model ( ) = + β δ + + β δ + + β δ + T t that I will use throughout this lecture. I assume ​ ​​ − ​ ​u​T​ that agents have present bias, which induces + ⋯ + β δ 2 3 a self-control­ problem. Present bias without 2 ​u​t​ ​[ ​ut​ 1​ ​ ​​ ​ ​ut​ 2​ ​ ​​ ​ ​ut​ 3​ commitment leads to low, socially inefficient( ( ) = + β δ + + δ + + δ + ) T t levels of effort. This section also describes the ​​​ − ​ ​uT​ ]​ ​.​ normative assumptions that are made throughout + ⋯ + δ this paper. Section II characterizes equilibrium Accordingly, from self t​​’s perspective, one util in an economy with ­forward-looking, sophisti- at date t​ 1​ is worth ​ ​ utils at date ​t.​ However, cated ­present-biased agents. Section III charac- one util at+ date ​t 2​ isβ worth δ utils at date ​t 1.​ terizes equilibrium, including ­employer-based Present bias is a +wedge betweenδ​ ​ the present+ and exploitation of workers, in an economy with the future, but, from today’s perspective, it does ­forward-looking, naïve present-biased­ agents. not affect trade-offs between two future dates. Section IV describes the commitment puzzle: The present bias parameter is ,​ and present the existence of shrouded private paternalism. bias is ​1 ​ the “extra” discountingβ​ between Section V resolves the commitment puzzle by the present− βand( the future . Accordingly, when ​ assuming that people choose employers by 1,​ discounting is purely) exponential, and using backward-looking­ experienced utility thereβ = is no present bias. instead of using forward-looking­ predicted Present bias induces dynamically inconsistent utility. I use the terms ­model-based forecast- preferences. For example, the preferences from ing and ­model-free forecasting to respec- the perspective of self t​​ don’t agree with the pref- tively distinguish the forward-looking­ and erences of self ​t 1.​ Self ​t 1​ has preferences ­backward-looking equilibrium concepts that + + I study e.g., Dayan and Niv 2008 . I call the 2 ( ) 3 ​​U​t 1​ ​u​t 1​ ​u​t 2​ ​ ​​ ​ ​ut​ 3​ equilibrium based on backward-looking­ fore- ( ) + = + + β δ + + βδ + casts a ­model-free equilibrium because it does 3 T t 1 ​ ​​ ​ ​ut​ 4​ ​ ​​ −( + )​ ​uT​ ​ not rely on a structural theory of one’s own + βδ + + ⋯ + βδ future behavior. The model contains both a for- mal sector, where forcing mechanisms endog- 2 3 4 ​ut​ 1​ ​ ​ut​ 2​ ​ ​​ ​ ​ut​ 3​ ​ ​​ ​ ​ut​ 4​ enously arise in equilibrium, and an informal ( ) = + + β[δ + + δ + + δ + household sector, where forcing mechanisms T ​ t 1 ​ ​​​ −( + )​ ​uT​ ​ ​. (have no traction.) Section VI relates these results + ⋯ + δ ] to the concept of private paternalism, pater- nalism implemented by private institutions. Section VII discusses the limits and scope of private paternalism, including a discussion of 12 See Strotz 1955 , Phelps and Pollak 1968 , Akerlof 1991 , Laibson( 1997) , and O’Donoghue( and) Rabin the relative strengths of public versus private ( ) ( ) paternalism, and the difficulty of generating 1999a . Laibson 1997 uses this functional form to cap- ture( the) properties (of hyperbolic) discounting see Chung and efficient private paternalism in the informal Herrnstein 1967; Ainslie 1975, 1992; and Loewenstein( and sector. Section VIII concludes and discusses Prelec 1992 . See Cohen et al. 2017 for a review of the directions for future work. intertemporal) choice literature. ( ) 4 AEA PAPERS AND PROCEEDINGS MAY 2018

From self ​t 1 ​’s perspective, one util at date ​ However, when she plans at date t​​ for her effort t 2​ is worth( + ​ )​ utils at date ​t 1,​ contradict- level at date t​ 1,​ she would prefer to maximize ing+ the preferencesβ δ of self ​t.​ + this objective+ In general, I assume that ​ 1​ and ​ ​ is close β < δ to 1. For a “daily model” in which each time _1 2 max​ ​ ​ ​ ​ ​ e​ t 1​ ​e​t 1​ ​.​ ( ​et​ 1​ period is one day , ​ ​ will be very close to unity. + β(−2 + + + ) With this case in) δmind, I assume ​ 1​ and ​ 1.​ Setting ​ 1​ also simplifiesβ the< nota- The associated FOC is tion.δ = I’ll note whereδ = this assumption affects my † argument. ​(​ ​e​ t 1​ 1)​ 0,​ From a normative perspective, it is common β − + + = † to treat present bias as a bias and not a norma- where ​​e​ t 1 ​ is her desired future effort level. It tive preference and strip it out( of the planner’s follows that+ objective.13 Indeed,) even the agent agrees with † ∗ this perspective when it comes to trade-offs that ​​e​ t 1 ​ 1 ​e​ t 1​ ​ .​ involve all future selves. Accordingly, when we + = > β = + ∗ discuss normative benchmarks we will assume Note that ​​e​ t 1​ ​ ​ for the same reason that​​ ∗ + = β that the planner’s objective is e​ t​ ​ .​ When= β ​ 1,​ self ​t​ and self ​t 1​ fully agree β = + 2 3 T on the best course of action for period t​ 1.​ ​U ​u0​​ ​u​1​ ​ ​​ ​ ​u2​​ ​ ​​ ​ ​u3​​ ​​​ ​ ​uT​ ​ .​ + = + δ + δ + δ + ⋯ + δ When ​ 1,​ self ​t​ wants self ​t 1​ to work β† < + hard ​ ​e​ t 1 ​ 1 ​ and self t​ 1​ prefers to work ( + =∗ ) + A. An Illustrative Example less hard ​ ​e​ t 1​ ​ 1 .​ This is the essence of the dynamic( +inconsistency= β < ) problem that afflicts Consider the following simple model of ­present-biased agents: self t​​ and self t​ 1​ don’t effort. When an agent exerts effort ​e​ , she will agree on how to behave. + experience immediate effort cost measured ( in utils of ​ ​ 1 ​ ​ e​​ 2​​ and she will experience a B. Beliefs about the Choices of Future Selves ) −_2 ­one-period delayed reward measured in utils ( ) Since the beginning of the contemporary liter- of ​e.​ Assume that effort is exerted one period 15 before the reward is experienced. I also assume ature on dynamically inconsistent preferences _ Strotz 1955 , there has been active discussion that e​​ has an upper bound, ​​ e ​,​ and this_ upper ( ) bound is larger than ​ .​ In other words, ​​ e ​ .​ 14 of the extent to which agents correctly antici- β > β pate their own future behavior. At one extreme, At date ​t,​ the worker’s current objective is sophisticated agents understand that their future selves will hold preferences that contradict the _1 2 preferences of the current self. In the illustrative ​​max​ ​ ​ ​ ​ e​ t​ ​ ​e​t​ ​.​ ​et​​ (−2 + β ) example that I have presented, early sophisti- cated selves will choose to commit later selves This implies that her effort level will be given by to exert effort ​e 1​ in every future period if the first-order condition FOC commitment is free.= If commitment is not free ( ) or if ­non-contractible uncertainty makes com- ( ​ ​e​ ∗​ ​ 0,​ mitment counterproductive , such commitments − t + β = may not be chosen see Laibson) 2015 . ( ) ∗ At the other extreme, naïve agents believe where ​​e​ t​ ​​ is her equilibrium level of effort. Accordingly, that their future selves will hold preferences that match the preferences of the current self ​​e​ ∗​ ​ .​ with respect to utility trade-offs in the future . t = β ( )

15 Many earlier scholars, including Adam Smith and 13 For example, see O’Donoghue and Rabin 1999b , David Hume, also understood these psychological forces and Choi et al. 2003 , or Carroll et al. 2009 . ( ) wrote about them. See Ashraf, Camerer, and Loewenstein 14 This bound( )will play a role in( Section) IIIB. 2005 and Cohen et al. 2017 for selected references. ( ) ( ) VOL. 108 RICHARD T. ELY LECTURE 5

