American Economic Association

Why ? Author(s): John Conlisk Source: Journal of Economic Literature, Vol. 34, No. 2 (Jun., 1996), pp. 669-700 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/2729218 Accessed: 20/10/2008 21:54

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=aea.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Literature.

http://www.jstor.org Journal of Economic Literature Vol. XXXIV (June 1996), pp. 669-700

Why Bounded Rationality?

JOHN CONLISK University of California, San Diego

Special thanksfor editorial suggestions are due to Vincent Crawford, Garey Rainey, Michael Rothschild, and three most helpful referees. Very special thanks for many years of helpfui insights are due to Richard Day and Luigi Ermini.

Hamlet: "What a piece of work is a man! how noble in reason! how infinite in faculties!" Hamlet, II.2.319. Puck: "Lord, whatfools these mortals be!" Midsummer Night's Dream, III.3.116.

NEARLY EVERYONE would see the ring theme. Most references are to the truth as between Hamlet and Puck. last 15 years, though many earlier works Including Hamlet and Puck. Hamlet is are also cited. A longer version of the feigning madness, and Puck is just being, survey, including many more references, well, puckish. Model-writing economists, is available from the author on request. however, tend not to the middle but to the "infinite in faculties" extreme. Al- I. Spoiling a Good Story: Evidence of though the postulate of unbounded ra- Bounds on Rationality tionality has dominated economic model- ing for several decades, the dominance is Should the facts be allowed to spoil a good story? relaxing. Is this encouraging? Why Michael Lovell (1986, p. 120) bounded rationality? Lovell asked this question about un- In this survey, four reasons are given bounded rationality in forecasting (about for incorporating bounded rationality in rational expectations). We can ask it economic models. First, there is abun- about unbounded rationality in general. dant empirical evidence that it is impor- We know there are critical physiological tant. Second, models of bounded ratio- limits on human cognition (Herbert Si- nality have proved themselves in a wide mon 1990, p. 7), but are the limits im- range of impressive work. Third, the portant to economics? Do they spoil any standard justifications for assuming un- of the good story told by the standard bounded rationality are unconvincing; theory of optimizing behavior? To be their logic cuts both ways. Fourth, delib- clear, the question is not whether eration about an economic decision is a bounds on rationality are always impor- costly activity, and good economics re- tant. They are not; there are many con- quires that we entertain all costs. These texts in which the hypothesis of un- four reasons, or categories of reasons, bounded rationality surely works well. are developed in the following four sec- Rather the questions are whether,bounds tions. Deliberation cost will be a recur- on rationality are often enough impor- 669 670 Journal of Economic Literature, Vol. XXXIV (June 1996) tant to include in economic analysis and, gerate confirming over disconfirming if so, when. evidence relative to initial beliefs; give The evidence sketched in this section answers that are highly sensitive to logi- will be put in two categories, direct evi- cally irrelevant changes in questions; do dence and confounded evidence, though redundant and ambiguous tests to con- the dividing line is vague. The "direct" firm an hypothesis at the expense of de- category will concern studies, mostly ex- cisive tests to disconfirm; make frequent perimental, which test economic ration- errors in deductive reasoning tasks such ality more or less directly by testing the as syllogisms; place higher value on an cognitive abilities relevant to economic opportunity if an experimenter rigs it to decisions. The "confounded" category be the "status quo" opportunity; fail to will concern tests in which rationality hy- discount the future consistently; fail to potheses are entertained jointly with adjust repeated choices to accommodate other hypotheses in economic settings. intertemporal connections; and more. In such experiments, the mental tasks A. Direct Evidence-Rationality Tests put to people are often simple, at least on Single Individuals relative to many economic decisions; There are many studies in which single whereas their responses are frequently individuals are faced with decisions way off. Most important, reasoning er- which have objectively correct answers rors are typically systematic. Psycholo- and which test the kinds of reasoning gists hypothesize that subjects make sys- frequently ascribed to agents in eco- tematic errors by using decision nomic theory. Do subjects do well in "heuristics,"or rules of thumb, which fail such tests? Often not. to accommodate the full logic of a deci- Hundreds of studies of this type have sion, as when a person makes systematic been done, mostly by psychologists but forecast errors by using adaptive rather more recently by experimental econo- than rational expectations. The system- mists also. There is a mountain of ex- atic errors are often referred to as "bi- periments in which people: display ases," and the general topic often carries intransitivity; misunderstand statistical the label "heuristics and biases." independence; mistake random data for The sheer number of experiments re- patterned data and vice versa; fail to ap- porting biases is so great that a sizable preciate law of large number effects; fail number of books and long survey papers to recognize statistical dominance; make have been written just to review the evi- errors in updating probabilities on the dence. For example, see the books by basis of new information; understate the Hal Arkes and Kenneth Hammond significance of given sample sizes; fail to (1986), Robin Hogarth (1980), Daniel understand covariation for even the sim- Kahneman, Paul Slovic, and Amos Tver- plest 2X2 contingency tables; make false sky (1982), and Richard Nisbett and Lee inferences about causality; ignore rele- Ross (1980); and see the survey papers vant information; use irrelevant informa- by John Payne, James Bettman, and Eric tion (as in sunk cost fallacies); exaggerate Johnson (1992), Gordon Pitz and Natalie the importance of vivid over pallid evi- Sachs (1984), and Slovic, Sarah Lichten- dence; exaggerate the importance of fal- stein, and Baruch Fischhoff (1988). For lible predictors; exaggerate the ex ante mini-surveys aimed at economists, see probability of a random event which has and already occurred; display overconfidence (1989), Tversky and Thaler (1990), and in judgment relative to evidence; exag- Kahneman, Jack Knetsch, and Thaler Conlisk: Why Bounded Rationality? 671

(1991). For examples of bias experiments initial expertise, better opportunities to by economists, see David Grether and learn, and the like. Although such design Charles Plott (1979) on preference re- conditions do attenuate biases, the at- versals, Grether (1992) on Bayes rule tenuation is typically limited. The pre- tests, and John Sterman (1989) and Rich- vailing overall impression is that biases ard Herrnstein and Drazen Prelec (1991) are not fragile effects which easily disap- on suboptimal decisions in the face of pear, but rather substantial and impor- dynamic complications. tant behavioral regularities. On debias- At the same time that psychologists ing, see the discussions in Raymond view heuristics as a source of bias, they Battalio, John Kagel, and Komain Jiran- also view heuristics as critical to problem yakul (1990, p. 28), Berndt Brehmer solving (Rudolf Groner, Marina Groner, (1980), Grether (1992), Hogarth (1980, and Walter Bischof 1983; Allan Newell ch. 5), Fischhoff (1982), Nisbett and Simon 1990, section II; and Payne, and Ross (1980, pp. 251-54), Payne, Johnson, and Bettman 1993). At first Bettman, and Johnson (1992, pp. 106, glance, this seems puzzling. Why not 114-16), Robert Slonim (1994), and condemn problem solving which leads to Slovic, Lichtenstein, and Fischhoff systematic error? The answer is simple. (1988, pp. 683-85, 688-89). Deliberation cost. For a boundedly ra- As Vernon Smith (1989, 1991) and tional individual, heuristics often provide Smith and James Walker (1993) empha- an adequate solution cheaply whereas size, market discipline, through repeated more elaborate approaches would be un- transactions with significant stakes, can duly expensive. As Pitz and Sachs (1984, be potent in attenuating discrepancies p. 152) put it, "a tradeoff exists between between optimizing and observed behav- cognitive effort and judgmental accu- ior; whereas psychological studies of de- racy." It is ironic that this economic biasing typically do not include such tradeoff should be better recognized in market forces. However, Smith and psychology than in economics. Experi- Walker also emphasize that attenuation mental and selected other economists is a matter of degree. They take a delib- recognize the tradeoff, but it tends to be eration cost view, arguing that decision pushed out of sight in economics by the makers try "to achieve a balance between emphasis on unbounded rationality. the benefits of better decision making This summary obviously stresses nega- and the effort cost of decision" (1993, p. tive evidence. There are also many ex- 260) and that "there are both low stake periments in which subjects reason accu- and high stake economic decisions in rately, especially after practice. In life, and all are of interest" (p. 249). Ex- principle, we expect that virtually any periments described in Smith and clear cut reasoning error can be made to Walker and in Mark Pingle (1992) verify disappear through an experiment which the importance of deliberation cost. See provides adequate incentive and which also Day and Pingle (forthcoming) and cleverly enough exposes or punishes the Bruno Frey and Reiner Eichenberger error. Ultimately, we would like to know (1994). when and why people get it right or There is one other source of (more or wrong. Psychologists have addressed this less) experimental evidence, vast in question through "debiasing"tests-tests quantity and intimately familiar to aca- of whether biases will diminish or disap- demic economists. Course exams. We pear when experiments are designed to carefully administer mental tests of eco- give subjects stronger incentives, greater nomic reasoning to many thousands of 672 Journal of Economic Literature, Vol. XXXIV (June 1996)

student subjects; and we provide sizable standard life cycle theory. Relative to the incentives for getting the right answers theory, with or without liquidity con- (namely grades, later redeemable for straints, people seem to be inefficient in scholarships, higher starting salaries, and smoothing consumption over the life cy- other large economic rewards). Though cle. Various studies report that the young we teach that agents act as if unbound- and the old consume too little, that con- edly rational, we use gallons of red ink to sumption is unduly sensitive to short run inform students that they do not. income fluctuations, that consumption is In summary, the bias evidence sug- not sensitive enough to expected future gests that people are capable of a wide changes in income, and that consump- variety of substantial and systematic rea- tion is improperly sensitive to the com- soning errors relevant to economic deci- position of wealth and income (Thaler sions. Further, the evidence suggests 1990; Christopher Carroll 1994; John that the magnitude and nature of the er- Shea 1995; and references there). Aggre- rors are themselves systematically re- gate consumption data also display ex- lated to economic conditions such as cess sensitivity of consumption to in- deliberation cost, incentives, and expe- come (Marjorie Flavin 1981, 1993); and rience. In this sense, investigaton of tests of representative agent models of bounded rationality is not a departure consumption, output, and asset prices from economic reasoning, but a needed are routinely rejected (Kenneth Single- extension of it. ton 1990). Angus Deaton (1992) dis- cusses some of these anomalies in his B. Confounded Evidence-Testing survey of consumption behavior. Economic Rationality Jointly In purchasing large appliances, con- with Other Hypotheses sumers tend to buy models with low Turn next to tests of predictions based price and high energy use even though, on both unbounded rationality and other at plausible discount rates, the initial hypotheses. If the predictions fail, expla- price saving does not compensate for the nations are confounded. We can't be later energy dissaving, as if consumers sure which hypothesis is at fault. What were myopic (Jerry Hausman 1979; Der- follows are examples, or "anomalies,"for mot Gately 1980; Loewenstein and which a case can be made (i) that con- Thaler 1989, pp. 182-83). In purchasing ventional economic theory is at odds flood and earthquake insurance, consum- with the evidence, (ii) that bounded ra- ers also appear to make inefficient tionality provides a possible reconcili- choices; see the large study by Howard ation, and (iii) that economists have not Kunreuther et al. (1978). In the fore- agreed on a better reconciliation. The word, Kenneth Arrow describes the re- examples are merely cited, not argued; sults as "certainly disconcerting from the the citations develop the arguments and point of view of generally accepted the- give many further references. Richard ory" (p. vii). Appliances and insurance Thaler is probably the leading anomaly are purchases for which consumers may hunter among economists. A number of have little experience or training, and for the citations are to Thaler's anomaly col- which the deliberation and other costs of umns in the The Journal of Economic expertise may be large relative to poten- Perspectives. Many of the columns are tial benefits. collected in Thaler (1992). Expectations. Survey data on expecta- Consumer behavior. Household con- tions of inflation and other variables sumption data are often at odds with commonly reject the unbiasedness and Conlisk:Why Bounded Rationality? 673