Accordingly, naïve agents have no motive to Second, the experimental participants who are commit their future selves, unless commitment willing to make commitments are rarely will- is the ancillary undesired consequence of an ing to pay to make those commitments. In other action that is actually( motivated) by some other words, some experimental participants are will- goal: e.g., pre-purchasing­ theatre tickets to a ing to restrict their choice set, but these subjects musical that is likely to sell out. are usually not willing to pay for this privilege.18 O’Donoghue and Rabin 1999a introduced If one counts only pure commitment mecha- the concept of partially naïve( agents) who have a nisms that have not been created by behavioral partial understanding of their dynamic inconsis- economists,19 the list is small. ­Market-based tency. Specifically, partially naïve agents believe examples include web-blocking­ software e.g., that future selves will have present bias char- the Freedom app and alarm clocks that (jump ) acterized by a ­present-bias parameter ​​ˆ ,​ where ​​ off nightstands.20 It is not easy to expand this β ​ ˆ ​ is strictly greater than the true value of ​ .​ As ​​ list if one keeps the conceptual bar high i.e., β​ β pure commitments, as opposed to commitments( ˆ ​ falls to ​ ,​ partially naïve agents become fully β​ β that have ancillary benefits that may explain the sophisticated. As ​​ ˆ ​ rises to 1,​ ​ partially naïve β ​ ­choice-reducing behavior . In Korea, consum- agents become fully naïve. Because partially ers turn down free credit )Cho and Rust 2017 . naïve agents recognize some of their dynamic Many people delete apps from( their smartphone) inconsistency ​ ​ ​​ ˆ 1​, they will also choose (β < β ​< ) or iPad to force themselves to stop using the app commitment devices when commitment is free. rather than freeing up memory .21 The list of pure( commitments is not long. ) II. Sophistication and Commitment When commitments exist in real markets, they are usually embedded in a broader set of If agents are sufficiently sophisticated, mechanisms policies, so they aren’t pure, and ­non-contractible uncertainty is sufficiently low, they aren’t even/ perceived as ­self-generated. In and commitment technologies are sufficiently other words, there is very little commitment for inexpensive, then agents will choose pure com- commitment’s sake, but people often find them- mitment technologies Laibson 1997, 2015 . In selves working studying in institutions that use a rapidly growing literature,( economists have) forcing mechanisms:/ e.g., a firm with frequent demonstrated that there is demand for a wide progress reports, supervisor ­check-ins, and bind- range of explicit commitment technologies. ing deadlines. In this sense, people don’t appear This literature is reviewed in Bryan, Karlan, and Nelson 2010 and Cohen et al. 2017 .16 ( ) ( ) This new empirical literature has two inter- future consumption. However, this may not be commitment, esting limitations. First, almost all of the papers but instead a forecast by the consumer that she only wants demonstrating a demand for pure commitment to eat a single serving of these foods and hence shouldn’t are lab or field experiments that study com- buy in volume at a per-unit­ discount . Naïve agents who do not seek commitment( will make such) ­low-consumption( mitment mechanisms designed by behavioral forecasts. ) economists, as opposed to analysis of pure 18 Schilbach 2017 is a notable exception. Beshears et al. commitments that already exist in markets.17 2017 is also an( exception) but the amounts that are being paid( for) commitment in this study are small in most cases, less than $1 per year . ( 19 The self-control­ ) website stickK was created by behav- 16 For example, see Ariely and Wertenbroch 2002 ; ioral economists Dean Karlan and Ian Ayres . Behavioral Ashraf, Karlan, and Yin 2006 ; Houser et al. 2010(; Giné,) economists also report( using a host of commitment) behav- Karlan, and Zinman 2010( ; Kaur,) Kremer, and (Mullainathan) iors of their own design: for example, a coauthor who com- 2010, 2015 ; Royer,( Stehr,) and Sydnor 2015 ; Bisin and mits to pay a penalty if he doesn’t finish a draft on time, Hyndman( )2014 ; Augenblick, Niederle,( and) Sprenger doctoral students who create a communal fund that they pay 2015 ; Schilbach( ) 2017 ; Beshears et al. 2017 ; Alsan et into whenever they oversleep, a doctoral student who signs al.( 2017) ; and Cho( and Rust) 2017 . ( ) up for a research presentation as a commitment device, and (17 One) exception is Cho and( Rust) 2017 , who show that a faculty member who bets that he will lose a ­prespecified many Korean consumers reject free( credit.) Wertenbroch amount of weight by a deadline. 1998 is another possible exception, depending on how one 20 For the jumping alarm clocks, see https://nandahome. interprets( ) this paper. Wertenbroch studies the fact that con- com/. sumers tend to buy candy and other “sin” goods in small 21 I have saved thousands of hours by deleting a chess packages e.g., “single serving” , implicitly limiting their app from my iPad. ( ) 6 AEA PAPERS AND PROCEEDINGS MAY 2018

ˆ to be choosing to commit themselves, but rather participants’ ​ ​ to be 0.87 i.e., ​​​ Other​ ​ 0.87​, accepting restrictions on their freedom as part of implying a substantialβ degree( ofβ ​interpersonal= ) a grand bargain with certain counterparties. sophistication. Note that perfect interpersonal ˆ sophistication would correspond to ​ ​​ Other​ ​ .​ III. Naïveté, Freedom, and Exploitation By contrast, experimental participantsβ = β ​per- ˆ 23 ceive ​​​ ​Self​ 1.03​ , which implies perfect intra- Naïve agents believe that the preferences they personalβ ​ naïveté.= Fedyk’s work provides formal hold today will be identical to the preferences evidence for a hypothesis that has long been that they will hold in the future. In other words, folk wisdom. As Mark Twain put it: “Nothing naïve agents believe that their preferences are so needs reforming as other people’s habits.”24 dynamically consistent and anchored by their Indirect evidence for naïveté comes from current preferences . So( the plans they make ) the absence of a thriving commitment industry. today will be faithfully executed by their future Very few pure commitment devices are cur- selves. W. C. Fields expressed this view suc- rently being marketed, with the exception of cinctly when he said, “Now don’t say you can’t the small set of products reviewed in Section II. swear off drinking; it’s easy. I’ve done it a thou- 22 When commitment is present in markets, it is sand times.” almost always bundled with other features. So For naïve agents, there is no reason to com- the worker student consumer is not buying mit themselves because they believe their future commitment/ purely/ for commitment’s sake, selves will always agree with their current but is instead buying commitment along with selves. Naïve agents will never choose com- some other features. For example, a mortgage mitment for its own sake. All else equal, naïve that provides credit for purchasing a home and agents prefer freedom. They would never choose an ancillary system of forced savings principal to bind themselves unless there was some com- repayments . Later in this lecture, I will( explain pensating trade-off. why commitments) are almost always bundled Most of the evidence from the field supports with other products and work arrangements, and a substantial degree of naïveté, and some stud- thereby shrouded. ies even find evidence for full naïveté: e.g., In light of the evidence for naïveté, I will here- Dellavigna and Malmendier 2006 and Acland after study the extreme case of perfect naïveté and Levy 2015 , gyms; Goda( et) al. 2015 , i.e., ​ 1​ but ​​​ˆ ​ ​ 1 .​ Though it is likely retirement ( savings;) Augenblick and ( Rabin) Self (that peopleβ << have someβ ​ =­domain-specific,) partial forthcoming , experimental effort task; Levy awareness of their own( present bias, I study the) et( al. 2017 ,) smoking cessation; and Kuchler ( ) limiting case, ​​​ˆ ​ ​ 1,​ for simplicity. and Pagel 2017 , credit cards. βSelf ​ = In a pair( of experiments,) one in a large class and one online, Fedyk 2017 reports evidence A. Naïve Agents in Home Production for asymmetric naïveté. (Specifically,) people are relatively sophisticated about others’ present If agents are naïve, their behavior will depend bias, but fail to anticipate their own future pres- upon the setting in which they are operating and ent bias. In the classroom experiment, students the counterparties with whom they negotiate. systematically underestimate how late they will If the counterparty is a firm, the agent may be turn in an assignment but hold much more accu- exploited, a possibility that I consider in the next rate beliefs about their classmates’ propensity subsection. In home production, by contrast, to procrastinate. In Fedyk’s online experiment, the agent will simply fail to live up to her good participants engage in a real-effort­ task cf. intentions for high effort. Augenblick, Niederle, and Sprenger 2015, and( I now revisit the earlier example, where I Augenblick and Rabin forthcoming . Fedyk esti- showed that an agent will expect one period mates ​ 0.82​ for her sample of experimental) ahead to work with effort ​​eˆ ​ 1,​ when( in fact, participants.β = These ­participants perceive other ) =

23 This estimate is statistically indistinguishable from unity. 22 W. C. Fields 1938 , The Temperance Lecture, radio 24 Mark Twain 1894 , Pudd’nhead Wilson, Chapter XV broadcast. See Shapiro( 2006,) p. 256 . prefatory quote. Charles( )L. Webster and Company. ( ) VOL. 108 RICHARD T. ELY LECTURE 7 she will work with effort e​ .​ Accordingly, Period 2.—Worker chooses between the she anticipates payoff = β firm’s wage schedule and the worker’s outside option, which has undiscounted payoff ​z​ and is received in period 4. In other words, the work- ​ ​ _1 ​ ​​ eˆ ​​ ​2​ ​eˆ ​ _1 ​ .​ −2 + = 2 er’s choice is

However, she will consistently underperform ​max​ ​max​ ​ ​ _1 ​ ​e​​ 2​ w e ​, z ​ relative to this benchmark. Her actual undis- {β e [−2 + ( )] β } counted payoff will be ( ) max​ ​max​ ​ ​ _1 ​ ​e​​ 2​ w e ​, z ​.​ ​ ​ _1 ​ ​ ​​ 2​ .​ = { e [−2 + ( )] } −2 β + β This payoff will be strictly less than the worker’s This choice is binding for the next period.26 anticipated payoff, 1 2 as long as ​ 1 .​25 / ( β ≠ ) Period 3.—If the worker chose the firm’s B. Naïve Agents in a Monopsony Labor Market wage schedule, she chooses effort e​​ subject to present bias that she had not previously antici- If a naïve, present-biased­ agent worker inter- pated. The cost of effort is experienced in this acts with a sophisticated firm i.e.,/ a firm that period. In other words, the worker chooses ​e​ in understands the worker’s present( bias , and the following maximization problem: the worker evaluates the firm’s offers) using a ­forward-looking structural theory of her max​ ​ ​ _1 ​ ​e​​ 2​ w e ​.​ e { 2 } own future actions, then( the firm) will typically − + β ( ) exploit the agent’s naïveté and may cause the Period 4.—If the worker chose the firm’s agent to earn less than her outside option e.g., wage schedule, she experiences a utility flow O’Donoghue and Rabin 1999b . I illustrate( derived from her wage payment, ​w e ,​ or, if she this exploitative equilibrium in the) current sub- chose her outside option, she experiences( ) utility section monopsony and the subsection that flow ​z.​ follows (competitive) equilibrium . In the next section of( the paper, I critique the) underlying In equilibrium, the firm may exploit the assumptions that support these exploitative worker, even if the firm has competition in the equilibria, and explain why these equilibria will labor market see the next subsection for the sometimes not arise in practice. competitive labor( market case . In this subsec- Readers who want to see the details of the tion, I study a monopsonist firm.) 27 exploitative equilibrium should read the cur- The firm will offer a wage schedule that gen- rent subsection. Readers wishing to skip ahead erates a perceived payoff that is at least as good can take the exploitative equilibria for granted as z​​. The firm’s problem can be reduced to the Sections IIIB and IIIC , and jump to Section IV, choice of four scalar values: e​​ˆ , ​ w eˆ ​, ​e​​ ∗​, w ​e​​ ∗​.​ where( I critique such )equilibria and argue that Effort level ​​eˆ ​ is the quantity of effort(​) that( the) they describe only a subset of the relationships worker expects to exert once she starts work- between firms and workers customers. ing the expectation that is formed in period 2 . To understand exploitative/ equilibria, con- Wage( w​ eˆ ​​ is the wage that the firm will pay the) sider the following game played between a mon- worker for(​) effort e​​ˆ . ​​ Effort level e​​​​ ∗​​ is the quan- opsonist firm and a worker. The game has four tity of effort that the worker will revert to once periods, which are summarized below.

Period 1.—Firm commits to an enforceable 26 The agent can choose again once the game “ends” after wage schedule mapping effort to wage: ​w e .​ period 4. There are many unmodeled reasons to believe ( ) that such temporary stickiness( exists, )including frictional transition costs. 27 In a more general setting, Eliaz and Spiegler 2006 describe the menu of contracts offered by a monopolist( to) 1 25 Note that ​ ​ _ ​ ​ ​​ 2​ ​ is a concave parabola with a max- −2 β + β dynamically inconsistent consumers that vary in their degree imum of ​1 2​ at ​ 1.​ of naïveté. / β = 8 AEA PAPERS AND PROCEEDINGS MAY 2018 the moment to work arrives period 3 ; she is But the firm allows the worker to opt out of this unexpectedly subject to present( bias, which) she arrangement once she is on the job and revert hadn’t foreseen when joining the firm. Finally, to a lower effort( level, e​​ ​​ ∗​ ,​ and receive) wage wage w​ ​e​​ ∗​​ is the wage the firm promises to pay = β ( ) e ∗ _ the worker for effort ​​​​ ​. Firms receive output ​​ 7 ​w ​e​​ ∗​ z _1 ​ _1 ​​ 1 _1 ​ ​ e ​2​ .​ e​​ ∗​​ and pay wage w​ ​e​​ ∗​​ , so firms maximize the ( ) ( ) = + 2 β + 2( − ) following expression( subject) to a participation β constraint P and an incentive compatibility Note that this is worse than the worker would constraint (IC) :28 have done, had she originally chosen her outside ( ) option, because 5 max​ ​ ​e​​ ∗​ w ​e​​ ∗​ , ​ ​e​​ ∗​, w ​e​​ ∗​, ​eˆ ​, w eˆ ​ ( ) ( ) (​) − ( ) 1 2 1 2 1 29 ​ ​ _ ​ ​​ ​e​​ ∗​ ​ ​ w ​e​​ ∗​ ​ _ ​ ​ ​​ ​ z _ ​ subject to the constraints −2 ( ) + ( ) = −2 β + + 2 β

1 2 _ P ​ ​ _ ​ ​​ ​eˆ ​ ​ ​ w eˆ ​ z​ , _1 ​​ 1 _1 ​ ​ e ​2​ ( ) −2 ( ) + (​) ≥ + 2 − ( β) 1 2 1 2 IC ​ ​ _ ​ ​​ ​e​​ ∗​ ​ ​ w e​ ​​ ∗​ ​ _ ​ ​​eˆ ​​ ​ ​ w eˆ ​.​ z _ 1 ​ ​ 1 ​ ( ) −2 ( ) + β ( ) ≥ −2 + β (​) < + 2 β( − β) In equilibrium, the firm recruits the worker with a _1 ​​ 1 _1 ​ ​ ​​ 2​ misleading offer: work with effort ​​eˆ ​ and get paid + 2( − ) β wage w​ eˆ ​.​ The firm sets up the contract so that β (​) z.​ the naïve worker has an unanticipated incentive = to switch after taking the job to work effort e​​ ​​ ∗​,​ with payment( ​w ​e​​ ∗​​. The firm) sets ​​eˆ ​ and ​w eˆ ​​ The _inequality follows from the assump- ( ) (​) tion ​​ e ​ .​ We now consider the payoff to so the worker is indifferent between the firm’s > β 1 the firm. Output is ​ ​ and wages are ​z ​_ ​ offer and the worker’s outside option, ​z​ this is β + 2 β + ( 1 1 _ the participation constraint, P . The firm sets ​​ ​ _ ​ 1 __ ​​ e ​2​ .​ So the firm’s profits are ( )) 2 e​​ ∗​​ with payment ​w ​e​​ ∗​​ so the worker is indiffer- ( − β​)​ ( ) ent between staying with her original effort plan _ and switching to this new level of effort this ​ ​ z _1 ​ _1 ​​ 1 _1 ​ ​ e ​2​ ​ ( π = β − + 2 β + 2 − is the incentive compatibility constraint, IC . [ ( β) ] Constraints P and IC are both satisfied( at ))the 1 1 1 _2 firm’s optimum.( ) ( ) _ ​ z _ ​​ ​ _ ​ 1 ​ e ​ ​ = 2 β − + 2 − The Appendix derives the solution to the (β ) firm’s problem. The Appendix shows that the 0​. ≥ firm exploits the worker by offering a high level of compensation for the maximal level of effort. The last inequality follows from a restriction on​ Specifically, the firm offers the_ worker the high- z​ see footnote 27 . est feasible level of effort, e​​ˆ ​ e ​,​ and a suffi- ( ) ciently high wage, ​w eˆ ​ ,​ to make= this level of C. Naïve Agents in a Competitive Labor Market effort appear to be desirable(​) ­ex ante to the naïve worker: The fundamental logic of these results doesn’t change when the ­present-biased agent interacts _ with firms that compete with one another.30 6 ​w eˆ ​ z _1 ​ ​​ e ​2​ .​ ( ) (​) = + 2 Returning to the illustrative example, two key properties also emerge in competitive equilib- 28 rium: i the worker is recruited expecting to To study the case of economic interest, we assume that​ ( ) _ 1 1 1 _ work at effort level ​​eˆ ​ e ​​ and then switches to z _ ​ _ ​ ​ __ 1 ​ e ​2​ .​ This restriction on z​ ​ implies that the ≤ 2 β + 2( ​− ) = firm is able toβ offer a contract to the worker that incents the worker to choose the firm over the worker’s outside option. 29 These constraints bind in the monopsonistic equilib- 30 However, the participation constraint now may not rium. In this subsection, we will express them as equalities. bind. VOL. 108 RICHARD T. ELY LECTURE 9 a less intensive effort level e​​​​ ∗​ ​ , and ii the D. Exploitative Contracts in Practice worker may end up with less =thanβ her outside( ) option. In competitive equilibrium I have the Such exploitative contracts frequently do same setup timing, but now there is more than arise, especially in product markets. For exam- one firm with/ Bertrand competition , so the ple, Dellavigna and Malmendier 2004, 2006 firms make( zero profit in equilibrium. ) and Shui and Ausubel 2005 convincingly( argue) This equilibrium causes worker payoffs to that exploitative contracts( arise) in the following rise weakly , but the analysis from the pre- service and goods markets: consumer credit vious( subsection) doesn’t change. To distin- e.g., credit cards , gambling, health clubs, guish the previous monopsony wage mapping life( insurance, mail) order businesses, mobile from the current competitive equilibrium wage phones, and vacation ­time-sharing. A rich litera- mapping, I let ​​w​​ c​ e ​ represent the competi- ture has developed to study the many ways that tive equilibrium wage( ) mapping. In competitive firms interact with naïve agents, including the equilibrium firms need to disgorge all of their role for regulation.32 rents in the second­ stage of the relationship. So However, exploitative equilibria of the sort ​​w​​ c​ ​e​​ ∗​ .​ Then incentive compatibility characterized in the previous subsections seem implies( ) that= β counterfactual. Workers rarely complain that they ended up working less hard then they thought ​ ​ _1 ​ ​​ ​e​​ ∗​ ​ 2​ ​w​​ c​ ​e​​ ∗​ ​ _1 ​ ​ ​​ 2​ ​ ​​ 2​ they would work and that this “underworking” −2 ( ) + β ( ) = −2 β + β _ _ is exploitative. Rather, workers complain about ​ _1 ​ ​​ e ​2​ ​w​​ c​ ​ e ​.​ being overworked. So the exploitative equilib- 2 = − + β ( ) rium in the previous subsection seems largely Accordingly, counterfactual as applied to labor markets. _ _ w​​ c​ ​ e ​ _ 1 ​ _1 ​ ​​ e ​2​ .​ IV. The Commitment Puzzle ( ) = 2 β + 2 β We can compare the realized competitive Until now, I have emphasized the perverse equilibrium wage to the realized wage in the properties of equilibria with naïve ­present-biased monopsony equilibrium. From equation 7 the agents. These equilibria feature misperceptions monopsony realized wage is ( ) e.g., bait and switch and exploitation. Such perverse( equilibria show) up in many product _ markets e.g., Dellavigna and Malmendier 2004, ​w ​e​​ ∗​ z _1 ​ _1 ​​ 1 _1 ​ ​ e ​2​ ( ( ) = + 2 β + 2( − ) 2006 , but there are other domains, especially β labor) markets, where such exploitation seems to ​w​​ c​ ​e​​ ∗​.​ ≤ β = ( ) be the exception rather than the rule. As discussed at the end of Section IIIB, the So ​​w​​ c​ ​e​​ ∗​​ has weakly risen relative to the exploitative equilibria are characterized by the realized( wage) in the monopsonist case.31 See expectation of high effort and high pay, followed Laibson and Yariv 2007 for a similar analy- by an unanticipated propensity for workers to sis, which shows that( firms) do not make excess switch to low effort and low pay. However, there ­profits in competitive equilibrium, even when are many firms where the work­environment agents are ­present-biased and naïve. is designed to coax high effort out of workers instead of tricking them into slacking off. Many work environments are carefully designed to 31 A similar inequality applies to the anticipated wage in competitive equilibrium. From equation 6 the anticipated monopsony wage is ( ) 32 For example, see Ausubel 1991 ; Hall 1997 ; Eliaz _1 _2 ​w eˆ ​ z ​ ​​ e ​ ​ and Spiegler 2006, 2008 ; Gabaix( and) Laibson( )2006 ; (​) = + 2 ( ) ( ) _ _ Gottlieb 2008 ; Bucks and Pence 2008 ; Stango and _ 1 ​ _1 ​​ ​ _1 ​ 1 ​ e ​2​ _1 ​ ​​ e ​2​ ( ) ( ) ≤ 2 β + 2( − ) + 2 Zinman 2009 ; Grubb 2009, 2015 ; Heidhues and Ko˝szegi β 2010, 2015,( 2017) ; Spiegler( 2011 ); Armstrong and Vickers _ ( ) ( ) _ 1 ​ _1 ​ ​​ e ​​​ 2​ 2012 ; Gerardi, Goette, and Meier 2013 ; Warren and = 2 β + 2 ( ) ( ) β Wood 2014 ; Grubb and Osborne 2015 ; Gurun, Matvos, ​w​​ c​ eˆ ​.​ and Seru( 2016) ; Ru and Schoar 2016( ; and) Johnen 2017 . = (​) ( ) ( ) ( ) 10 AEA PAPERS AND PROCEEDINGS MAY 2018 help workers with ­self-control problems reach If people are naïve, they should not want com- a high level of productivity. These work envi- mitment devices, and firms won’t provide them in ronments deploy forcing mechanisms like the settings we have described so far. deadlines, meetings, reviews, progress reports, This appears to leave us in a conceptual cul de etc., to push workers to be more productive. sac. What could resolve the commitment puzzle? Employers also provide many paternalistic ben- efits like highly subsidized health and retirement V. When Do Naïve Agents “Choose” programs, although ­tax-arbitrage and screening Commitment? provide another rationale for such subsidies. Work environments are one of many private Economists tend to assume that people eval- institutions that are characterized by bundled uate contracts e.g., a job offer by building a forcing mechanisms. As I reviewed above, col- complete probability-weighted­ ( prediction) of their leges feature myriad mechanisms that paternal- future actions under the specific terms of that istically encourage student effort. Even some contract. I’ll refer to this as ­model-based forecast- private financial arrangements have bundled ing. In simple settings, like the employment rela- commitment features: e.g., mortgages with a tionship discussed in this lecture see Sections fixed repayment stream and mandatory principal IIIB and IIIC , it is plausible that (new employ- repayment. Commitment features are also bun- ees might enter) the employment relationship dled into the structure operation of many social with a model-based­ forecast of their actions: ​​eˆ . ​​ organizations: e.g., /religious groups, social The illustrative employment setting that I have clubs, and marriage. described has no uncertainty and no dynamics None of this would be puzzling if these forc- one deterministic state , and the action space ing mechanisms were widely advertised. But, is( ­one-dimensional effort,) e​ ​. The contract is a by and large, these forcing mechanisms aren’t simple univariate function( mapping) effort levels being advertised or marketed. Firms don’t go to wages: ​w e ​. out of their way to tell applicants that the firm Real employment( ) relationships are incom- is going to reduce the scope of procrastination parably more complex. The action space and with deadlines and progress reports. Likewise, the dynamic state space both have numerous colleges don’t boast about all of the work they dimensions,( and) the contingent compensation are going to coax out of their students. We fear function has so many branches that nobody commitment, in marriage and in many other could even write it down, let alone implement a domains, rather than seeking it out. ­probability-weighted model of future states and Herein lies the commitment puzzle. On one actions. Few potential employees would ever try hand, lots of thriving institutions have bundled to gather all of the necessary information to cal- commitment features that appear to be specif- culate this probability-weighted­ model: onboard- ically designed to help agents overcome their ing programs, human resource manuals, travel ­self-control problems. On the other hand, these expense reimbursement policies, workplace institutions generally don’t market these com- training programs, quarterly review processes, mitment features: i.e., the forcing mechanisms performance measurement systems, coworker are shrouded. assignment policies, promotion guidelines, ben- If people are sophisticated or only partially efits manuals, rules for medical leaves, mental naïve , they should appreciate (commitment fea- health programs, etc. Gathering all of this infor- tures )and firms should be eager to market them. mation and then building a best-response­ func- So sophistication implies that the shrouding of tion to all states of nature is a daunting task. forcing mechanisms is a puzzle. Is it then possi- In( complex settings, people) rely instead on ble to resolve the commitment puzzle by assum- other types of information: reputation, ratings, ing that people are completely naïve? Here too and summary measures of experienced utility.33 we encounter a problem. If people are naïve, they should not appreciate the commitment­ ­features, and it will be more profitable for firms to exploit 33 Experienced utility is a measure of subjective ­well-being during some period of time. Experienced util- these agents than to help them overcome their ity is typically measured with questions about affect, ­self-control problems see the preceding two sub- ( mood, emotions, and or life satisfaction. This differs from sections and the exploitative equilibria therein . Spiegler’s 2015 consideration/ of product-use disclosures, ) ( ) VOL. 108 RICHARD T. ELY LECTURE 11