efficiency predictions of rational expec- ties (Charles Lee, Andrei Shleifer, and tations (John Cragg and Burton Malkiel Thaler 1991); excess trading volume on 1982; K. Holden, D. A. Peel, and J. L. shares that have risen in price relative to Thompson 1985, ch. 3; Lovell 1986; Jef- volume on shares that have fallen in frey Frankel and Kenneth Froot 1987; price ( and Meir Statman Werner De Bondt and Thaler 1990; 1985); predictability from lagged insider Takatoshi Ito 1990). Rational expecta- trading data (Nejat Seyhun 1992); and tions can also be tested, jointly with more (see the De Bondt and Thaler 1994 other hypotheses, in experiments. The survey). On the significance of anomalies classic probability matching experiments for market-beating portfolios, see in psychology rejected rational expecta- Hashem Pesaran and Allan Timmermann tions as early as the 1950s (Sidney Win- (1995). Various related anomalies are ter 1982). Data from recent experimen- found in foreign exchange rate markets; tal asset markets favor adaptive over see the surveys by Froot and Thaler rational expectations (Smith, Gerry (1990), Karen Lewis (1994), and Frankel Suchanek, and Arlington Williams 1988; and Andrew Rose (forthcoming). Plott and Shyam Sunder 1988; Ramon David Cutler, James Poterba, and Marimon and Sunder 1993; Steven Pe- Lawrence Summers (1991, p. 529) sug- terson 1993; John Hey 1994); although gest four stylized facts as giving overall the experimenters note that experienced pattern to the anomalous price behavior subjects move toward rational expecta- of stocks, bonds, foreign exchange, and tions. The evidence suggests that expec- some real assets (housing, collectibles, tations may or may not be rational, de- and precious metals): pending on experience, difficulty of the First, returns tend to be positively serially forecasting task, and other conditions. correlated at high frequency. Second, they Asset prices. Despite the presence of are weakly negatively serially correlated over long horizons. Third, deviations of asset val- highly experienced and motivated trad- ues from proxies for fundamental value have ers, financial markets generate numerous predictive power for returns. Fourth, short anomalies. According to the efficient term interest rates are negatively correlated with excess returns on other assets. markets hypothesis, arbitrage should force predictability out of stock price Decision experiments. The "asset inte- changes. Yet stock prices display: slow gration" hypothesis-that an agent's ob- mean reversion (De Bondt and Thaler jective function is defined on total 1985; Eugene Fama and Kenneth wealth rather than on changes in French 1988); predictable end-of-week, wealth-has the status of a rationality end-of-year, seasonal, and holiday effects postulate in the sense that total wealth, (Thaler 1987; Josef Lakonishok and Sey- not change in wealth, is what dictates the mour Smidt 1988); excess fluctuation in agent's opportunity set for consumption, prices relative to fluctuation in funda- the ultimate conveyor of utility. There is mentals (Stephen LeRoy 1989; Robert substantial evidence from decision ex- Shiller 1989); dramatic bubbles unex- periments that asset integration often plained by changes in fundamental val- fails (Camerer 1992; Battalio, Kagel, and ues ( 1989; and Smith, Jitanyakul 1990; Robert Gertner 1993). Suchanek, and Williams 1988); excess The issue is greatly complicated by the risk premia relative to bonds (Rajnish fact that the relevant wealth concept is Mehra and Edward Prescott 1985); sys- lifetime wealth, implying that asset inte- tematic deviation of mutual fund prices gration effects are confounded with in- from the values of the component securi- tertemporal choice effects. 674 Journal of Economic Literature, Vol. XXXIV (June 1996)

Individuals often express a substan- (1987) on second-price auctions. For tially lower willingness to pay than will- first price auctions, Smith emphasizes ingness to accept for a marginal unit of a that the suggestion of overbidding is commodity. Although there is contro- relative to risk neutrality and that the versy over magnitudes, and although ef- bidding pattern, over wide variation in fects diminish with practice (as in re- experimental stakes, can be interpreted peat-trial experiments), the body of coherently as a risk aversion effect. This results is hard to reconcile with standard interpretation, however, strains other di- models of economic rationality. Tversky mensions of standard theory. The stakes and Kahneman (1991, p. 1054) and Ray- in the auctions, though varying widely, mond Hartman, Michael Doane, and are still small relative to subjects' base Chi-Keung Woo (1991) emphasize the wealth, whereas risk aversion is a second magnitude of the anomaly. Jason order effect. Thus, substantial risk aver- Shogren et al. (1994) and Robert Fran- sion effects require that we either give ciosi et al. (1995) emphasize the mitigat- up asset integration or give up declining ing effects of experience. As a related absolute risk aversion. Further, there is anomaly, individuals seem to place the conflict that, in other small-stake ex- higher value on an opportunity if it is as- periments, risk-seeking, risk-neutrality, sociated with the status quo; see William and risk-aversion are all found, both over Samuelson and Richard Zeckhauser losses and over gains (Battalio, Kagel, (1988). For example, Knetsch (1989) and Jiranyakul 1990, section 3.2; refer- finds that most students first given a ences in Conlisk 1993a, p. 259). fancy mug then refuse to trade it for a Many experiments test whether sub- large chocolate bar, whereas most stu- jects will contribute to public goods or dents first given a large chocolate bar will free ride (John Ledyard 1995; Robyn then refuse to trade it for a fancy mug. Dawes and Thaler 1988). The optimal Experimental auctions and games. In selfish strategy is free riding, whereas experiments on common value auctions, the experiments show substantial contri- there is evidence of systematic overbid- butions. Although most experiments can- ding relative to theoretical predictions. not distinguish whether departures from This "winner's curse" is also found in selfish optimality are due to decision er- some real auctions (Alvin Roth 1988, ror or to altruism, Thomas Palfrey and section III; Thaler 1988; Orley Ashenfel- Jeffrey Prisbrey (1993) design an ingen- ter and David Genesove 1992). How- ious experiment which allows the distinc- ever, Susan Garvin and Kagel (1994) find tion. They find substantial decision error that the winner's curse disappears in ex- and little altruism. periments when subjects gain substantial In the large experimental literature on experience through repeated auctions game theory, predictions based on the with the same design settings; and James usual strong rationality postulates are Cox, Samuel Dinken, and Smith (1995) often violated (and often not). See Ana- find a stronger disappearance effect tol Rapoport, Melvin Guyer, and David when subjects are allowed to withdraw Gordon (1976), Roth (1988), Camerer from bidding to an alternative activity (1990), Camerer et al. (1993), Dale Stahl yielding a positive safe payoff. and Paul Wilson (1994), and references Related experiments suggest overbid- there. ding in private-value auctions. See Smith What are we to make of such anoma- (1989, p. 158) on first price auctions and lies? Some may yield to further optimiz- Kagel, Ronald Harstad, and Dan Levin ing theory. Others, however, seem to Conlisk: Whu Bounded Rationalittl? 675 achieve anomalous status only because need to economize on various transac- economists push optimizing theory too tion costs. Williamson (1986, p. 110) far. For example, there seems little traces transaction costs to agents' limited doubt that consumption-smoothing be- cognitive abilities: "Economizing on havior is observed and that competition transaction costs essentially reduces to is a powerful force in squeezing predict- economizing on bounded rationality. ability out of stock price changes. How- Williamson's work has had huge impact ever, as theories of these successful ideas on the literature of industrial organiza- push to finer and finer margins of opti- tion (Richard Schmalensee and Robert mality, predictions begin to defy the Willig 1989) and organizational design data, as if rationality, a matter of degree, (Radner 1992). Although many organiza- is being pushed too hard. Arrow (1986) tional theorists avoid mention of discusses how much the computational bounded rationality, preferring imper- power attributed to agents has increased fect information hypotheses to imperfect as economic theory has evolved. Anoma- rationality hypotheses, some do not. For lies are not surprising relative to theories example, Raaj Kumar Sah and Joseph which neglect deliberation cost, experi- Stiglitz (1988) and Joel Sobel (1992) ana- ence, and other conditions bearing on lyze organizational designs to protect how close to unbounded rationality it is against the mistakes of fallible workers possible or sensible to be. Fortunately, and decision makers. the anomalies suggest not just shortcom- X-Inefficiency. An organization can be ings of standard theory but also direc- inefficient because its outputs lie at the tions for improved theory. Many of the wrong point on an efficiency frontier or models surveyed in the next section are because its outputs lie inside the effi- motivated by anomalies. ciency frontier. The latter was dubbed "X-inefficiency" by Harvey Leibenstein II. Bounded Rationality in Economic (1966), who pioneered in its study. Models: A Sampler There is now a sizable body of theo- retical and empirical work on X-ineffi- Though a small fraction of the total lit- ciency, much of it rooted in notions of erature on economic theory, there are bounded rationality. See Leibenstein many models which allow for bounded (1987), Leibenstein and Shlomo Maital rationality. This section is a sampler. The (1994), Roger Frantz (1992), and refer- models spread in all directions, making ences there. them hard to categorize. The categories Boundedly rational choice-early used, though each has its own logic, models. In standard optimizing theory, overlap in various ways. agents act as if they perform exhaustive Firms, organizations, and institutions. searches over all possible decisions and (1937, 1992), Alfred Chan- then pick the best. Simon (1955, 1987) dler (1962), Richard Cyert and James hypothesizes that agents instead perform March (1963), March and Simon (1968), limited searches, accepting the first sat- Oliver Williamson (1985, 1986), and Ja- isfactory decision. This "" hy- cob Marschak and Roy Radner (1972) pothesis is the direct inspiration for a are pioneers in analyzing the nature of number of the models cited below (in- firms, organizations, and economic insti- cluding Day and Herbert Tinney 1968; tutions. A central insight is that the exis- Winter 1971; and Radner and Rothschild tence, size, structure, and workings of 1975), and the spirit of the idea is perva- organizations are critically shaped by a sive. 676 Journal of Economic Literature, Vol. XXXIV (June 1996)

A related idea is suboptimization. A (1972) theory of "elimination by as- decision maker who finds optimization pects," an individual chooses among al- impossible or unduly costly may instead ternatives, not by comparing alternatives solve a simpler, approximate optimiza- in all their aspects at once, but rather by tion problem. Because errors due to the heuristic of comparing alternatives suboptimization in one period may call one randomly chosen aspect at a time, for adjustments the next, it is natural to eliminating alternatives along the way. embed suboptimization in a dynamic Heuristics are rational in the sense that context which generates feedback. Al- they appeal to intuition and avoid delib- though suboptimization with feedback eration cost, but boundedly rational in has a long history (for example, dynamic the sense that they often lead to biased Cournot models), Day and colleagues choices. Sociologists and anthropologists were the first to develop the idea into a also study behavioral rules relevant to broad and coherent approach, calling it economics, often in the form of social "recursive programming." Such models norms and conventions (Jon Elster generate rich dynamics, foreshadowing 1989). Biases, heuristics, and norms have the recent interest in complex economic been used in various economic models to dynamics. For theory and numerous em- explain otherwise puzzling behavior. pirical applications, see Day (1963), Day Winter (1982) uses learning heuristics and Alessandro Cigno (1978), and refer- to explain experimental results on "prob- ences there. Hierarchical decision mod- ability matching" violations of rational els (Ermini 1987, 1991) also have a re- expectations. George Akerlof and Wil- cursive programming form. liam Dickens (1982) and Markup pricing, adaptive expectations, (1994) use cognitive dissonance (the bias partial adjustment, imitation, and sto- of fitting beliefs to convenience) to chastic choice are examples of more pas- model worker safety, innovations, adver- sive decision making. In the John Cross tising, social security, crime, and morally (1973, 1983) and Susan Himmelweit dubious behavior. Akerlof (1991) uses sa- (1976) models of stochastic choice, for lience (the bias of attaching undue example, an agent chooses at random weight to recent or vivid events) to ex- among a list of possible actions, where plain why people may procrastinate or the choice probabilities evolve according show excessive obedience to authorities, to the historical performances of the and he shows how small effects of this various possibilities. Cross and Himmel- sort may cumulate into inadequate sav- weit apply the theory to store choice, lot- ing, organizational failure, addiction, and tery choice, advertising, supply deci- crime. Akerlof and Janet Yellen (1987) sions, and other issues. Rajiv Sarin use biases to sketch microfoundations for (1994) applies the theory to evolutionary traditional Keynesian analysis. Tversky game theory. The approach has prece- and Kahneman (1991) use loss aversion dents in psychological learning theory (greater marginal sensitivity to losses and is a rough precedent for the eco- than to gains) to explain various behav- nomic classifier models discussed below. ioral puzzles representable as deforma- Boundedly rational choice-heuristics, tions of an indifference map in the norms, and other imports from sister dis- neighborhood of a current consumption ciplines. Psychologists and cognitive sci- point. Shlomo Benartzi and Thaler entists study heuristics (rules of thumb) (1995) add myopia to loss aversion to by which people deal with their cognitive propose a resolution of the equity pre- limitations. For example, in Tversky's mium puzzle. Brian Arthur (1994) con- Conlisk:Why Bounded Rationality? 677 siders a model in which individuals shift norms in question might be inherited among a menu of possible heuristics as either biologically or culturally; see the experience dictates. Leibenstein and dual inheritance models of Robert Boyd Maital (1994) use defensive bias (ration- and Peter Richerson (1985). Norms alization of error) in a game theoretic might be the cause of bounds on indi- model of X-inefficiency. vidualistic rationality. Or norms might be Time inconsistency can be viewed as the effect of bounded rationality; Simon multiple selves bounding each other's ra- (1993) argues that docility to social tional choices; the Doer Self wants des- norms improves economic fitness by in- sert whereas the Planner Self wants to ducing people to augment their limited stick to the diet (terminology from rationality with the collective wisdom of Thaler and Shefrin 1981). Many behav- their social group. ioral rules (for example, don't keep des- Ronald Heiner (1983, 1989) develops sert in the house) arise as responses to the stimulating hypothesis that economic such conflicts (Thomas Schelling 1984). behavior is predictable in large part be- Negative time preference is a major cause bounded rationality leads people source of time inconsistency. Loewen- to adopt rules of thumb which display stein and Prelec (1992) use loss aversion greater regularity than does optimiza- to explain time inconsistency flowing tion. Thus, Heiner argues, standard eco- from negative time preference. To ex- nomics is subject to an ironic misspe- plain consumption anomalies, Shefrin cification problem: "the observed and Thaler (1988) and Thaler (1990) hy- regularities that economics has tried to pothesize budgeting heuristics; income is explain on the basis of optimization allocated to different accounts, such as a would disappear if agents could actually current spending account and a retire- maximize"(1983, pp. 586-96). ment account, with transfers across ac- . Many of the counts not allowed. models cited above and below are evolu- Some economists argue that inherited tionary models-dynamic models in emotions and social norms (anger, em- which more successful agents and ac- barrassment, sensitivity to relative posi- tivites gradually increase their share of tion, loyalty, altruism) can improve eco- the economy at the expense of less suc- nomic performance in ways outside the cessful agents and activities. The evolu- scope of standard theory. For example, tionary approach to economics has a long loyal individuals cooperate better, and a history, is currently experiencing an up- person who involuntary blushes at a lie is swing of interest, and might be the better able to win trust. Amartya Sen longer run mainstream to which econo- (1977) refers to the selfishly rational mists return, encompassing optimization agents of economic theory as "rational models as a special case. Core ideas of fools" because they lack these advantages the evolutionary approach have been sur- of emotions and norms. For models of veyed recently by Richard Nelson (1995) the advantages, see Akerlof (1984) on in this Journal. The approach is espe- loyalty and on gift exchanges, Jack cially well suited, for example, to analyz- Hirshleifer (1987) on emotions as guar- ing growth and technical change. Be- antors of threats and promises, Robert cause the rate of technical change is Frank (1985, 1988) on emotions and on limited in large part by agents' bounded sensitivity to relative position, and ability to perceive and exploit opportuni- Schelling (1978) and Simon (1993) on ties for improving production processes, the fitness of altruism. The emotions and technical change relates more naturally 678 Journal of Economic Literature, Vol. XXXIV (June 1996)

to bounded than to unbounded rational- severity of fluctuations in a market di- ity, and more naturally to evolutionary rectly to the deliberation cost of individ- than to equilibrium approaches. The pio- ual firms; depending on conditions, fluc- neering work is Nelson and Winter tuations may increase or decrease when (1982); more recent models include Sil- deliberation cost increases. verberg, Giovanni Dosi, and Luigi Stimulated by empirical anomalies, a Orsenigo (1988), Conlisk (1989), Elias- number of authors investigate the effect son (1991), and various other models of boundedly rational traders in asset cited in Nelson (1995, section IV). markets. Bradford De Long et al. (1990) Bounded rationality and market out- show how boundedly rational traders, by comes. The effect of bounded rationality accepting "too much" risk and thus earn- on market outcomes is a central ques- ing higher returns than unboundedly ra- tion. Does it make a difference, or are tional traders, may come to dominate an outcomes the same as if all agents opti- asset market. Shefrin and Statman mized? What are the differences and (1994) develop a "behavioral capital as- when do they occur? The issues are ex- set pricing model" in which some traders plored in many models. Some are evolu- display reasoning errors suggested by the tionary models with full dynamics; others bias literature from psychology. The investigate only equilibria. model provides a broad theory which In a classic early paper, Winter (1971) covers return anomalies and the survival shows how, under strong conditions, of boundedly rational agents, along with market competition may select for sur- the usual asset pricing issues. Timothy vival only those firms which display "as Cason (1992) shows how learning may if' optimization. However, Ulrich Witt lead to market efficiency, whereas Tim- (1986) addresses the same issue and mermann (1995) shows how learning finds convergence to optimization under may itself be the mechanism leading to some circumstances and not under oth- various anomalies. ers. Conlisk (1983) shows how a broad In an old and very simple model, array of firm reaction functions may lead James Meade (1964, ch. V) showed that, a market to approximate perfect compe- as a result of saving and other effects, tition; and Dhananjay Gode and Sunder individuals with superior investment effi- (1993) give examples of double auction ciency need not accumulate wealth faster markets in which "zero-intelligence" than other individuals. Lawrence Blume traders (computers which bid randomly and David Easley (1992) elaborate the subject only to budget constraints) may logic with modern methods, finding that, achieve near perfect market efficiency. in an asset market, "fit rules need not be However, Thomas Russell and Thaler rational, and rational rules [need] not be (1985) show that a small reasoning error fit" (p. 9). by a fraction of the consumers in a mar- Timur Kuran (1991) explains how cog- ket may alter market equilibrium; and nitive limitations may influence the evo- John Haltiwanger and Michael Waldman lution of preferences, in which case the (1985, 1991) show how a small propor- whole notion of evolution to optimality tion of boundedly rational agents in a becomes problematic. market may have more than proportion- Evolution to rational expectations in ate influence on market equilibrium due markets. Among strong rationality hy- to congestion effects (or less than pro- potheses, rational expectations are spe- portionate effect under opposite condi- cial, for at least three reasons. First, tions). Conlisk (forthcoming) relates the expectations are critical to market out- Conlisk: Why Bounded Rationality? 679

comes, for example to macroeconomic choice. Second, there may be correlation policy. Second, because departures from across individuals in these decision er- rational expectations can be detected rors, due to common responses to without knowing utility functions, ra- changes in the economy. From these in- tional expectations are easier to test than sights, Akerlof and Yellen demonstrate most implications of unbounded rational- that a fraction of boundedly rational ity. Third, because an agent's rational ex- agents in an economy, though suffering pectation depends on knowledge of an utility or profit losses which are only sec- entire market or economy, not just on ond order small, may cause first order knowledge of the agent's own narrow cir- effects on market outcomes. They give cumstances, calculation of a rational ex- various examples; other examples are in pectation may involve high deliberation Stephen Jones and James Stock (1987), cost. Howard Naish (1993), and references There has been great interest in there. whether adaptation might lead agents to Near rationality models suggest that rational expectations. The answer de- the benefit of upgrading from bounded pends on exact conditions-on whether to unbounded rationality may be small. the context is simple enough, on whether At the same time, computational com- agents' prior beliefs are compatible with plexity models suggest that the delibera- the context, on how agents process new tion cost of upgrading may be sizable, information, and so on. Overall, authors even astronomical. For example, in many are quite cautious about claiming sup- integer programming problems (such as port for rational expectations on adaptive scheduling, capital budgeting, cargo grounds. See Margaret Bray (1982), loading, and itinerary problems), compu- Blume and Easley (1982), Bray and tational complexity increases exponen- David Kreps (1987), Thomas Sargent tially with problem size; see Christos (1993), Jerome Detemple and Shashid- Papadimitriou and Kenneth Steiglitz har Murthy (1994), and George Evans (1982) and Silvano Martello and Paolo and Seppo Honkapohja (1995). In an Toth (1990). A classic example from empirical extension of this work, Tim- game theory is chess. The optimal strat- mermann (1994) asks whether stock mar- egy in chess is conceptually simple, just ket investors in the U.K. could have as in tic-tac-toe, because both games in- learned rational expectations from the volve only a finite number of possible se- historical record. He makes a case that quences of play. However, for chess this they could not unless they had good number is "comparableto the number of prior information about long-run proper- molecules in the universe" (Simon and ties of stock price series. Jonathan Schaeffer 1992, p. 2). Simon Near rationality, complexity, and mar- (1990, p. 6) concludes, "If the game of ket outcome. Akerlof and Yellen (1985) chess, limited to its 64 squares and six combine two insights to argue the impor- kinds of pieces, is beyond exact computa- tance of bounded rationality to market tion, then we may expect the same of al- outcomes. First, because objective func- most any real-world problem . . ." tions are often flat at their optima, an Self-organizing markets. Suppose, in a agent may be "near rational"in utility or society with no organized market for a profit achieved, but far from unbounded certain good, potential buyers and sellers rationality in terms of action taken. See meet pairwise at random, agreeing to John Cochrane (1989) for a striking ex- trade if their reservation prices allow, ample on intertemporal consumption otherwise adjusting their reservation 680 Journal of Economic Literature, Vol. XXXIV (June 1996) prices before the next period. Because some sort, making- adaptive transitions each new period produces new pairings, among the categories as time passes. there is an expanding web of effects con- Transitions are governed by imitation, necting the whole population. It may oc- fitness-sensitive reproduction, or other cur that all reservation prices converge mechanisms. (Evolutionary game models to an equilibrium price, in which case a often fit this description, but are dis- market is born, despite the absence of an cussed under separate heading below.) organizing institution. The market is In Michael Farrell (1970), investors dis- self-organizing. Jacques Lesourne (1992, tribute over wealth states. In Winter 1993) and colleagues have developed an (1971), firms distribute over profitability enlightening series of models of self- states. In the older diffusion models sur- organizing markets. The models deal veyed in David Bartholomew (1982), and with the birth of intermediaries, the in the newer models of Arthur (1989) emergence of speculators, the formation and Glenn Ellison and Drew Fudenberg of opinions,- the generation of sunspot (1993), agents distribute over techno- equilibria, the founding of unions, logical states, informational states, dis- changes in the structure of competition, ease states, or the like. In Conlisk and the effects of critical maverick (1980), Waldman (1994), and John Har- agents. Such models, with their rich in- rington (1994), individuals distribute teractions among adaptive agents, are at over decision-making rules. In Edmund an opposite pole from representative Phelps and Winter (1970), Dennis Small- agent models, which sacrifice nearly all wood and Conlisk (1979), Mark Gra- interactions to pursue optimization. novetter and Roland Soong (1986), Alan In summary of the last four topics, Kirman (1993), and Ellison and Fuden- does bounded rationality alter market berg (1995), buyers distribute over sell- outcomes? The models answer with a re- ers (among other interpretations). Al- sounding maybe. Depending on circum- though most of these models view stances, boundedly rational agents may behavior as boundedly rational, some or may not self-organize into markets. If treat agents as perfectly rational but im- a market is already organized, boundedly perfectly informed. See Abhijit Banerjee rational agents may have no special ef- (1992, 1993), Sushil Bikhchandani, fect at all, may affect either the level or David Hirschleifer, and Ivo Welch variability of price and output, may have (1992), and Welch (1992) on herding, effects that are less or more than propor- fads, and informational cascades. tionate to their numbers, and may have Such models are especially useful for second order effects on themselves but investigating direct interactions among first order effects on the market (or the individuals, as opposed to indirect inter- opposite). With experience, boundedly actions through market prices, a distinc- rational agents may or may not learn tion stressed in Kirman (1994). Among more accurate behavioral rules, may do the interactions considered are imitation, better or worse than unboundedly ra- word-of-mouth communication, fads and tional agents in the short run, and may fashions, bandwagons, threshold effects, disappear, dominate, or coexist in the herding, increasing returns, lock-ins, and long run. Thus, bounded rationality mat- informational cascades. ters, but not in a simple way. Various themes emerge from this lit- Population distribution models. In a erature, but not an encompassing pat- common type of model, a population of tern. In a number of the models, one can individuals distributes over categories of ask whether the population converges to Conlisk: Why Bounded Rationality? 681 a well behaved outcome, such as a mar- rematchings from a population. Depend- ket equilibrium, highest quality brand, or ing on conditions, models may or may best technology. The typical answer is not evolve to Nash equilibria. For re- maybe, depending on exact conditions. views of evolutionary games, see John There may be convergence to a unique Creedy (1992), Daniel Friedman (1993), best outcome. Or there may be multiple and Marimon and Ellen McGrattan stable equilibria such that even the worst (1995); and, for symposia, see George outcome can attract the entire popula- Mailath (1992), Crawford (1993), and tion when initial conditions or random Day (1993b). Especially interesting are events dictate. For example, increasing papers which investigate evolutionary ra- returns to brand popularity may lead any tionales for rules of thumb and conven- brand, regardless of quality, to dominate tions. For example, see the theoretical a market (as in Smallwood and Conlisk discussions of Crawford (1993, 1995), 1979); or increasing returns to usage may Robert Rosenthal (1993), Peyton Young lead a single technology, regardless of (1993), Fernando Vega-Redondo (1993), objective efficiency, to dominate an in- and Joel Watson (1994); and see the em- dustry (as in Arthur 1989). In some pirical studies of John Van Huyck, cases, more boundedly rational behavior Joseph Cook, and Battalio (1994). leads to better equilibria (Smallwood In a different game approach, and Conlisk 1979; David Lane and bounded rationality takes the form of re- Roberta Vescovini 1995). Or there may strictions on available strategies. For ex- be no equilibria at all, leaving the popu- ample, players may be restricted in the lation to fluctuate forever, either peri- complexity of strategies they are able to odically or chaotically. There may be implement (Dilip Abreu and Ariel Ru- quasi-equilibria in which the model rests binstein 1988; Vega-Redondo 1994); for substantial periods of time, only to players may be restricted to a subset of cascade off to another quasi-equilibrium actions (Sobel 1991); or players may be on the occurrence of a small-probability restricted in the type of inference they event. A long-standing theme, more re- display (Stahl 1993). cently labelled "path dependence," is Dynamics and simulation. Bounded that initial conditions and chance events rationality is often modeled as some may dictate outcomes. Thus, "history form of dynamic adaptation. Using ob- matters" in determining "emergent servation and intuition, modelers endow structures" (Paul David 1985, 1986; Ar- agents with adaptive behavioral rules for thur 1989). interacting within some assumed envi- Games. Game theorists have recently ronment, then set the dynamic in mo- turned to bounded rationality with en- tion. Due to model complexity, simula- thusiasm, either to address experimental tion is common. Such models include the anomalies, or to provide a dynamic for large macromodels predating the "ra- selection among multiple equilibria, or tional expectations revolution," the perhaps simply because game theory, tatonement price adjustment models having pushed rationality to the furthest now out of fashion, various micro-simu- extreme, was ripest for a revision. In lation models (Guy Orcutt, Joachim evolutionary game models, a game is Merz, and Hermann Quinke 1986; played repeatedly, and players modify Robert Bennett and Barbara Bergmann their strategies in light of payoff experi- 1986; Eliasson 1991), dynamic versions ence. The repetitions of play may involve of computational general equilibrium the same pair of opponents or random models (surveyed in Alfredo Pereira and 682 Journal of Economic Literature, Vol. XXXIV (June 1996)