For example, college applicants can easily look contracts that O’Donoghue and Rabin 1999b up a school’s rating compiled by US News and characterize in their analysis. ( ) World Report or one of its competitors . That takes a minute( or two. Reading a student) hand- A. ­Model-Free Equilibrium book takes hours. Reading and evaluating hun- dreds of course syllabi would take weeks. I now study a variant of the effort model from There are many proxy measures that provide above. I change the model slightly to focus on the estimates of what life would be like with a new key issues that I want to highlight. First, I make employer: Glassdoor online ratings, reports of the action choices more complex and the produc- how many hours how( “hard”) existing employ- tion payoff functions arbitrary, emphasizing the ers work each week,( answers) to broad ques- general/ structure of the setting. Second, I omit tions about job quality e.g., “How do you like the choice between formal sector production and working here?” . Firms( report that these repu- home production. Now everyone works in a firm tational factors )play an important role in suc- and also works at home e.g., house cleaning, cessfully recruiting new workers.34 Potential meal preparation, yard work,( home repairs, and employees can imperfectly forecast their own or child care . This change is made to highlight/ future work satisfaction( from) such reports, with- the differences) in efficiency in formal sector pro- out developing a ­model-based forecast of their duction and informal sector home production. future behavior. Likewise, existing employees Third, I incorporate model-free­ ( choice) and call might use their own experienced utility at their the resulting framework a ­model-free equilibrium. employer, rather than a ­model-based theory of I model the labor market as a game among their behavior, to forecast their future utility identical firms in a competitive labor market and flows at their employer. Memories don’t reflect a naïve, ­present-biased worker or workers . ­forward-looking present bias. Instead, memories I now describe the notation.( The agent) reflect an opposite weighting bias: recent events chooses a formal sector action, ​​af​​ ,​ from a com- get more weight than events further back in time pact space: ​​af​​ ​Af​​​ .​ Actions in the formal sector e.g., Malmendier and Nagel 2016 .35 have immediate∈ utility payoff for the formal sec- ( ) f I want to contrast ­model-based forecasting the tor employee of ​ ​e​​ ​ ​af​​ .​ Actions in the formal nearly universal assumption that we make in eco( - − ( ) sector produce real-valued­ output F​ ​af​​ ​ for the nomic analysis , with backward-looking,­ expe- firm. Payments from the formal sector( )employer rientially grounded) model-free­ forecasting. The f have delayed utility benefits​​u ​​ ​w ​af​​ ,​ where terms ­model-based and ­model-free are taken from ( ( )) ​w ​af​​ ​ is a formal sector ­real-valued compensa- the neuroscience literature e.g., Dayan and Niv tion( schedule.) The firm’s( profits are) 2008 . If people make decisions( based on ratings, reputation,) memory, proxy measures, and expe- ​F ​a​f​ w ​a​f​ .​ rienced utility in other words, ­model-free fore- ( ) − ( ) casting , then (equilibrium will sharply diverge The agent also chooses a home produc- ) from the analysis that I’ve described in previous tion action from a compact space, ​​ah​​ ​Ah​​ .​ sections. The equilibrium analysis in Sections Actions at home have immediate utility payoff∈ h IIIB and IIIC assumed that agents generate ​ ​e​​ ​ ​a​h​ .​ Payoffs at home have delayed utility − ( ) h 36 ­model-based forecasts of their future behavior. benefits ​​u​​ ​ ​ah​​ .​ If people use model-free­ forecasting, equilibrium The agent’s( ) total payoff omitting present will instead align with the constrained efficient bias combines her separable( payoff from ( ) formal) sector production( and informal) sector home production: which are specifically meant to help consumers proxy con- ( ) sumption levels. f f 34 8 ​U​ ​a​​ , a​​ ​ ​ ​e​​ ​ ​a​​ ​u​​ ​ w ​a​​ ​ Wade Burgess, “A Bad Reputation Costs a Company ( ) ( f h) ≡ [− ( f ) + ( ( f ))] at Least 10% More Per Hire.” Harvard Business Review, h h March 29, 2016 https://hbr.org/2016/03/a-bad-reputation- ​[ ​e​​ ​ ​a​h​ ​u​​ ​ ​a​h​ ]​.​ costs-company-at-least-10-more-per-hire( . + − ( ) + ( ) 35 See Kahneman et al. 1993 for other) memory dis- tortions, including end effects( experiences) at the end of a 36 Some actions may have immediate payoffs. I focus on stream of experienced utility get( more weight than experi- the leading case in which actions have delayed payoffs: e.g., ences in the rest of the stream . grocery shopping at ​t​ yields a ­well-stocked kitchen at ​t 1​. ) + 12 AEA PAPERS AND PROCEEDINGS MAY 2018