John Shoven 1988), some parts of the by which a decision maker turns scarce system dynamics literature (Michael cognitive and other resources into better Radzicki and Sterman 1994), and the decisions. The deliberation cost theme classifier models discussed under the pervades the discussion above; yet very next heading. Day (1994) relates dynam- few explicit models of deliberation tech- ics and simulation to the mathematical nology and deliberation cost have ap- field of complex dynamics. peared. Day and Tinney (1968, Appen- Classifier systems. In a classifier dix), Marschak and Radner (1972, model of bounded rationality, each agent sections 9.6-9.7), and Reinhard Selten in each period chooses one among a dis- (1978) sketch deliberation cost models, crete list of actions. Choices are sensitive but do not develop them. to the historical rewards of the actions The first full model of deliberation according to primitive evolutionary rules cost (though not presented in such with a genetic flavor. Classifier systems terms) seems to be the model developed originated in the machine learning litera- by Radner and Rothschild (1975), Rad- ture as models of how a machine, start- ner (1975b), and Rothschild (1975). In ing from extreme ignorance, might by their model, a decision maker, facing trial and error come to adopt effective several planning activities, does not have decision rules in performing some task enough time to optimize every activity in (John Holland and John Miller 1991). every period. Thus, the implicit delibera- Classifier models have been used in eco- tion cost in attending to any one activity nomics, for example, by Marimon, is the reduced performance of other ac- McGrattan, and Sargent (1990) to study tivities. Most of the few other models of the emergence of a medium of exchange, deliberation cost are more recent. Con- by Nicolas Vriend (1994) to study self- lisk (1988, forthcoming), Evans and Ra- organized markets, and by David mey (1992, 1994, 1995), Smith and Midgley, Robert Marks, and Lee Cooper Walker (1993), and Andre De Palma, (1995) to study competitive strategies. A Gordon Myers, and Yorgos Papageorgiou controversial issue is whether classifier (1994) consider full deliberation tech- models, by assuming such elementary nologies in which agents choose the mag- trial-and-error learning, set the "dial of nitude of a costly deliberation input. A rationality" too far toward the primitive simpler approach is to suppose that extreme. Arthur (1993) argues that the agents choose among distinct behavioral dial can be calibrated in these models to rules, each carrying its own fixed delib- match empirically observed behavior, al- eration cost. This approach has been though his only example is a multi-arm used to study the fitness of cheap imita- bandit context, which by its nature virtu- tion relative to costly optimization (Con- ally excludes anything but simple trial- lisk 1980), the effect of bounded ratio- and-error learning. nality on game equilibria (Abreu and Economy of the mind-deliberation Rubinstein 1988; and Rosenthal 1993), technologies and deliberation cost. If ra- and aggregate technical change (Conlisk tionality is scarce, good decisions are 1993b). Ermini (1991) relates delibera- costly. There is a tradeoff between effort tion cost to hierarchical decision making; devoted to deliberation and effort de- and Gordon Winston (1989) discusses voted to other activities, reflecting what decision cost more broadly. Day (1993a) calls the "economy of the Collectively, these models show how a mind." A model of the tradeoff requires deliberation technology can merge stan- some form of "deliberation technology" dard modeling ingredients (optimization, Conlisk: Why Bounded Rationality? 683

rational expectations, market equilib- pers in the journal Economics and Phi- rium) with boundedly rational ingredi- losophy. ents (satisficing, learning, rules of Argument 1. "As if" rationality. The thumb). In such a context the "degree of question is not whether people are un- rationality" of a decision, relative to the boundedly rational; of course they are decision that would prevail under un- not. The question is whether they act ap- bounded rationality, is endogenously proximately as if unboundedly rational; determined, along with other model they do. outcomes, by economic forces. Section Comment. This hugely influential ar- IV returns to the deliberation cost gument of Milton Friedman (1953) is a theme. conditional argument. Do people in fact act as if unboundedly rational? Accord- III. Yes, But As If: Argumentsfor ing to the evidence cited in Section I, Unbounded Rationality they sometimes do and sometimes do not. Under the latter condition, by the The case for investigating bounded ra- logic of the "as if" argument, we should tionality has not been convincing to most investigate bounded rationality. economists. Arguments for optimiza- Argument 2. Learning. Though peo- tions-only modeling have held powerful ple's rationality is bounded, they learn sway, shaping the research, the teaching, optima through practice, in the end act- and the everyday conversations of econo- ing as if unboundedly rational. Econo- mists. The arguments are so familiar that mists can take a shortcut to the outcome a few code words are enough to conjure by assuming unbounded rationality from one up, as in, "Yes, but you don't under- the start. stand; no one assumes that people are Comment. Learning extends Argument unboundedly rational, only that they act 1 by suggesting how people come to act as if unboundedly rational." This section "as if" smarter than they are. However, works quickly through a list of eight the learning logic cuts both ways. Learn- prominent arguments for unbounded ra- ing is promoted by favorable conditions tionality, giving a brief comment on such as rewards, repeated opportunities each. The purpose is partly to review for practice, small deliberation cost at ideas which have made the literature each repetition, good feedback, un- what it is and partly to suggest more con- changing circumstances, and a simple structive versions of the arguments. A context. Conversely, learning is hindered number of the arguments in effect de- or blocked by the opposite conditions. scribe conditions under which un- That is the message of the numerous ex- bounded rationality seems a sensible as- periments cited in Section I and the nu- sumption. By inspecting the conditions merous models cited in Section II. The and their opposites, we can turn the ar- learning logic makes us expect that Argu- guments toward the more constructive ment 2 will sometimes apply and some- question of when and why bounded ra- times not. Economic issues involving tionality is likely to be important. For long horizons, such as life cycle decisions more extensive methodological discus- by individuals and technological evolu- sions, see Milton Friedman (1953), Tjall- tion by firms, are among the most impor- ing Koopmans (1957), Winter (1975), tant in economics, yet are among the Gary Becker (1976, 1993), Elster (1979, least likely to meet the conditions for ef- 1983), Sen (1977, 1987), Ermini (1987), fective learning. A young person making Robert Sugden (1991), and various pa- a life cycle plan gets no practice and 684 Journal of Economic Literature, Vol. XXXIV (June 1996) therefore no feedback; the problem is using such rules. Thus, theorists often enormously complex; and the environ- suggest that a good theory of an individ- ment is likely to change dramaticallyand ual's decisions must, as a survival condi- unpredictably during the person's life- tion for the individual, disallow the pos- time. The famous M. Friedman and sibility that the individual uses pumpable Leonard Savage (1948) billiards expert rules of thumb (Mark Machina 1989, pp. plays as if a master of the laws of physics. 1623-24; Hirshleifer and John Riley But what of a beginner taking the first 1992, section 1.6). However, the non- shot, in poor light, on a badly warped pumbability criterion is easily chal- and randomly moving table, with as- lenged. Although tricksters abound-at sorted friends and relatives guiding the the door, on the phone, and elsewhere- cue stick? Is a young person making life people can easily protect themselves, cycle decisions more like the expert with their pumpable rules intact, by such player or more like the beginner? simple devices as slamming the door and Argument 3. Survivors and tricksters. hanging up the phone. The issue is again Agents who do not optimize will not sur- a matter of circumstance and degree. vive. Argument 4. Don't quarrel with suc- Comment. The survival argument is as- cess. Economics is built on the postulate sociated with the classic papers by Ar- of unbounded rationality. Utility maximi- men Alchian (1950) and M. Friedman zation has been a powerful generator of (1953), and has been critically evaluated successful hypotheses. It is foolish to in general terms by many authors, nota- quarrel with such success. bly Winter (1964, 1975) and Nelson and Comment. Becker, never shy about Winter (1982). Early on, Koopmans this argument, claims that "all human be- (1957, pp. 140-41) advised formal mod- havior can be viewed as involving par- eling of the hypothesis; and there are by ticipants who maximize their utility from now the numerous models cited in Sec- a stable set of preferences and accumu- tion II. The models show Argument 3 to late an optimal amount of information" be highly conditional. Nonoptimizing (1976, p. 14, emphasis added). To accept firms survive under some conditions but the argument, however, we have to grant not under others. In the presence of de- (a) that the existing success of economics liberation cost, for example, survival should be credited to optimization hy- logic may favor a cheap rule of thumb potheses and (b) that expanded treat- over a costly optimization. ments of rationality would not lead to The survival argument carries lesser greater success. The models of the pre- force for individuals than for firms. We ceding section are the rebuttal to (b). commonly read in the financial pages Consider (a) in more detail. that firms fail for lack of profits, but we Can strong rationality postulates really seldom read in the obituary pages that take major credit for the successes of people die of suboptimization. Consum- economics, or do the successes originate ers who display wasteful shopping pat- in ideas consistent with much broader terns can survive at a lower standard of notions of rationality? Empirical practice living, and workers who use their talents tends to neglect this question. Instead of wastefully can survive at a lower wage. testing the predicted effects of optimiza- There is a more subtle survival argu- tion against the predicted effects of com- ment for individuals. Rules of thumb are peting theories, we tend to test against typically exploitable by "tricksters,"who the nonsubstantive null hypothesis of no can in principle "money pump" a person effect. This is somewhat like arm wres- Conlisk: Why Bounded Rationality? 685 tling a rag doll; it doesn't prove any- hidden in one of hundreds of cracks in thing-unless the rag doll wins. the sidewalk and that, to know which As an example of his view, Becker crack, the walker must have reasoned (1976, p. 10) credits a utility-maximiza- through a complex pattern of logical tion model of Michael Grossman (1975) clues. In the face of such deliberation with predicting various correlations cost, the walker may walk on by. Simi- among health and economic variables for larly, deliberation cost may make rational individuals. However, Grossman himself expectations cost more than they are notes (p. 148) that variants on his model worth. Unboundedly rational optima, by "can be used to rationalize any observed neglecting deliberation cost, may iden- correlation between two variables." Did tify false opportunities for gain. utility maximization suggest the patterns Argument 6. Discipline and "ad ho- Grossman found, or did it merely pack- cery." Without the discipline of optimiz- age them? Whatever the truth about the ing models, economic theory would de- particular case, economic research often generate into a hodge podge of ad hoc seems to work backwards from empirical hypotheses which cover every fact but findings to whatever utility maximization which lack overall cohesion and scientific will work. Where the empirical arrow refutability. falls, there we paint the utility bullseye. Comment. Discipline comes from good Putting the issue a bit differently, Ar- scientific practice, not embrace of a par- thur Goldberger (1989), Simon (1986), ticular approach. Any approach, includ- and Arrow (1986) note that utility maxi- ing the optimization approach, can lead mization has little empirical content to an undisciplined proliferation of hy- without strong auxiliary assumptions on potheses to cover all facts. Conversely, a the utility functions and other model in- bounded rationality hypothesis might gredients. Because a trained economist produce a parsimonious explanation of a can see through a utility maximization, variety of empirical patterns. For exam- stating auxiliary assumptions is often lit- ple, Shefrin and Statman (1994) use tle different from stating empirical pre- their behavioral capital asset pricing dictions outright, as, say, a sociologist model to address various financial might. In this sense, the utility maximi- anomalies as a group. A merit of the de- zation merely packages the prediction. liberation cost idea is that it suggests a Argument 5. Sidewalk twenties. A discipline for models of bounded ratio- model of unbounded rationality identi- nality-that departures from unbounded fies an agent's best opportunity for gain. rationality be systematically related to Because it is implausible for an agent to the deliberation cost involved. forgo opportunities for gain, unbounded Argument 7. Tractability and definite rationality identifies the agent's likely ac- outcomes. The unbounded rationality tion. postulate, because it can be formulated Comment. Forgoing an opportunity for through well understood mathematical gain, it is claimed, is like failing to pick optimizations, confers tractable analysis up a $20 bill lying on the sidewalk. In and definite outcomes. the rational expectations literature, the Comment. Consider tractability. Be- sidewalk twenties argument appears as cause optimizations may be arbitrarily the claim that an agent with suboptimal complex, whereas bounded rationality expectations would be "consistently may be represented by simple rules of fooled" into forgoing opportunities for thumb, optimization-based models are gain. However, suppose that the $20 is sometimes more and sometimes less 686 Journal of Economic Literature, Vol. XXXIV (June 1996) tractable than adaptation-based models. plines are in fact clusters of activity, not A spectacular example of the latter is re- provinces protected by border guards. cent macrotheory. By insisting on ra- Whenever theory and evidence suggest a tional expectations and intertemporal op- need to settle the sparsely populated ar- timization, which are quite intractable in eas between clusters, science says wel- general settings, macrotheory is often re- come. duced to considering only a single "rep- In summary, the standard arguments resentative" agent. We model Robinson for unbounded rationality, despite their Crusoe and pretend he's a $7 trillion great influence, are too extreme to be economy. Arrow (1986), James Tobin convincing. Put in more flexible form, (1989), Robert Solow (1989), and Kir- however, the arguments contain many man (1992) note the strange sacrifices useful insights about conditions favoring required for the "ritualpurity" of optimi- one or another treatment of rationality. zation-only models (Akerlof and Yellen's Fortunately, economists are coming to phrase, 1987, p, 137). adopt more flexible interpretations. Consider definite outcomes. For an Even Becker, perhaps pushed by his agent with a well behaved objective long-standing interest in nonstandard function, it is argued that an optimiza- costs, has recently opened the door a tion gives one model and one outcome, crack for deliberation cost. In his Nobel whereas adaptation, which may take dif- lecture (1993), Becker says: "Actions are ferent forms, may give many models and constrained by income, time, imperfect many outcomes. The main response is memory and calculating capacities, and that, even if we insist on looking at only other limited resources" (p. 386, empha- one model, evidence and plausibility sis added); and he concludes, "My work should be the criteria, not prior bias to- may have sometimes assumed too much ward optimizations. In any case, the one rationality"(p. 402). optimization model may generate multi- ple equilibria and thus multiple out- IV. No Free Lunch, Yes Bounded comes, whereas the adaptive models may Rationality all converge to the same one of the mul- tiple equilibria and thus generate a sin- It is evident that the rational thing to do is to gle outcome. This equilibrium-selection be irrational, where deliberation and estima- issue motivates a number of the adaptive tion cost more than they are worth. models cited in Section II. Frank Knight (1921, p. 67, footnote) Argument 8. Definition. Economicsis Human cognition is a scarce resource, by definition the study of optimizing be- implying that deliberation about eco- havior; bounded rationality is the prov- nomic decisions is a costly activity. To ince of other disciplines. avoid a free lunch fallacy, it can be ar- Comment. By its most common defini- gued, we are forced to incorporate delib- tion, economics concerns scarcity. Be- eration cost, and thus bounded rational- cause human reasoning ability is scarce, ity, in economic models. There are one could as well argue that economists special problems. are by definition required to study bounded rationality. More important, A. Economizing Economizing: economics as a science must view every The-Regress Issue theory, including optimization theory, as open to empirical challenge. Regarding Unbounded rationality is typically for- the province metaphor, scientific disci- mulated as the assumption that a Conlisk: Why Bounded Rationality? 687

decision maker optimizes an objective The question of how far to go . . . is in itself function subject to cost and other con- an optimization problem, but a peculiar one in that it can itself not be subjected to analy- straints. Because it is a routine exercise sis . . . at least in the last instance. Should to include one more cost in an optimiza- one try to analyse the question of how to tion model, a treatment of deliberation strike an optimal balance , . . . then the same cost seems straightforward at first question could be raised in relation to this glance. Simply include that extra cost. question, and so on. At some point a decision must be taken on intuitive grounds. (Leif Jo- However, we quickly collide with a per- hansen 1977, p. 144) plexing obstacle. Suppose that we first formulate a deci- Other early mentions of the regress issue sion problem as a conventional optimiza- are in Howard Raiffa (1968, p. 266), tion based on the assumption of un- Radner (1968, p. 56, and 1975, p. 266), bounded rationality and thus on the Marschak and Radner (1972, sections assumption of zero deliberation cost. 9.6-9.7), and Hans Gottinger (1982). Suppose we then recognize that delib- Perhaps the most succinct summary of eration cost is positive; so we fold this the issue is Day and Pingle's phrase further cost into the original problem. "economizing economizing" (1991, p. The difficulty is that the augmented op- 509). If we can economize on economiz- timization problem will itself be costly to ing, then we can economize on econo- analyze; and this new deliberation cost mizing on economizing, and so on. Given will be neglected. We can then formu- the vast number of expositions of choice late a third problem which includes the theory, it is remarkable how infrequently cost of solving the second, and then a the regress issue is mentioned. I have fourth problem, and so on. We quickly found only three papers-Philippe find ourselves in an infinite and seem- Mongin and Bernard Walliser (1988), ingly intractable regress. In rough nota- Holly Smith (1991), and Barton Lipman tion, let P denote the initial problem, (1991)-which discuss the regress issue and let F(.) denote the operation of fold- in any detail. ing deliberation cost into a problem. The regress problem seems to block Then the regress of problems is P, F(P), any effort to maintain optimization as F2(P), ... the ultimate logical basis for all behav- There are two difficult issues here: (i) ioral modeling. How can we formulate what the operator F looks like and (ii) an optimization problem which takes full how to deal with the regress. Start with account of the cost of its own solution? (ii). Few authors mention the regress is- There is no reason to suppose that se- sue, and most mentions are little more. quences like P, F(P), F2(P), . . . will often Examples: converge (though Lipman 1991 discusses convergence of a related sequence) or, if It might ... be stimulating, and it is certainly convergence occurs, that the limit corre- more realistic, to think of consideration or sponds to any problem descriptive of a calculation as itself an act on which the person must decide. Though I have not ex- decision maker. We seemingly must plored the latter possibility carefully, I sus- yield to the idea that some behavioral hy- ect that any attempt to do so leads to fruit- pothesis other than optimization, such as I?essand endless regression. (Savage 1954, p. 30). learning or adaptation, is needed to es- cape the regress. In Johansen's words, an optimization whose scope covers all con- "At some point a decision must be taken siderations including its own costs . . . sounds like it may involve the logical difficulties of on intuitive grounds." self-reference. (Winter 1975, p. 83) In practical modeling, then, what 688 Journal of Economic Literature, Vol. XXXIV (June 1996)

should economists do about the regress senting the unpredictability of delibera- of problems P, F(P), F2(P), . . . ? It tion (else the agent would know the an- seems sensible to focus on only the swer to begin with). A deliberation tech- first two problems, P and F(P). Problem nology might then be specified as a P asks what the perfect decision is, function and problem F(P) asks in addition how much costly deliberation the decision X(T) = G(T, X*, Xo, u). (1) maker should expend in approximating the perfect decision. These are sensible It would be naturalto give G(T, X*, Xo, u) behavioral questions. Problem F2(P) properties such that X(T) moves sto- asks in addition how much delibera- chastically from Xo toward X* as T in- tion the decision maker should expend creases from 0 to oo. The formal assump- deciding how much deliberation to ex- tions might be G(A, X*, Xo, u) = Xo, pend approximating a perfect decision. (a / a T)E[G(T, X*, Xo, u) - X*]2 < 0, This problem seems overly convoluted, and G( o0 , X*, Xo, u) = X*. and F3(P), F4(P), . . . are more so. Consider the intuition of (1) under Although the regress as a whole is these assumptions. At one extreme, if de- worthwhile to notice, because it helps liberation is prohibitively costly (C very to put issues in perspective, practical large), the decision maker is motivated modeling might, at least initially, neglect to do no deliberation (T = 0), and (1) all problems beyond P and F(P). In yields the pure rule-of-thumb decision any case, that is what economists have X = Xo, perhaps a simple adaptive re- done. sponse to experience. At the opposite ex- treme, if deliberation is free (C = 0), B. An P and Example of F(P) the decision maker is motivated to do in- Consider a decision maker choosing a finite deliberation (T = oo), and (1) decision variable X (scalar or vector) to yields the unboundedly rational choice make a payoff function H (X) large. Let X = X*. In between, (1) gives a mix 11(X)have unique optimizer X*. Suppose among rule-of-thumb behavior, delibera- that the decision maker has enough in- tion, and random noise. The mix dictates formation in principle to compute the the decision maker's "degree of rational- value of 11(X)for any X and thus to find ity" for the problem at hand. Algebrai- X*. Then X = X* is the unboundedly ra- cally specific deliberation technologies tional choice. However, suppose that of form (1) are used in Conlisk (1988, Hl(X) is so complex a function that the forthcoming) and Evans and Ramey deliberations in finding X = X* would (1992). be prohibitively costly. Thus, consider a To the possible criticism that (1) deliberation technology by which the doesn't look much like human cognition, decision maker "produces" an approxi- we might note that a CES production mation X to the perfect decision X*. function doesn't look much like a factory Let T be the costly effort devoted to floor. In representing a deliberation approximatingX*, where C is the cost of technology as in representing a produc- one unit of T. Let X(T) be the actual tion technology, the object is not faith- decision resulting from this costly fulness to cognitive science or to engi- deliberation, and let Xo be a rule- neering. Rather the object is a simple of-thumb decision that the agent could relationship for representing economic use for free (zero deliberation). Finally, tradeoffs. let u be a random disturbance repre- In this example, the original problem Conlisk: Why Bounded Rationality? 689