I assume that all of the payoff functions Period 4.—The agent experiences utility flows​​ f h described above are continuous, so all of the u​​ ​ w ​af​​ ​ and u​​ ​​ ​ ​ah​​ .​ These are the agent’s pay- maxima defined below are attained.37 offs,( which( )) result (from) the actions that she chose The game has four periods that match the in period 3. periods in the previous game Section IIIB . ( ) Before analyzing this model, it is import- Period 1.—Each firmi ​​ commits to an enforce- ant to note that the formal sector in this model able wage schedule mapping ​​af​​ to wage ​​wi​​ ​af​​ .​ would generically feature an inefficient equilib- Agents observe the realized remembered utility( ) rium if agents i.e., formal sector workers used flows associated with each /of these wage sched- ­model-based forecasts.( This is a corollary )of the ules by talking to peers, viewing ratings rank- analysis in Section IIIC. Specifically, the equi- ings,( gathering other proxy data, or using/ their librium would not generally maximize social own personal experience memories in the for- welfare equation 8 . Instead, the equilibrium mal sector . In other words,/ agents gather data would feature( bait-and-switch­ ( )) tactics of the type that proxy )for analyzed in Section IIIC. However, I am now assuming that agents do f f ∗ f ∗ Ui​ ​ ​ ​e​​ ​ ​a​ f, i​ ​ ​u​​ ​ ​wi​​ ​a​ f, i​ ​ ,​ not use ­model-based forecasting. Instead, they ≡ − ( ) + ( ( )) use ­model-free forecasting, which exploits peer ∗ where ​​a​ f, i​ ​​ is the equilibrium empirical expe- reports, rankings, memories, and other proxies rienced ­formal-sector behavior/ generated/ by for experienced utility see the assumptions ( wage schedule ​​wi​​ ​af​​ .​ Note that the ­model-free associated with period 1 . Under the assump- ( ) f ) summary measure, U​​ i​ ​ ​,​ places equal weight on​ tion of model-free­ forecasting, socially efficient f ∗ f ∗ ​e​​ ​ ​a​ f, i​ ​ ​ and ​​u​​ ​ ​wi​​ ​a​ f, i​ ​ ​ because the data are actions are implemented in the formal sector −gathered( ) from backward-looking­ ( ( )) memories and i.e., actions are not affected by present bias .39 evaluations.38 (Specifically, in equilibrium in the formal sector,) ∗ workers choose an action ​​a​ f​ ​,​ which satisfies Period 2.—The agent chooses a firm using the model-free data collected in period 1. In other ∗ f f a​ f​ ​ arg max​ ​ ​e​​ ​ ​af​​ ​u​​ ​ F ​af​​ ​.​ words, the agent uses model-free data to choose ∈ ​af​​ [− ( ) + ( ( ))] the most appealing firm: Moreover, equilibrium in the formal sector is not f f arg max​ ​ ​U​ i​ arg max​ ​ Ui​ ​ ​.​ generically characterized by bait-and-switch­ i β = i tactics.( ) This choice is binding for the next period. This property derives from the fact that a firm in a competitive market will not attract a worker Period 3.—Agent chooses actions ​​a​f​ and ​​a​h​ unless the firm delivers the weakly highest subject to a ­present-bias wedge that she had not total payoff to the worker, as( reflected) by the previously anticipated i.e., the agent is naïve payoff measure that the worker uses when she with respect to her present( bias . The cost of picks the firm in period 2 . Using this metric, effort is experienced in period 3.) Formally, the the highest feasible( payoff to) the worker is worker chooses ​​af​​ and ​​a​h​ in the maximization f f problem: max​ ​ ​e​​ ​ ​af​​ ​u​​ ​ F ​af​​ ​.​ ​af​​ [− ( ) + ( ( ))] f f max​ ​ ​ ​e​​ ​ ​af​​ ​u​​ ​ w ​af​​ ​ If we use this equilibrium concept to study the ​af​​, ah​​ {[ ] − ( ) + β ( ( )) original functional forms of Section I, we see that the firms induce their employees to exert ​ ​e​​ h​ ​a​​ ​u​​ h​ ​a​​ ​ ​.​ + [− ( h ) + β ( h )]} effort e​ 1​ in period 3 by not giving them the option to= slack off and choose a lower effort

37 This is an application of the Weierstrass extreme value theorem. Recall too that the action spaces are compact. 39 The assumption ​ 1​ is being used to generate this 38 f ∗ δ = If there were recency bias, then u​​​​ ​ ​wi​​ ​a​ f, i​ ​ would efficiency result. With ​ 1,​ this equilibrium would not be f ∗ ( ( )) δ < actually be weighted more than ​ ​e​​ ​ ​a​ f, i​ ​ ​ and not vice socially efficient because the agent isn’t using​ ​-discounted versa . − ( ) ( flows in Period 2 to choose among firms. δ ) ( ) VOL. 108 RICHARD T. ELY LECTURE 13 level i.e., ​e ​. Firms coax high effort out of the worker in period 3 by using an incentive their employees( = β by) making the low effort option mechanism that punishes( low effort output . unappealing. For example, firms can sufficiently This forcing mechanism is not present/ in home) sanction workers who exert low effort so that the production. Accordingly, the worker chooses firms obtaine ​ 1​ in equilibrium. In this equi- actions in home production that are influenced librium, firms =push their employees to work hard by present bias and accordingly not socially and pay them high wages that go along with this efficient. high effort. This is the opposite of the exploit- To illustrate this point, assume that a worker ative equilibrium in which firms effectively has two production systems, formal and infor- encourage their workers to slack off so they can mal home, that are identical in functional form. pay them very little recall Sections IIIB and Using/ our earlier example again, suppose that IIIC . ( these symmetric effort costs are respectively​ ) _1 2 _1 2 To summarize, if prospective employees use ​ ​ ​e​ f​ ​ and ​ ​ ​ ​e​ h​ ​ .​ Suppose also that firms’ ­model-free forecasts to choose among employ- −2 −2 production functions are linear in ​​e​f​ ,​ so that ers, equilibrium contracts will be characterized ​F ​e​​ ​e​​​ .​ Finally, suppose that u​​​​ w w​ ( f ) = f f ( ) = by efficient forcing mechanisms i.e., commit- and ​​u​​ ​e​​ ​e​​ .​ In equilibrium, ​​e​​ 1​ and ment , even if agents are naïve. In( competitive h ( h ) = h f = ) 41 equilibrium, workers will choose a​​ ​ ∗​ ​ arg ​max​ ​ ​​e​h​ .​ The socially efficient level of produc- f ​af​​ = β ​e​​ f​ ​a​​ ​u​​ f​ F ​a​​ ​​ and get paid ​F∈ ​a​ ∗​ ​ .​ tion is only achieved in the formal sector. Home [− ( f ) + ( ( f ))] ( f ) production is characterized by an inefficiently B. Resolution of the Commitment Puzzle low level of production unless some other forc- ing mechanism is present( at home, a question that I will return to in Section VII . This framework provides a potential resolu- ) tion to the commitment puzzle. With ­model-free forecasting, firms provide forcing mechanisms VI. Private Paternalism that look like commitment devices. Because these forcing mechanisms are being pro- This lecture has discussed two channels by vided to naïve agents, firms do not highlight which private, ­for-profit firms choose policies the fact that their incentive contracts will turn that advance workers’ interests by restrict- out to be strictly binding. The commitment is ing workers’ choices. The first channel of pri- implicit. Commitment is supported in equilib- vate paternalism is frequently discussed in the rium because it produces high levels of experi- intertemporal choice literature: if workers are enced remembered utility for workers, thereby ­present-biased and sophisticated, and they use enabling/ the firms to recruit more workers under ­model-based reasoning, they will explicitly ask that successful compensation work schedule. firms to limit their choices e.g., Laibson 1997, / 2015; Lin 2017 . While this( channel exists in C. Inefficiency in Home Production principle and has) been elicited in lab and field experiments see Section II , it does not seem In the current section, I have focused on equi- to be a strong( force in most) real markets. Very librium outcomes in the formal sector. As I have few firms recruit workers or customers by boast- argued, those outcomes will be socially efficient. ing about the firm’s ­choice-limiting policies But the forces that create this efficiency do not products services. / apply to home production. In the formal sector, There /is also a second channel of private the agent chooses an employer in period 2 and paternalism that will arise for both sophisticated is stuck with that employer through period 4.40 and naïve agents. This channel depends on two Because the worker can’t instantaneously switch conditions: i agents workers use ­model-free employers in period 3, the employer is able to data to choose( ) among/ prospective firms, and force a high efficient level of costly effort from ii those agent worker choices have some ( ) degree( ) of temporary/ stickiness so the worker who chooses a firm in period ( 2 stays at that 40 At the end of the fourth period, the entire game can repeat and the worker can choose a new firm in the next iter- ation of period 2. 41 Equilibrium formal sector wage is ​w 1.​ = 14 AEA PAPERS AND PROCEEDINGS MAY 2018 firm in period 3 . When these two conditions e.g., the cost of switching jobs. Hence, private are met, firms )will choose ­freedom-restricting paternalism is not libertarian, because agents policies that engender social efficiency. These would opt out at period 3 if the cost were small forcing mechanisms will be shrouded if there enough. It is this( friction )that keeps the equilib- are enough naïve workers in the population. rium from unraveling and serves to bind them Naïve workers will not fully understand why unexpectedly, in this instance to the mast. these restrictive employer( policies) are desirable. ( ) Using model-free­ learning, naïve agents will VII. The Limits and Virtues of Private be able to identify the firms that are associated Paternalism with good outcomes for their employees, but the naïve agents won’t understand the causal mech- Private paternalism has numerous limitations anisms. Naïve agents view freedom-restricting­ that prevent it from serving as a general counter- policies as intrinsically adverse to the extent weight to present bias. I review those limitations that they use at least some model-based­ ( rea- first. Then I conclude this section by describing soning . Accordingly, firms will have no reason the benefits of private paternalism and speculate to trumpet) ­freedom-restricting policies, even about constructive ways of expanding its practi- though firms have an incentive to adopt such cal scope. policies. These two channels demonstrate that private A. Limits of Private Paternalism paternalism i.e., paternalism that is imple- mented by private( institutions without gov- The model that I used to illustrate private ernment intervention is a coherent concept. paternalism makes many special simplifying The second channel )demonstrates that private assumptions. Some of these assumptions are paternalism may occur even if agents are naïve. necessary for the efficiency result that the model Moreover, the second channel explains why pri- generated. vate paternalism is often shrouded. Under the First, I assumed that the agents’ payoff func- second channel, firms will want to highlight tions in the formal sector are separable from employee satisfaction and gloss over the forcing those in the informal sector home production . mechanisms that make those employees produc- In general, this won’t be true.( For example,) tive and ­well compensated. there are only a fixed number of hours in the day, so more hours spent working in the formal A. Private Paternalism versus Libertarian sector will crowd out hours spent in the infor- Paternalism mal sector. Effort may work the same way: an intense day at formal sector work may crowd out From a linguistic perspective, private pater- effort at home even if hours at the two activ- nalism looks like libertarian paternalism ities don’t change( . Dropping separability will Thaler and Sunstein 2003, 2008 . However, break the clean result) that workplace actions are unlike( libertarian paternalism, private) pater- socially efficient. Naturally, one can still solve nalism as defined in this lecture incorporates for model-free­ equilibria in the absence of sep- forcing (mechanisms that are ­ex )post binding, arability. A characterization of the nonseparable­ although they are adopted on a voluntary ex­ ante case is an open problem. basis. Consider the model-free­ equilibrium Second, unobserved heterogeneity in present Section VA and the specific illustrative exam- bias will break the social efficiency result, even ple( that I have) been carrying through this lec- in the formal sector, because screening is now ture. Agents in period 3 would like to opt out of necessary e.g., O’Donoghue and Rabin 1999b, the contract that they accepted in period 2. They 2006; Galperti( 2015; Lin 2017 . Efficient joined the firm in period 2 to work at effort level​ screening will require repeated interactions) or e 1​ for payment w​ 1,​ but they would prefer historical data if agents are naïve about their( to= costlessly opt out =in period 3 and work at present bias. ) another firm at effort( level ​e )​ for payment ​ Third, the efficiency results will break down w .​ What prevents them from= β doing this? when benefits or costs occur at horizons that Some= β friction or cost that is large enough to fall beyond the horizon of experienced util- temporarily bind them to their current employer: ity measures. For example, coal miners may VOL. 108 RICHARD T. ELY LECTURE 15 experience health problems that occur at long data. In addition, data on experienced utility are delays. Even mundane tasks, like deliver- a public good, so they will be undersupplied ing packages or working on an assembly line, even if agents did understand their value. For all may generate ­long-horizon, adverse conse- of these reasons, experienced utility is not pro- quences that are missed in short-horizon­ mea- vided, though it would be inexpensive to collect sures of experienced utility. Likewise, some in principle: e.g., gyms generally do not disclose ­employer-based benefits may havelong-horizon ­ the average frequency of attendance of their consequences. Health investments, like a smok- members, though they could report these data ing cessation program, generate ­long-run ben- because many gyms now use swipe cards for efits. Retirement benefits accrue for decades entry e.g., Dellavigna and Malmendier 2006 . before they are paid out. Experienced utility In the (model that I have discussed, such infor)- needs to capture all of these long-run­ channels mation would be useful for naïve, ­present-biased to generate social efficiency in the ­model-free agents using ­model-free forecasting. However, equilibrium. those agents don’t realize the value of this infor- Fourth, even when the benefits and costs mation and gyms do not have an incentive to occur in the ­short run, it is unlikely that expe- provide it. rienced utility proxies will fully reflect all of the actual utility flows. The model assumed that B. Home Production agents have access to information sources that jointly provide an unbiased, comprehensive, and Private paternalism is particularly ineffec- noiseless measure of experienced utility. But all tive in the domain of home production. There of these properties are problematic. Glassdoor is no “firm” to provide enforcement. However, reviews are prone to selection bias and some some domestic partnerships e.g., marriages employers have been accused of pressuring pay- operate in a way that generates( an efficient forc)- ing employees to write positive reviews. Current/ ing mechanism. For example, my spouse does employees may distort their reports to potential an excellent job of forcing me to turn off my employees who are being recruited. Employers computer phone iPad at certain times of the and employees may collude in providing infor- day. Nevertheless,/ / marriage is a crude tool for mation to prospective employees. Job panels achieving the efficiency gains that I modeled in e.g., at career fairs usually don’t include dis- the formal sector. Partner selection isn’t efficient gruntled( employees) and supervisors typically for many different reasons. First, it is difficult to monitor the reports of ­entry-level employees gather ­forecast-relevant, ­experienced-utility data at these events. Marketing videos for corporate on one’s partner before settling down with them: recruitment usually feature the most excited would you ask your fiancée’s ­ex-spouse for this positive current employees, or at least those who/ information and, if so, how would you interpret are willing to express such opinions. Finally, the report? By contrast, a firm has many current many subtle aspects of a job may be hard to cap- and ex-employees­ who can serve as credible ture in rankings or ratings. Experienced utility informants. Even good partners aren’t always may be particularly hard to gather in novel set- present to act as a commitment mechanism: I go tings, where very few people have experience to McDonald’s whenever my spouse isn’t with with the good service being sold or job being me. Good partners aren’t scalable, whereas a offered. In these/ settings, agents may be more firm with a good commitment forcing technol- prone to rely on model-based­ forecasts because ogy can scale its operations. In/ addition, choices they effectively have no ­experienced-based( data of romantic partners are sometimes heavily on which to rely . influenced by considerations that have little Fifth, the costly) collection of data on expe- bearing on long-run­ happiness. Finally, a large rienced utility( will) also be inefficiently low. and growing fraction of the world’s adults don’t Because naïve, present-biased­ agents overes- have partners at all. timate the accuracy of their own ­model-based forecasts and ­underestimate the relevance of C. Public Paternalism other people’s behavior as a proxy for their own behavior Fedyk 2017 , they will not fully real- When private paternalism is likely to fail, ize the benefits( of gathering) experienced utility there may be a case for public paternalism i.e., ( 16 AEA PAPERS AND PROCEEDINGS MAY 2018 paternalism initiated by some part of the gov- D. Virtues of Private Paternalism ernment . As discussed above, private paternal- ism is unlikely) to efficiently address institutional I now discuss six benefits of private arrangements that have ­long-run consequences paternalism. e.g., retirement, health, and insurance . First, private paternalism can enable agents to Experienced( utility of a mid-life­ worker may fail) overcome present bias, even when they have naïve to capture payoffs that occur later in life. Defined beliefs. Private paternalism is optional at low fre- benefit or defined contribution pension plans quencies e.g., the choice of an employer , but generate payoffs that aren’t viscerally or hedon- binding at( high frequencies e.g., effort choices) ically experienced until retirement.42 Likewise, over the course of a work day( , which is an ideal a firm’s health investments in its workforce gen- correction for agents with present) bias. erate benefits that may take a lifetime to be fully Second, private paternalism can enable agents realized. Experienced utility, which is captured to overcome a host of biases other than present by backward-looking­ memories, is unlikely bias. The examples discussed in this paper and to reflect these future payoffs. Hence, the in O’Donoghue and Rabin 1999b, 2006 focus( ­model-free equilibrium that I discussed Section on present bias, but ­model-free learning) is a VA won’t create efficient ­long-run investments.( ­general-purpose technique for avoiding some Accordingly,) long-run­ investments present a types of misforecasting and the costs that usu- natural case for public paternalism. Indeed, a ally accompany misforecasting( . For example, substantial fraction of public paternalism is con- irrationally optimistic agents may) be less eas- centrated in such domains, including tax and ily exploited if they use ­model-free forecasts regulatory incentives for retirement savings,43 instead of using the overoptimistic structural Social Security forced retirement savings , and theory that distorts their forecasts. “The average Medicare forced( savings during working) life person loses money in a casino, so I shouldn’t that funds health( insurance in retirement . expect to do better than that despite my struc- Likewise, private paternalism is likely) to be tural theory that there exists (an exploitable hot less relevant outside of formal sector employ- hand effect .”44 ment relationships. Indeed, in the ­model-free Third, private) paternalism is not imposed by equilibrium Section VA , there were no forcing the government. Many scholars are skeptical mechanisms( in the home) production sector. By of strict, government-controlled­ paternalism.45 contrast, in a firm, monitoring and forcing mech- Private paternalism avoids further empowering a anisms e.g., deadlines and progress reports are Leviathan, which may not have our best interests natural (features of the environment. In home) at heart or may not be nimble smart enough to production, monitors are often absent, and, even be optimally paternalistic. / when some potential monitor is present, it is not Fourth, private paternalism allows for innova- clear that they have the necessary authority or tion and improvement, because it is generated appropriately aligned incentives. Hence, home and continuously reinvented by agents in the production is less likely to be characterized by private sector and subject to competitive forces. efficient private paternalism thanformal-sector ­ Fifth, private paternalism will appeal to both production. Accordingly, many types of public naïve agents if they use ­model-free forecasting paternalism are targeted at behavior outside of as well as sophisticated( agents whether they use) firms: e.g., drug criminalization, cigarette and ­model-free or ­model-based forecasting( . Hence, soda taxes, helmet and seatbelt laws and, in most private paternalism will be successful for) a wide countries, mandatory ­state-funded unemploy- range of cognitive types. ment insurance. ( )

44 This argument has its limits. Inferences about expe- rienced utility must sometimes be made contingently to be 42 This raises the thorny question of whether a place- useful. One wouldn’t want to reason that the average kay- holder for a future flow of utils, say a worker’s rising account aker on Class V rapids doesn’t get injured so I can kayak balance in a 401 k , serves as an accurate forecasting proxy. down Class V rapids despite my complete lack of experi- 43 For example,( ) the tax treatment of IRAs and 401 k s, ence as a kayaker . ( as well as regulatory incentives for firms to offer matching( ) 45 For example,) see Thaler and Sunstein 2003, 2008 and contributions in 401 k plans. Glaeser 2006 . ( ) ( ) ( ) VOL. 108 RICHARD T. ELY LECTURE 17