P is to choose X to make Hl(X) large. deliberation cost. This category includes Adding risk neutrality, the augmented the vast majority of models of bounded problem F(P) may be defined as the rationality surveyed in Section II. Cate- problem of choosing the deliberation gories 3 and 4, which require specifica- effort T to make the expected net pay- tion of a deliberation technology, contain off E{Hn[X(T)]} - CT large. Summariz- only the very few models surveyed in the ing: final three paragraphsof Section II. Original problem P. Choose X to make Consider Category 3. It supposes that we 'have specified a deliberation technol- Hl(X)large. ogy and that the decision maker chooses Augmented problem F(P). Choose T to the optimal amount of deliberation. In make E{F1[X(T)]} - CT large. the example, the decision maker chooses the T, call it T*, which maximizes C. Four Rationalities E{n[X(T)]} - CT. Thirty years ago, Wil- The problems P and F(P) suggest a liam Baumol and Richard Quandt (1964, rough way of categorizing treatments of p. 23) dubbed this "optimal imperfec- rationality in the literature. Most models tion." In their words, treat the original decision problem P. One can easily formulatethe appropriate... Only a few add a deliberation technology marginalconditions for what one maycall an and treat the augmented problem F(P). optimallyimperfect decision, which requires Among models treating P, there are two that the marginalcost of . . . more refined calculation be equal to its marginal (ex- ways to close the model. Either the deci- pected) grossyield. sion maker optimizes, or the decision maker uses some other behavioral rule, We might quarrel with the words "easily call it an adaptive rule. Among models formulate," because Baumol and Quandt treating F(P), there are the same two did not in fact present a model of opti- ways to close the model. Either the deci- mal imperfection, nor have many authors sion maker optimizes in the sense of since. In terms of the F(P) example, the finding the optimal deliberation effort to marginal condition referred to by Bau- devote to the choice, or the decision mol and Quandt is that the marginal cost maker follows some adaptive rule in of deliberation C equal the expected choosing deliberation effort. This gives marginal benefit aE{fn[X(T)]}/aT. If F(P) four categories of models. is viewed as a stopping problem (when to stop deliberating and take final action), 1. Treat problem P. Optimal closure. then optimal imperfection means opti- 2. Treat problem P. Adaptive closure. mal stopping. However, there is a prob- 3. Treat problem F(P). Optimal clo- lem. Why would a decision maker who sure. cannot optimize relative to problem P be 4. Treat problem F(P). Adaptive clo- able to optimize relative to problem sure. F(P), which will often be more compli- The categories are in decreasing order of cated? Yet, if we fold in the cost of delib- size. Category 1 comprises models of un- erating about F(P), we are off again into boundedly rational choice, the vast ma- the regress P, F(P), F2(P), . . . Optimal jority of models in the literature. Cate- imperfection returns us to the regress. gory 2 includes models of bounded Nonetheless, in the literature, the few rationality in which adaptive choice rules models which treat problem F(P) often are specified outright, with no delibera- do invoke optimal imperfection. What is tion technology or explicit treatment of the defense? Taking a dynamic view, we 690 Journal of Economic Literature, Vol. XXXIV (June 1996)

might justify optimal imperfection as an D. Ex Ante vs Ex Post Posts: equilibrium condition. A model in Cate- Similarities of Deliberation Cost gory 4 might adapt over time into a and Information Cost model in Category 3, just as, by more fa- miliar adaptive logic, a model in Cate- When I walked into a post while gory 2 might adapt over time into Cate- watching a bird, my family called it a gory 1. However, the conditions for such dumb move. Among economists, how- convergence from Category 4 to Cate- ever, I could have claimed that, given gory 3 seem delicate. We must suppose the spatial distribution of lamp posts, the that the decision maker faces a time se- expected utility of bird watching ex- quence of original problems {Pt} suffi- ceeded the expected disutility of a colli- ciently complex that deliberation cost re- sion. Ex ante, the post probably was not mains important, thus leading to a there, and it is entirely rational to collide sequence {Ft(Pt)} of deliberation cost with an ex post post. This example illus- problems. We must then assume that the trates the confounding of rationality is- optimal deliberation effort is the same sues with information issues. Am I dumb for each problem in the sequence to walk into a post or merely a rational {Ft(Pt)),so that there is an invariant opti- victim of imperfect information? mal effort T* to which the actual efforts Expanding the deliberation technology {Tt} might in principle converge. Finally, idea, it is natural to view decisions as we must assume that the decision maker "produced" by a decision technology does manage to converge. with two inputs, costly information-gath- Because these conditions are delicate, ering and costly deliberation. The simi- a modeler may have to justify optimal larity of information-gathering and delib- imperfection-Category 3-as nothing eration, as joint inputs in producing a more profound than a compromise of ex- decision, suggests that models of delib- pedience. Category 3, by taking direct eration, as they evolve in economics, will and explicit account of deliberation cost inevitably have a general resemblance to and the tradeoffs it implies, is already a existing models of information collec- big improvement over Categories 1 and tion. For example, the illustrative delib- 2, which comprise most of the existing eration technology above resembles literature. some sampling models, with T the analog Though optimal imperfection closes a of a sample size. model with an optimization, it is not a It is curious that such similar eco- retreat to some new form of unbounded nomic issues, costly deliberation and rationality. An unboundedly rational de- costly information collection, have been cision maker optimizes every setting; treated so differently in standard eco- whereas an optimally imperfect agent nomics, one avoided and the other em- does not. In the example above, an un- braced. In practice, the difference in boundedly rational decision maker hits treatment has required that anything re- both settings X = X* and T = T* = sembling imperfect deliberation be (where T* = oo because deliberation is passed off instead as imperfect informa- free), whereas an optimally imperfect tion. decision maker hits only T = T*. This For example, Williamson (1985, 1986) difference is large. In the example, an is a towering figure in industrial organi- optimally imperfect X mixes rule-of- zation for his insights about transactions thumb behavior, deliberation, and ran- cost. Although he sees these costs as dom noise. rooted in bounded rationality, formal Conlisk: Why Bounded Rationality? 691 theories based on his ideas tend to por- proper technique by dissertation supervi- tray the costs as information costs. An- sors and journal referees. No one other example is the famous Gang of claimed that information was literally Four explanation of why we observe co- perfect in real life, merely that agents operation in finitely repeated prisoner's learned their own situations well enough dilemma games even though the familiar to act "as if' perfectly informed; after all, unraveling argument of game theory pre- those who didn't would be driven out of dicts failure to cooperate. Although the business by those who did. observed behavior appears to be bound- E. Elephants in the Living Room edly rational, Kreps et al. (1982) suggest a possible rescue of standard theory by Deliberation cost and bounded ratio- putting the bound on information in- nality, like elephants in a living room, stead. They assume that, although both are sometimes just too much to ignore. players really are unboundedly rational, Standard economics is forced to recog- one player thinks the other might be nize their presence, if not to refer to boundedly rational. This clever (and them by name. Consider two examples, strained?) informational twist is enough human capital and technical change. to induce cooperation within the usual People spend much on human capital, rationality assumptions. The Gang of in large part through schooling. The in- Four approach is in sharp contrast to vestment is partly information collection Selten's (1978) approach to the chain (names and dates), partly skill acquisi- store game, another game in which the tion (typing), and partly general cogni- unraveling logic produces a counterintui- tive investment ("learning to think"). tive prediction. Selten faces the bounded The cognitive investment must be a 're- rationality issue directly and sketches a sponse to bounded rationality. Consider theory of bounded rationality, including a deliberation cost interpretation. Delib- a brief discussion of deliberation cost eration cost can be specific to a particu- and, implicitly, of the regress issue. See lar decision, as in the F(P) illustration also Selten and Rolf Stoecker (1986) and above; or it can be the general cost of Selten (1991, especially p. 18) on coop- all-purpose cognitive training used in eration, unravelling, and the Gang of many decisions over many years. The Four. part of schooling cost which goes into To gain perspective, it is entertaining general cognitive development is general to imagine an accidentally different his- deliberation cost, and human capital the- tory for economic theory. Imagine that ory is implicitly concerned with bounded modern decision theory began, not with rationality. The assumption that students perfect rationality and imperfect infor- invest optimally in schooling is an unusu- mation, but with the opposite. Observed ally strong example of optimal imperfec- behavior that seemed to be the result of tion. Explicit recognition of the relation imperfect information was instead of human capital theory to bounded ra- passed off by clever economists as the tionality might bring new insights to the result of bounded rationality. As the idea theory. caught on, strict conventions for proper As a second example, consider techni- treatment of bounded rationality devel- cal change. Many technological innova- oped. Scholars departing from the con- tions result from insights that would ventions, or even worse from the perfect have been made years earlier if people information postulate, were chastised as really could draw all possible inferences "ad hoc" and were firmly guided back to from existing information. In this sense, 692 Journal of Economic Literature, Vol. XXXIV (June 1996)

the rate of technical change is deter- cases. As with other model ingredients, mined largely by bounds on rationality however, we in practice want to work di- and by the resulting delays in exploiting rectly with the most convenient special economic opportunities. Yet, according case which does justice to the context. to various models of research and devel- The evidence and models surveyed sug- opment, decision makers engage in opti- gest that a sensible rationality assump- mal amounts of search for the unex- tion will vary by context, depending on ploited opportunities, as if unboundedly such conditions as deliberation cost, rational on that dimension. We can view complexity, incentives, experience, and the search cost as (in part) deliberation market discipline. cost, and we can view the optimal search Beyond the four reasons given, there assumption as an example of optimal im- is one more reason for studying bounded perfection. If the relation of technical rationality. It is simply a fascinating change to bounded rationality were rec- thing to do. We can mix some Puck with ognized openly (as in the evolutionary our Hamlet.

models surveyed in Nelson 1995), stan- REFERENCES dard models of technical change might ABREU, DILIP AND RUBINSTEIN, ARIEL. "The be better. Structure of Nash Equilibrium in Repeated Games with Finite Automata," Econometrica, Nov. 1988, 56(6), pp. 1259-81. V. Final Words AKERLOF, GEORGE A. An economic theorist's book of tales. Cambridge: Cambridge U. Press, Why bounded rationality? In four 1984. words (one for each section above): evi- . "Procrastination and Obedience," Amer. dence, success, methodology, and scar- Econ. Rev., May 1991, 81(2), pp. 1-19. AKERLOF, GEORGE A. AND DICKENS, WILLIAM city. In more words: Psychology and eco- T. "The Economic Consequences of Cognitive nomics provide wide-ranging evidence Dissonance," Amer. Econ. Rev., June 1982, that bounded rationality is important 72(3), pp. 307-19; reprinted in AKERLOF 1984, pp. 123-44. (Section I). Economists who include AKERLOF, GEORGE A. AND YELLEN, JANET L. bounds on rationality in their models "Can Small Deviations from Rationality Make have excellent success in describing eco- Significant Differences to Economic Equi- libria?"Amer. Econ. Rev., Sept. 1985, 75(4), pp. nomic behavior beyond the coverage of 708-20. standard theory (Section II). The tradi- . "Rational Models of Irrational Behavior," tional appeals to economic methodology Amer. Econ. Rev., May 1987, 77(2), pp. 137-42. ALCHIAN, ARMEN A. "Uncertainty, Evolution, and cut both ways; the conditions of a par- Economic Theory," J. Polit. Econ., June 1950, ticular context may favor either bounded 58(3), pp. 211-21. or unbounded rationality (Section III). ARKES, HAL R. AND HAMMOND, KENNETH R., eds. Judgment and decision making: An inter- Models of bounded rationality adhere to disciplinary reader. Cambridge: Cambridge U. a fundamental tenet of economics, re- Press, 1986. spect for scarcity. Human cognition, as a ARROW, KENNETH J. "Rationalityof Self and Oth- ers in an Economic System,"J. Bus., Oct. 1986, scarce resource, should be treated as 59(4, pt. 2), pp. S385-99. such (Section IV). ARTHUR, W. BRIAN. "Competing Technologies, The survey stresses throughout that an Increasing Returns, and Lock-In by Historical is not Events," Econ. J., Mar. 1989, 99(394), pp. 116- appropriate rationality assumption 31. something to decide once for all con- . "On Designing Economic Agents that Be- texts. In principle, we might suppose have Like Human Agents," J. Evolutionary there is an encompassing single theory Econ., Feb. 1993, 3(1), pp. 1-22. . "Inductive Reasoning and Bounded Ra- which takes various forms of bounded tionality," Amer. Econ. Rev., May 1994, 84(2), and unbounded rationality as special pp. 406-11. Conlisk: Why Bounded Rationality? 693