Sixth, private paternalism will only suc- developing and developed economies lies in the ceed in gaining greater market share if it raises relative role of the formal sector. In developing social welfare. Unlike government paternalism, countries, the formal sector is relatively small, which is imposed by fiat, private paternalism which implies that the scope for private pater- is ­choice-based and, if agents use accurate nalism is small. In this sense, the lack of private ­model-free forecasts, prone to maximize( flows) paternalism in the informal sector can serve as a of experienced utility. poverty trap. To gain intuition for this effect, think of the VIII. Conclusion: The Scope of Private life of a typical university employee. Most uni- Paternalism versities provide health clinics, health insurance, retirement benefits, life insurance, disability In this lecture, I have characterized the insurance, medical leave, a structured work scope of private paternalism in an economy week, performance benchmarks, frequent feed- with ­present-biased agents. In this setting, pri- back, deadlines, and many other mechanisms vate paternalism arises through two channels: that keep us productive. A ­self-employed worker i agents who seek out commitment because has little or none of these supporting structures. they( ) hold sophisticated beliefs about their pres- Accordingly, I believe that a substantial frac- ent bias and use ­model-based forecasting, and tion of my productivity owes to the paternalistic ii agents naïve or sophisticated who use environments in which I am embedded, rather ­model-free( ) forecasting( to choose )firms that than my intrinsic productivity. I believe that if generate high experienced utility flows by pre- I moved to the self-employment­ sector, my pro- venting workers from making present-biased­ ductivity and life satisfaction would plummet, choices. When the second channel is active, in part because my present bias would have free and naïve consumers are a significant part of rein to distort my behavior. the population, then private paternalism will be Future research should empirically evalu- shrouded. Firms will deploy private paternalism, ate these conceptual conjectures. It should also but they won’t highlight it. This combination of identify the feasibility and consequences of private paternalism and shrouding resolves the expanding the scope of private paternalism. The commitment puzzle. Forcing mechanisms are role of private paternalism is enlarged i when commonplace in the labor market but neverthe- we find ways to shift workers from the( informal) less shrouded because they are being deployed sector to the formal sector or make the informal to attract naïve workers who like the resulting sector more structured so (it can support private flows of high experienced utility, and don’t rec- paternalism ; ii when we make agents more ognize the paternalistic roots of that success. sophisticated,) (so) they will explicitly seek out I have argued that private paternalism arising commitment devices; and iii when we increase from model-free­ learning will be more socially the quality of and access (to data) on ­model-free efficient in the following settings: i when pay- experienced utility e.g., requiring that gyms offs are clearly tied to actions, so( measures) of report the exercise frequency( of their members, experienced utility are more accurate; ii when or requiring that firms disclose standardized sat- payoffs aren’t too delayed relative( )to the isfaction surveys of their employees . The pro- actions that generate those payoffs; iii when vision of data on experienced utility )is a public employment relationships are ­long-lived,( ) so good that could be subsidized or mandated. that ­long-run costs and benefits are integrated into available measures of experienced utility; Appendix: Solving for the Exploitative and iv when productive activity occurs in the Equilibrium formal( )sector, where monitoring and forcing mechanisms are easy to deploy, instead of the Here I derive equilibrium for the case of a informal home production sector, where mon- monopsony employer Section IIIB . Following itors are not( always present) and not always able the discussion in the text,( the firm’s )optimization or willing to implement socially efficient forcing problem is mechanisms. This last point leads me to speculate that some 9 max​ ​ ​e​​ ∗​ w ​e​​ ∗​ ​ ( ) ​e​​ ∗​, w ​e​​ ∗​, ​eˆ ​, w eˆ ​ − ( ) of the difference between the quality of life in ( ) (​) 18 AEA PAPERS AND PROCEEDINGS MAY 2018 subject to the participation constraint P and But the actual wage is incentive compatibility constraint IC : ( ) ( ) _ _ ​w ​e​​ ∗​ _ 1 ​ ​ ​​ 2​ ​ ​ _1 ​ ​​ e ​2​ z ​ _1 ​ ​​ e ​2​ .​ P ​ ​ _1 ​ ​​eˆ ​​ ​2​ w eˆ ​ z​ , ( ) = 2 β + (2 + ) − 2 ( ) −2 + (​) ≥ β β

1 2 1 2 IC ​ ​ _ ​ ​​ ​e​​ ∗​ ​ ​ w ​e​​ ∗​ ​ _ ​ ​​ eˆ ​​ ​ ​ w eˆ ​.​ REFERENCES ( ) −2 ( ) + β ( ) ≥ −2 + β (​) Note that the participation and incentive com- Acland, Dan, and Matthew R. Levy. 2015. patibility constraints will both bind in equi- “Naïveté, Projection Bias, and Habit Forma- librium. If the participation constraint didn’t tion in Gym Attendance.” Management Sci- bind, the firm could increase profit by lowering ence 61 1 : 146–60. ​w eˆ ​​ and ​w ​e​​ ∗​ ,​ holding all else equal. If the Ainslie, George.( ) 1975. “Specious Reward: A incentive(​) compatibility( ) constraint didn’t bind, Behavioral Theory of Impulsiveness and the firm could increase profit by lowering Impulse Control.” Psychological Bulletin 82 ​w ​e​​ ∗​,​ holding all else equal. Henceforth, I will 4 : 463–96. treat( )these as binding constraints. Rearrange the Ainslie,( ) George. 1992. Picoeconomics. Cam- participation constraint to find bridge: Cambridge University Press. Akerlof, George A. 1991. “Procrastination and ​w eˆ ​ _ 1 ​ ​​eˆ ​​ ​2​ z.​ Obedience.” American Economic Review 81 2 (​) = + 2 : 1–19. Now take this wage and substitute it into the Alsan,( ) Marcella, John Beshears, Wendy S. Arm- incentive compatibility constraint to find strong, James J. Choi, Brigitte C. Madrian, Minh Ly T. Nguyen, Carlos Del Rio, David Laibson, and Vincent C. Marconi. 2017. “A ​w ​e​​ ∗​ _1 ​ ​​ ​e​​ ∗​ ​ 2​ ​ _1 ​ ​​eˆ ​​ ​2​ z ​ _1 ​ ​​eˆ ​​ ​2​ .​ β ( ) − 2 ( ) = β[​ 2 + ] − 2 Commitment Contract to Achieve Virologic Suppression in Poorly Adherent Patients with Rearranging this expression, I express the final HIV AIDS.” AIDS 31 12 : 1765–69. wage, ​w ​e​​ ∗​​ , as a function of ​​e​​ ∗​ and ​​eˆ :​​ Ariely, /Dan, and Klaus Wertenbroch.( ) 2002. “Pro- ( ) crastination, Deadlines, and Performance: Self-Control by Precommitment.” Psycholog- ​w ​e​​ ∗​ _1 ​ ​​ ​e​​ ∗​ ​ 2​ ​ ​ _1 ​ ​​eˆ ​​ ​2​ z ​ _1 ​ ​​eˆ ​​ ​2​ .​ ( ) = 2 ( ) + [2 + ] − 2 ical Science 13 3 : 219–24. β β Armstrong, Mark, (and) John Vickers. 2012. “Con- Accordingly, I can rewrite the firm’s objective as sumer Protection and Contingent Charges.” Journal of Economic Literature 50 2 : 477– 93. ( ) max​ ​ e​​ ∗​ ​ ​ _1 ​ ​​ ​e​​ ∗​ ​ 2​ ​ ​ _1 ​ ​​eˆ ​​ ​2​ z ​ _1 ​ ​​eˆ ​​ ​2​ ​.​ ​e​​ ∗​, eˆ ​ − 2 ( ) + [2 + ] − 2 Ashraf, Nava, Colin F. Camerer, and George Loe- { β β } wenstein. 2005. “Adam Smith, Behavioral This maximization generates two FOCs. The Economist.” Journal of Economic Perspectives FOC for ​​e​​ ∗​ is 19 3 : 131–45. Ashraf,( Nava,) Dean Karlan, and Wesley Yin. 2006. ​1 _ 1 ​ ​e​​ ∗​ 0,​ “Tying Odysseus to the Mast: Evidence from − = β a Commitment Savings Product in the Philip- which implies that ​​e​​ ∗​ .​ The FOC for ​​eˆ ​ is pines.” Quarterly Journal of Economics 121 = β 2 : 635–72. ​ ​eˆ ​ _1 ​​eˆ ​ 0.​ Augenblick,( ) Ned, Muriel Niederle, and Charles − + = β Sprenger. 2015. “Working over Time: Dynamic This first derivative with respect toe ​​ˆ ​ is every- Inconsistency in Real Effort Tasks.” Quarterly where strictly positive because_ ​ 1.​ So ​​eˆ ​ is set Journal of Economics 130 3 : 1067–115. equal to its upper bound ​​ e ​​. Soβ the< anticipated Augenblick, Ned, and Matthew( Rabin.) Forthcom - wage is ing. “An Experiment on Time Preference and _ Misprediction in Unpleasant Tasks.” Review of ​w eˆ ​ _ 1 ​ ​​ e ​2​ z.​ Economic Studies. (​) = 2 + VOL. 108 RICHARD T. ELY LECTURE 19