ASHENFELTER, ORLEY AND GENESOVE, DAVID. CAMERER, COLIN F. "Bubbles and Fads in Asset "Testing for Price Anomalies in Real-Estate Prices,"J. Econ. Surveys, 1989, 3(1), pp. 3-41. Auctions," Amer. Econ. Rev., May 1992, 82(2), . "Behavioral Game Theory," in Insights in pp. 501-05. decision making: A tribute to Hillel J. Einhorn. BANERJEE, ABHIJIT V. "A Simple Model of Herd Ed.: ROBIN M. HOGARTH. Chicago: U. of Chi- Behavior," Quart. J. Econ., Aug. 1992, 107(3), cago Press, 1990, pp. 311-36. pp. 797-817. . "Recent Tests of Generalized Utility . "The Economics of Rumours," Rev. Econ. Theories," in Utility theories: Measurements Stud., Apr. 1993, 60(2), pp. 309-27. and applications. Ed.: WARD EDWARDS. Dor- BARTHOLOMEW, DAVID J. Stochastic models for drecht: Kluwer Academic, 1992, pp. 207-51. social processes. 3rd ed. New York:Wiley, 1982. CAMERER, COLIN F. ET AL. "Cognition and Fram- BATTALIO, RAYMOND C.; KAGEL, JOHN H. AND ing in Sequential Bargaining for Gains and JIRANYAKUL, KOMAIN. "Testing between Alter- Losses," in Frontiers of game theory. Eds.: KEN native Models of Choice under Uncertainty: BINMORE, ALAN KIRMAN, AND PIERO TANI. Some Initial Results,"J. Risk Uncertainty, Mar. Cambridge: MIT Press, 1993, pp. 27-47. 1990, 3(1), pp. 25-50. CARROLL, CHRISTOPHER D. "How Does Future BAUMOL, WILLIAM J. AND QUANDT, RICHARD E. Income Affect Current Consumption?" Quart. "Rules of Thumb and Optimally Imperfect De- J. Econ., Feb. 1994, 109(1), pp. 111-47. cisions," Amer. Econ. Rev., Mar. 1964, 54(1), CASON, TIMOTHYN. "Call Market Efficiency with pp. 23-46. Simple Adaptive Learning,"Econ. Letters, Sept. BECKER, GARY S. The economic approach to hu- 1992, 40(1), pp. 27-32. man behavior. Chicago: U. of Chicago Press, CHANDLER, ALFRED D. Strategy and structure. 1976. Cambridge: MIT Press, 1962. . "Nobel Lecture: The Economic Way of COASE, RONALD H. "The Nature of the Firm," Looking at Behavior,"J. Polit. Econ., June 1993, Economica, Nov. 1937, 4(16), pp. 386-405. 101(3), pp. 385-409. . "The Institutional Structure of Produc- BENARTZI, SHLOMO AND THALER, RICHARD H. tion," Amer. Econ. Rev., Sept. 1992, 82(4), pp. "Myopic Loss Aversion and the Equity Pre- 713-19. mium Puzzle," Quart. J. Econ., Feb. 1995, COCHRANE, JOHN H. "The Sensitivity of Tests of 110(1), pp. 73-92. the Intertemporal Allocation of Consumption to BENNETT, ROBERT L. AND BERGMANN, BARBARA Near-Rational Alternatives," Amer. Econ. Rev., R. Microsimulated transactions model of the June 1989, 79(3), pp. 319-37. United States economy. Baltimore: Johns Hop- CONLISK, JOHN. "Costly Optimizers versus Cheap kins U. Press, 1986. Imitators,"J. Econ. Behav. Organ., Sept. 1980, BIKHCHANDANI, SUSHIL; HIRSHLEIFER, DAVID 1(3), pp. 275-93. AND WELCH, IVO. "A Theory of Fads, Fashion, . "Competitive Approximationof a Cournot Custom, and Cultural Change as Informational Market,"Rev. Econ. Stud., Oct. 1983, 50(4), pp. Cascades," J. Polit. Econ., Oct. 1992, 100(5), 597-607. pp. 992-1026. . "Optimization Cost," J. Econ. Behav. Or- BLUME, LAWRENCE AND EASLEY, DAVID. gan., Apr. 1988, 9(3), pp. 213-28. "Learning to Be Rational," J. Econ. Theory, __ "'An Aggregate Model of Technical Aug. 1982, 27(2), pp. 340-51. Change," Quart. J. Econ., Nov. 1989, 104(4), . "Evolution and Market Behavior," J. Econ. pp. 787-21. Theory, Oct. 1992, 58(1), pp. 9-40. _ "The Utility of Gambling," J. Risk Uncer- BOYD, ROBERT AND RICHERSON, PETER J. Cul- tainty, June 1993a, 6(3), pp 255-75. ture and the evolutionary process. Chicago: U. "AdaptiveFirms and Random Innovations of Chicago Press, 1985. in a Model of Cyclical Output Growth,"in Non- BRAY, MARGARET M. "Learning, Estimation and linear dynamics and evolutionary economics. the Stability of Rational Expectations," J. Econ. Eds.: RICHARD H. DAY AND PING CHEN. Ox- Theory, Apr. 1982, 26(2), pp. 318-39. ford: Oxford U. Press, 1993b, pp. 113-30. BRAY, MARGARET AND KREPS, DAVID M. "Ra- . "Bounded Rationality and Market Fluc- tional Learning and Rational Expectations," in tuations,"J. Econ. Behav. Organ., forthcoming. Arrow and the ascent of modern economic the- COX, JAMES C.; DINKIN, SAMUEL H. AND SMITH, ory. Vol. 1. Ed.: GEORGE R. FEIWEL. New VERNONL. "Endogenous Entry and Exit in York: New York U. Press, 1987, pp. 597-625. Common Value Auctions." Manuscript. U. of BREHMER, BERNDT. "In One Word: Not from Ex- Arizona, 1995. perience," Acta Psychologica, Aug. 1980, 45(1- CRAGG, JOHN G. AND MALKIEL, BURTON G. Ex- 3), pp. 223-41; reprinted in HAL R. ARKES AND pectations and the structure of share prices. KENNETH R. HAMMOND 1986, pp. 705-19. Chicago: U. of Chicago Press, 1982. BROCK, WILLIAM A. AND HOMMES, CARS H. "Ra- CRAWFORD,VINCENT P., ed. Adaptive Dynamics. tional Routes to Randomness." Economics Parts I and II. Special issues of Games Econ. Working Paper 95-03-029. Santa Fe Institute, Behav., July-Oct. 1993, 5(3-4). 1995. . "Adaptive Dynamics in Coordination 694 Journal of Economic Literature, Vol. XXXIV (June 1996)

Games," Econometrica, Jan. 1995, 63(1), pp. ARD H. "Does the Stock Market Overreact?"J. 103-43. Finance, July 1985, 40(3), pp. 793-805. CREEDY, JOHN; BORLAND, JEFF AND EICHBER- . "Do Security Analysts Overreact?"Amer. GER, JURGEN, eds. Recent developments in Econ. Rev., May 1990, 80(2), pp. 52-57. game theory. Aldershot, England: Edward El- . "Financial Decision-Making in Markets gar, 1992. and Firms: A Behavioral Perspective." Working JAMES M. AND CUTLER, DAVID M.; POTERBA, Daper 4777. NBER, 1994. SUMMERS, LAWRENCE H. "Speculative Dynam- DK LONG, J. BRADFORD ET AL. "Noise Trader ics," Rev. Econ. Stud., May 1991, 58(3), pp. Risk in Financial Markets,"J. Polit. Econ., Aug. 529-46. 1990, 98(4), pp. 703-38. CROSS, JOHN G. "A Stochastic Learning Model of DETEMPLE, JEROME AND MURTHY, SHASHID- Economic Behavior," Quart. J. Econ., May HAR."Intertemp oral Asset Pricing with Hetero- 1973, 87(2), pp. 239-66. geneous Beliefs," J. Econ. Theory, Apr.1994, . A theory of adaptive economic behavior. 62(2), pp. 294-320. Cambridge: Cambridge U. Press, 1983. ELIASSON, GUNNAR. "Modeling the Experimen- CYERT, RICHARD M. AND MARCH, JAMES G. A tally Organized Economy: Complex Dynamics behavioral theory of the firm. Englewood Cliffs, in an Empirical Micro-Macro Model of En- NJ: Prentice-Hall, 1963. dogenous Economic Growth," J. Econ. Behav. DAVID, PAUL A. "Clio and the Economics of Organ., July 1991, 16(1-2), pp. 153-82. QWERTY,"Amer. Econ. Rev., May 1985, 75(2), ELLISON, GLENN AND FUDENBERG, DREW. pp. 332-37. "Rules of Thumb for Social Learning,"J. Polit. . "Understanding the Economics of Econ., Aug. 1993,101(4), pp. 612-43. QWERTY: the Necessity of History," in Eco- . "Word-of-Mouth Communication and So- nomic history and the modern economist. Ed.: cial Learning" Quart. J. Econ., Feb. 1995, WILLIAM N. PARKER. Oxford: Blackwell, 1986, 110(1), pp.93'125. pp. 30-49. ELSTER, JON. Ulysses and the sirens: Studies in DAWES, ROBYN M. AND THALER, RICHARD H. rationality an irrationality. Cambridge: Cam- "Anomalies: Cooperation," J. Econ. Perspec- bridge U. Press, 1979. tives, Summer 1988, 2(3), pp. 187-97. . Sour grapes: Studies in the subversion of DAY, RICHARD H. Recursive programming and rationality. Cambridge: Cambridge U. Press, production response. Amsterdam: North-Hol- 1983. land, 1963. "Social Norms and Economic Theory," J. ed. Mathematical Dynamics and Eco- Econ. Perspectives, Fall 1989, 3(4), pp. 99-117. nomic Process. Special issue of J. Econ. Behav. ERMINI, LUIGI."Hierarchical Decomposition in Organ., July 1991, 16(1-2). Economic Analysis." Ph.D. dissertation. U. of "Bounded Rationality and the Coevolution California, San Diego, 1987. of Market and State," in The marketsfor inno- . "HierarchicalDecomposition in Economic vation, ownership and control. Eds.: RICHARD Analysis,"in Behavioral decision making: Hand- H. DAY, GUNNAR ELIASSON, AND CLAS book of . Vol. 2B. Eds.: WIHLBORG. Amsterdam: North-Holland, 1993a. ROGER FRANTZ, HARINDER SINGH, AND ed. Adaptive Games. Special issue of J. JAMES GERBER. Greenwich, CT: JAI Press, Econ. Behav. Organ., Sept. 1993b, 22(1). 1991, pp. 547-62. Complex economic dynamics. Cambridge: EVANS, GEORGE W. AND HONKAPOHJA,SEPPO. MIT Press, 1994. "Adaptive Learning and Expectational Stability: DAY, RICHARD H. AND CIGNO, ALESSANDRO. an Introduction," in Learning and Rationality Modelling economic change: The recursive pro- in Economics. Eds.: ALAN KIRMAN AND MARK ramming approach. Amsterdam: North-Hol- SALMON. Oxford: Basil Blackwell, 1995. land, 1978. EVANS, GEORGE W. AND RAMEY,GAREY. "Expec- DAY, RICHARD H. AND PINGLE, MARK A. tation Calculation and Macroeconomic Dynam- "Economizing Economizing," in Behavioral de- ics," Amer. Econ. Rev., Mar. 1992, 82(3), pp. cision making: Handbook of behavioral econom- 207-24. ics. Vol. 2B. Eds.: ROGER FRANTZ, HARINDER . "Calculation,Adaptation and Rational Ex- SINGH, AND JAMES GERBER. Greenwich, CT: pectations." Manuscript. U. of California, San JAI Press, 1991, pp. 509-22. - Diego, 1994. "If People Optimize, How Do They Do ."Expectation Calculation, Hyperinflation It?"J. Econ. Behav. Organ., forthcoming. and Currency Collapse," in The new macro- DAY, RICHARD H. AND TINNEY, E. HERBERT. economics: Imperfect markets and policy effec- "How to Co-operate in Business without Really tiveness. Eds.: Huw DAVID DIXON AND NEIL Trying: A Learning Model of Decentralized De- RANKIN. Cambridge: Cambridge U. Press, cision Making," J. Polit. Econ., July-Aug. 1968, 1995, pp. 307-36. 76(4), pp. 583-600. FAMA, EUGENE F. AND FRENCH, KENNETH R. DEATON, ANGUS. Understanding consumption. "Permanent and Temporary Components of Oxford: Clarendon Press, 1992. Stock Prices," J. Polit. Econ., Apr. 1988, 96(2), DE BONDT, WERNER F. M. AND THALER, RICH- pp. 246-73. Conlisk: Why Bounded Rationality? 695