Ausubel, Lawrence M. 1991. “The Failure of Dayan, Peter, and Yael Niv. 2008. “Reinforcement Competition in the Credit Card Market.” Amer- Learning: The Good, the Bad and the Ugly.” ican Economic Review 81 1 : 50–81. Current Opinion in Neurobiology 18 2 : 185– Beshears, John, James J. Choi,( )Christopher Har- 96. ( ) ris, David Laibson, Brigitte C. Madrian, and DellaVigna, Stefano, and Ulrike Malmendier. Jung Sakong. 2017. “Which Early Withdrawal 2004. “Contract Design and Self-Control: The- Penalty—10%, 20%, or ∞—Attracts the Most ory and Evidence.” Quarterly Journal of Eco- Deposits to a Commitment Savings Account?” nomics 119 2 : 353–402. NBER Working Paper 21474. DellaVigna, Stefano,( ) and Ulrike Malmendier. Bisin, Alberto, and Kyle Hyndman. 2014. “Pres- 2006. “Paying Not to Go to the Gym.” Ameri- ent-Bias, Procrastination and Deadlines in can Economic Review 96 3 : 694–719. a Field Experiment.” NBER Working Paper Eliaz, Kfir, and Ran Spiegler.( 2006.) “Contracting 19874. with Diversely Naïve Agents.” Review of Eco- Brown, George W. 1951. “Iterative Solutions of nomic Studies 73 3 : 689–714. Games by Fictitious Play.” In Activity Anal- Eliaz, Kfir, and Ran( )Spiegler.2008. “Consumer ysis of Production and Allocation, edited by Optimism and Price Discrimination.” Theoret- Tjalling C. Koopmans, 374–76. New York: ical Economics 3 4 : 459–97. Wiley. Ely, Jeffrey C., and (Juuso) Välimäki. 2003. “Bad Bryan, Gharad, Dean Karlan, and Scott Nelson. Reputation.” Quarterly Journal of Economics 2010. “Commitment Devices.” Annual Review 118 3 : 785–814. of Economics 2 1 : 671–98. Fedyk, (Anastassia.) 2017. “Asymmetric Naïveté: Bucks, Brian, and (Karen) Pence. 2008. “Do Bor- Beliefs about Self-Control.” Unpublished. rowers Know Their Mortgage Terms?” Journal Fudenberg, Drew, and David K. Levine. 1998. of Urban Economics 64 2 : 218–33. The Theory of Learning in Games. Cambridge, Cadena, Ximena, Antoinette( ) Schoar, Alexan- MA: MIT Press. dra Cristea, and Héber M. Delgado-Medrano. Gabaix, Xavier, and David Laibson. 2006. 2011. “Fighting Procrastination in the Work- “Shrouded Attributes, Consumer Myopia, and place: An Experiment.” NBER Working Paper Information Suppression in Competitive Mar- 16944. kets.” Quarterly Journal of Economics 121 2 : Carroll, Gabriel D., James J. Choi, David Laib- 505–40. ( ) son, Brigitte C. Madrian, and Andrew Metrick. Galperti, Simone. 2015. “Commitment, Flexibil- 2009. “Optimal Defaults and Active Deci- ity, and Optimal Screening of Time Inconsis- sions.” Quarterly Journal of Economics 124 tency.” Econometrica 83 4 : 1425–65. 4 : 1639–74. Gerardi, Kristopher, Lorenz( Goette,) and Stephan Cho,( )SungJin, and John Rust. 2017. “Precommit- Meier. 2013. “Numerical Ability Predicts Mort- ments for Financial Self-Control? Micro Evi- gage Default.” Proceedings of the National dence from the 2003 Korean Credit Crisis.” Academy of Sciences 110 28 : 11267–71. Journal of Political Economy 125 5 : 1413– Giné, Xavier, Dean Karlan, (and) Jonathan Zin- 64. ( ) man. 2010. “Put Your Money Where Your Butt Choi, James J., David Laibson, Brigitte C. Is: A Commitment Contract for Smoking Ces- Madrian, and Andrew Metrick. 2003. “Opti- sation.” American Economic Journal: Applied mal Defaults.” American Economic Review 93 Economics 2 4 : 213–35. 2 : 180–85. Glaeser, Edward( L.) 2006. “Paternalism and Psy- Chung,( ) Shin-Ho, and Richard J. Herrnstein. 1967. chology.” University of Chicago Law Review “Choice and Delay of Reinforcement.” Jour- 73 1 : 133–56. nal of the Experimental Analysis of Behavior Goda,( Gopi) Shah, Matthew R. Levy, Colleen Fla- 10 1 : 67–74. herty Manchester, Aaron Sojourner, and Joshua Clark,( Gregory.) 1994. “Factory Discipline.” Jour- Tasoff. 2015. “The Role of Time Preferences nal of Economic History 54 1 : 128–63. and Exponential-Growth Bias in Retirement Cohen, Jonathan D., Keith Marzilli( ) Ericson, David Savings.” NBER Working Paper 21482. Laibson, and John Myles White. 2017. “Mea- Gottlieb, Daniel. 2008. “Competition over suring Time Preferences.” NBER Working Time-Inconsistent Consumers.” Journal of Paper 22455. Public Economic Theory 10 4 : 673–84. ( ) 20 AEA PAPERS AND PROCEEDINGS MAY 2018

Grubb, Michael D. 2009. “Selling to Overconfi- Kuchler, Theresa, and Michaela Pagel. 2017. dent Consumers.” American Economic Review “Sticking to Your Plan: The Role of Present 99 5 : 1770–807. Bias for Credit Card Paydown.” Unpublished. Grubb,( )Michael D. 2015. “Consumer Inattention Laibson, David. 1997. “Golden Eggs and Hyper- and Bill-Shock Regulation.” Review of Eco- bolic Discounting.” Quarterly Journal of Eco- nomic Studies 82 1 : 219–57. nomics 112 2 : 443–77. Grubb, Michael D., and( ) Matthew Osborne. 2015. Laibson, David.( )2015. “Why Don’t Present-Bi- “Cellular Service Demand: Biased Beliefs, ased Agents Make Commitments?” American Learning, and Bill Shock.” American Eco- Economic Review 105 5 : 267–72. nomic Review 105 1 : 234–71. Laibson, David, and Leeat( Yariv.) 2007. “Safety in Gurun, Umit G., Gregor( ) Matvos, and Amit Seru. Markets: An Impossibility Theorem for Dutch 2016. “Advertising Expensive Mortgages.” Books.” Unpublished. Journal of Finance 71 5 : 2371–416. Levin, Jonathan. 2003. “Relational Incentive Halac, Marina, and Andrea( Prat.) 2016. “Manage- Contracts.” American Economic Review 93 rial Attention and Worker Performance.” Amer- 3 : 835–57. ican Economic Review 106 10 : 3104–32. Levy,( ) David T., Ron Borland, Eric N. ­Lindblom, Hall, Robert E. 1997. “The Inkjet( ) Aftermarket: Maciej L. Goniewicz, Rafael Meza, Theo- An Economic Analysis.” Unpublished. dore R. Holford, Zhe Yuan, Yuying Luo, Rich- Heidhues, Paul, and Botond Koszegi. 2010. ard J. O’Connor, Raymond Niaura, and “Exploiting Naïveté about Self-Control in the David B. Abrams. 2017. “Potential Deaths Credit Market.” American Economic Review Averted in USA by Replacing Cigarettes 100 5 : 2279–303. with ­E-Cigarettes.” Tobacco Control 27: Heidhues,( ) Paul, and Botond Koszegi. 2015. “On 18–25. the Welfare Costs of Naïveté in the US Cred- Lin, Hong. 2017. “Firm as a Self-Commitment it-Card Market.” Review of Industrial Organi- Device.” Unpublished. zation 47 3 : 341–54. Loewenstein, George, and Drazen Prelec. 1992. Heidhues, Paul,( ) and Botond Koszegi. 2017. “Anomalies in Intertemporal Choice: Evidence “Naïveté-Based Discrimination.” Quarterly and an Interpretation.” Quarterly Journal of Journal of Economics 132 2 : 1019–54. Economics 107 2 : 573–97. Holmström, Bengt. 1999. “Managerial( ) Incentive Malmendier, Ulrike,( ) and Stefan Nagel. 2016. Problems: A Dynamic Perspective.” Review of “Learning from Inflation Experiences.” Quar- Economic Studies 66 1 : 169–82. terly Journal of Economics 131 1 : 53–87. Houser, Daniel, Daniel Schunk,( ) Joachim K. Win- O’Donoghue, Ted, and Matthew Rabin.( ) 1999a. ter, and Erte Xiao. 2010. “Temptation and “Doing It Now or Later.” American Economic Commitment in the Laboratory.” Unpublished. Review 89 1 : 103–24. Johnen, Johannes. 2017. “Dynamic Competition O’Donoghue, (Ted,) and Matthew Rabin. 1999b. in Deceptive Markets.” Unpublished. “Incentives for Procrastinators.” Quarterly Kahneman, Daniel, Barbara L. Fredrickson, Journal of Economics 114 3 : 769–816. Charles A. Schreiber, and Donald A. Redelmeier. O’Donoghue, Ted, and Matthew( ) Rabin. 2006. 1993. “When More Pain Is Preferred to Less: “Incentives and Self-Control.” In Advances Adding a Better End.” Psychological Science in Economics and Econometrics: Theory and 4 6 : 401–5. Applications, Ninth World Congress, Vol. Kanemoto,( ) Yoshitsugu, and W. Bentley MacLeod. 2, edited by Richard Blundell, Whitney K. 1992. “Firm Reputation and Self-Enforcing Newey, and Torsten Persson, 215–45. Cam- Labor Contracts.” Journal of the Japanese and bridge: Cambridge University Press. International Economies 6 2 : 144–62. Phelps, E. S., and R. A. Pollak. 1968. “On Sec- Kaur, Supreet, Michael Kremer,( )and Sendhil Mul- ond-Best National Saving and Game-Equilib- lainathan. 2010. “Self-Control and the Devel- rium Growth.” Review of Economic Studies 35 opment of Work Arrangements.” American 2 : 185–99. Economic Review 100 2 : 624–28. Royer,( ) Heather, Mark Stehr, and Justin Sydnor. Kaur, Supreet, Michael Kremer,( ) and Sendhil Mul- 2015. “Incentives, Commitments, and Habit lainathan. 2015. “Self-Control at Work.” Jour- Formation in Exercise: Evidence from a Field nal of Political Economy 123 6 : 1227–77. Experiment with Workers at a Fortune-500 ( ) VOL. 108 RICHARD T. ELY LECTURE 21

Company.” American Economic Journal: Implicit, and Avoidable Costs.” American Eco- Applied Economics 7 3 : 51–84. nomic Review 99 2 : 424–29. Ru, Hong, and Antoinette( ) Schoar. 2016. “Do Strotz, Robert H.( ) 1955. “Myopia and Credit Card Companies Screen for Behavioral ­Inconsistency in Dynamic Utility Maximi- Biases?” NBER Working Paper 22360. zation.” Review of Economic Studies 23 3 : Schilbach, Frank. 2017. “Alcohol and Self- 165–80. ( ) Control: A Field Experiment in India.” Thaler, Richard H., and Cass R. Sunstein. 2003. ­Unpublished. “Libertarian Paternalism.” American Eco- Shapiro, Fred R., ed. 2006. The Yale Book of Quo- nomic Review 93 2 : 175–79. tations. New Haven, CT: Yale University Press. Thaler, Richard H., (and) Cass R. Sunstein. 2008. Shui, Haiyan, and Lawrence M. Ausubel. 2005. Nudge: Improving Decisions about Health, “Time Inconsistency in the Credit Card Mar- Wealth, and Happiness. New Haven, CT: Yale ket.” Unpublished. University Press. Spiegler, Ran. 2011. Bounded Rationality and Warren, Patrick L., and Daniel H. Wood. 2014. Industrial Organization. Oxford: Oxford Uni- “The Political Economy of Regulation in versity Press. ­Markets with Naïve Consumers.” Journal of Spiegler, Ran. 2015. “On the Equilibrium Effects the European Economic Association 12 6 : of Nudging.” Journal of Legal Studies 44 2 : 1617–42. ( ) 389–416. ( ) Wertenbroch, Klaus. 1998. “Consumption Stango, Victor, and Jonathan Zinman. 2009. Self-Control by Rationing Purchase Quantities “What Do Consumers Really Pay on Their of Virtue and Vice.” Marketing Science 17 4 : Checking and Credit Card Accounts? Explicit, 317–37. ( )