FARRELL, MICHAEL J. "Some Elementary Selec- for Individual Rationality,"J. Polit. Econ., Feb. tion Processes in Economics," Rev. Econ. Stud., 1993, 101(1), pp. 119-37. July 1970, 32(3), pp. 305-19. GOLDBERGER, ARTHUR S. "Economic and Me- FISCHHOFF, BARUCH. "Debiasing," in DANIEL chanical Models of Intergenerational Transmis- KAHNEMAN, PAUL SLOVIC, AND AMOS TVER- sion," Amer. Econ. Rev., June 1989, 79(3), pp. SKY 1982, pp. 422-44. 504-13. FLAVIN, MARJORIE A. "The Adjustment of Con- GOTTINGER, HANS W. "ComputationalCosts and sumption to Changing Expectations about Fu- Bounded Rationality," in Philosophy of eco- ture Income," J. Polit. Econ., Oct. 1981, 89(5), nomics. Eds.: WOLFGANG STEGMULLER, pp. 974-1009. WOLFGANG BALZER, AND WOLFGANG SPOHN. . "The Excess Smoothness of Consumption: Berlin: Springer-Verlag,1982, pp. 223-38. Identification and Interpretation," Rev. Econ. GRANOVETTER, MARK AND SOONG, ROLAND. Stud., July 1993, 60(3), pp. 651-66. "Threshold Models of Interpersonal Effects in FRANCIOSI, ROBERT ET AL. "Experimental Tests Consumer Demand," J. Econ. Behav. Organ., of the Endowment Effect." Manuscript. Eco- Mar. 1986, 7(1), pp. 83-99. nomic Science Laboratory, U. of Arizona, 1995. GRETHER, DAVID M. "Testing Bayes Rule and the FRANK, ROBERT H. Choosing the right pond. Ox- Representativeness Heuristic: Some Experi- ford: Oxford U. Press, 1985. mental Evidence," J. Econ. Behav. Organ., Jan. . Passions within reasons. New York: Nor- 1992, 17(1), pp. 31-57. ton, 1988. GRETHER, DAVID M. AND PLOTT, CHARLES R. FRANKEL, JEFFREY A. AND FROOT, KENNETH A. "Economic Theory of Choice and the Prefer- "Using Survey Data to Test Standard Proposi- ence Reversal Phenomenon," Amer. Econ. Rev., tions Regarding Exchange Rate Expectations," Sept. 1979, 69(4), pp. 623-38. Amer. Econ. Rev., Mar. 1987, 77(1), pp. 133- GRONER, RUDOLF; GRONER, MARINA AND 53. BISCHOF, WALTER F., eds. Methods of heuris- FRANKEL, JEFFREY A. AND RoSE, ANDREW K. "A tics. Hillsdale, NJ: Lawrence Erlbaum, 1983. Survey of Empirical Research on Nominal Ex- GROSSMAN, MICHAEL. "The Correlation Between change Rates," in The handbook of interna- Health and Schooling," in Household product- tional economics.2d ed. Eds.: GENE M. GROSS- ion and consumption. Ed.: NESTOR E. TER- MAN AND KENNETH S. ROGOFF. Amsterdam: LECKYJ. New York:Columbia U. Press, 1973. North Holland, forthcoming. HALTIWANGER, JOHN AND WALDMAN, MICHAEL. FRANTZ, ROGER. "X-Efficiency and Allocative Ef- "Rational Expectations and the Limits of Ra- ficiency: What Have We Learned?" Amer. tionality: An Analysis of Heterogeneity," Amer. Econ. Rev., May 1992, 82(2), pp. 434-38. Econ. Rev., June 1985, 75(3), pp. 326-40. FREY, BRUNO S. AND EICHENBERGER, REINER. . "Responders versus Non-responders: A "Economic Incentives Transform Psychological New Perspective on Heterogeneit," Econ. J., Anomalies,"J. Econ. Behav. Organ., Mar. 1994, Sept. 1991, 101(408), pp. 1085-110. 23(2), pp. 215-34. HARRINGTON, JOHN. The Social Selection of FRIEDMAN, DANIEL. "On Economic Applications Adaptive and Rigid Agents." Manuscript. Johns of Evolutionary Games." Manuscript. U. of Hop ins U., 1994. California, Santa Cruz, 1993. HARTMAN, RAYMOND S.; DOANE, MICHAEL J. FRIEDMAN, MILTON. Essays in positive econom- AND WOO, CHI-KEUNG. "Consumer Rationality ics. Chicago: U. of Chicago Press, 1953. and the Status Quo," Quart. J. Econ., Feb. FRIEDMAN, MILTON AND SAVAGE, LEONARD, J. 1991, 106(1), pp. 141-62. "The Utility Analysis of Choices Involving HAUSMAN, JERRY A. "Individual Discount Rates Risk,"J. Polit. Econ., Aug. 1948, 56(4), pp. 279- and the Purchase and Utilization of Energy- 304. Using Durables," Bell J. Econ., Spring 1979, FROOT, KENNETH A. AND THALER, RICHARD H. 10(l), pp. 33-54. "Foreign Exchange," J. Econ. Perspectives, HEINER, RONALD A. "The Origin of Predictable Summer 1990, 4(3), pp. 179-92. Behavior,"Amer. Econ. Rev., Sept. 1983, 73(4), GARVIN, SUSAN AND KAGEL, JOHN H. "Learning pp. 560-95. in Common Value Auctions: Some Initial Ob- . "The Origin of Predictable Dynamic Be- servations,"J. Econ. Behav. Organ., Dec. 1994, havior," J. Econ. Behav. Organ., Oct. 1989, 25(3), pp. 351-72. 12(2), pp. 233-58. GATELY, DERMOT. "Individual Discount Rates HERRNSTEIN, RICHARD J. AND PRELEC, and the Purchase and Utilization of Energy- DRAZEN. "Melioration: A Theory of Distributed Using Durables: Comment," Bell J. Econ., Choice," J. Econ. Perspectives, Summer 1991, Spring 1980, 11(1), pp. 373-74. 5(3), pp. 137-56. GERTNER, ROBERT. "Game Shows and Economic HEY, JOHN D. "Expectations Formation: Rational Behavior: Risk-Taking on 'Card Sharks'," Quart. or Adaptive or. . . ?" J. Econ. Behav. Organ., J. Econ., May 1993,108(2), pp. 507-21. Dec. 1994, 25(3), pp. 329-49. GODE, DHANANJAY K. AND SUNDER, SHYAM. "Al- HIMMELWEIT, SUSAN. "A Behavioural Model of locative Efficiency of Markets with Zero-Intelli- Learning in Production,"Rev. Econ. Stud., June gence Traders: Market as a Partial Substitute 1976, 43(2), pp. 329-46. 696 Journal of Economic Literature, Vol. XXXIV (June 1996)

HIRSHLEIFER, JACK. "On the Emotions as Guar- the Finitely Repeated Prisoners' Dilemma," J. antors of Threats and Promises," in The latest Econ. Theory, Aug. 1982, 27(2), pp. 245-52. on the best: Essays on evolution and optimality. KUNREUTHER, HOWARD ET AL. Disaster insur- Ed.: JOHN DUPRE. Cambridge: MIT Press, ance protection. New York:Wiley, 1978. 1987, pp. 307-26. KURAN, TIMUR. "Cognitive Limitations and Pref- HIRSHLEIFER, JACK AND RILEY, JOHN G. The erence Evolution," J. Inst. Theoretical Econ., analytics of uncertainty and information. Cam- June 1991, 147(2), pp. 241-73. bridge: Cambridge U. Press, 1992. LAKONISHOK, JOSEF AND SMIDT, SEYMOUR. "Are HOGARTH, ROBIN M. Judgement and choice: The Seasonal Anomalies Real? A Ninety Year Per- psychology of decision. New York:Wiley, 1980. spective," Rev. Finan. Stud., Winter 1988, 1(4), HOLDEN, K.; PEEL, D. A. AND THOMPSON, J. L. pp. 403-26. Expectations: Theory and evidence. London: LANE, DAVID AND VESCOVINI, ROBERTA. "When Macmillan, 1985. Optimization Isn't Optimal: Aggregation and HOLLAND, JOHN H. AND MILLER, JOHN H. "Arti- Information Contagion." Economics Working ficial and Adaptive Agents in Economic The- Paper 95-04-044. Santa Fe Institute, 1995. ory," Amer. Econ. Rev., May 1991, 81(2), pp. LEDYARD, JOHN 0. "Public Goods: A Survey of 365-70. Experimental Research," in The handbook of ITO, TAKATOSHI. "Foreign Exchange Rate Expec- experimental economics. Eds.: JOHN A. KAGEL tations: Micro Survey Data," Amer. Econ. Rev., AND ALVIN E. ROTH. Princeton: Princeton U. June 1990, 80(3), pp. 434-49. Press, 1995, pp. 111-94. JOHANSEN, LEIF. Lectures on macroeconomic LEE, CHARLES M. C; SHLEIFER, ANDREI AND planning. Part 1. General aspects. Amsterdam: THALER, RICHARD H. "Investor Sentiment and North-Holland, 1977. the Closed-End Fund Puzzle," J. Finance, Mar. JONES, STEPHEN R. G. AND STOCK, JAMES H. 1991, 46(1), pp. 75-109. "Demand Disturbances and Aggregate Fluctua- LEIBENSTEIN, HARVEY. "Allocative Efficiency vs. tions: The Implications of Near Rationality," 'X-Efficiency'," Amer. Econ. Rev., June 1966, Econ. J., Mar. 1987, 97(385), pp. 49-64. 56(3), pp. 392-415. Inside the firm: The inefficiencies of hier- KAGEL, JOHN H.; HARSTAD, RONALD M. AND archy. Cambridge: Harvard U. Press, 1987. LEVIN, DAN. "Information Impact and Alloca- tion Rules in Auctions with Affiliated Private LEIBENSTEIN, HARVEY AND MAITAL, SHLOMO. Values: A Laboratory Study," Econometrica, "The Organizational Foundations of X-Ineffi- Nov. 1987, 55(6), pp. 1275-1304. ciency: A Game-Theoretic Interpretation of Argyris' Model of Organizational Learning," KAHNEMAN, DANIEL; KNETSCH, JACK L. AND J. Econ. Behav. Organ., May 1994, 23(3), pp. THALER, RICHARD H. "Experimental Tests of 251-68. the Endowment Effect and the Coase Theo- LEROY, STEPHEN F. "Efficient Capital Markets rem," J. Polit. Econ., Dec. 1990, 98(6), pp. and Martingales," J. Econ. Lit., Dec. 1989, 1325-48. 27(4), pp. 1583-1621. . "Anomalies:The Endowment Effect, Loss LESOURNE, JACQUES. The economics of order and Aversion, and Status Quo Bias," J. Econ. Per- disorder. Oxford: Clarendon Press, 1992. spectives, Winter 1991, 5(1), pp. 193-206. . "Self-Organization as a Process in Evolu- KAHNEMAN, DANIEL; SLOVIC, PAUL AND TVER- tion of Economic Systems," in Nonlinear dy- SKY, AMos, eds. Judgment under uncertainty: namics and evolutionary economics. Eds.: Heuristics and biases. Cambridge: Cambridge RICHARD H. DAY AND PING CHEN. Oxford: U. Press, 1982. Oxford U. Press, 1993, pp. 150-66. KIRMAN, ALAN P. "Whomor What Does the Rep- LEWIS, KAREN K. "Puzzles in International Finan- resentative Individual Represent?"J. Econ. Per- cial Markets." Working Paper No. 4951. NBER, spectives, Spring 1992, 6(2), pp. 117-36. 1994. - "Ants, Rationality, and Recruitment," LIPMAN, BARTON L. "How to Decide How to De- Quart. J. Econ., Feb. 1993, 108(1), pp. 137-56. cide How to . . .: Modeling Limited Rational- . "Economies with Interacting Agents." ity," Econometrica, July 1991, 59(4), pp. 1105- Economics Working Paper 94-05-030. Santa Fe 25. Institute, 1994. LOEWENSTEIN, GEORGE AND PRELEC, DRAZEN. KNETSCH, JACK L. "The Endowment Effect and "Anomalies in Intertemporal Choice: Evidence Evidence of Nonreversible Indifference and an Interpretation," Quart. J. Econ., May Curves," Amer. Econ. Rev., Dec. 1989, 79(5), 1992, 107(2), pp. 573-97. pp. 1277-84. LOEWENSTEIN, GEORGE AND THALER, RICHARD KNIGHT, FRANK H. Risk, uncertainty and profit. H. "Intertemporal Choice," J. Econ. Perspec- Boston: Houghton-Mifflin, 1921; reprinted New tives, Fall 1989, 3(4), pp. 181-93. York:Sentry Press, 1964. LOVELL, MICHAEL C. "Tests of the Rational Ex- KOOPMANS, TJALLING C. Three essays on the state pectations Hypothesis," Amer. Econ. Rev., Mar. of economic science. New York: McGraw-Hill, 1986, 76(1), pp. 110-24. 1957. MACHINA, MARK J. "Dynamic Consistency and KREPS, DAVID M. ET AL. "RationalCooperation in Non-expected Utility Models of Choice under Conlisk: Why Bounded Rationality? 697

Uncertainty," J. Econ. Lit., Dec. 1989, 27(4), PALFREY, THOMAS R. AND PRISBREY, JEFFREY E. pp. 1622-68. "Altruism, Reputation, and Noise in Linear MAILATH, GEORGE J., ed. Evolutionarygame the- Public Goods Experiments." Economics work- ory. Special issue of J. Econ. Theory, Aug. 1992, ing paper 864. California Institute of Technol- 57(2). ogy, 1993. MARCH, JAMES G. AND SIMON, HERBERT A. Or- DE PALMA, ANDREt; MYERS, GORDON M. AND ganizations. New York:Wiley, 1968. PAPAGEORGIOU,YORGOS Y. "Rational Choice MARIMON, RAMON AND MCGRATTAN, ELLEN. Under an Imperfect Ability to Choose," Amer. "On Adaptive Learning in Strategic Games," in Econ. Rev., June 1994, 84(3), pp. 419-40. Learning and rationality in economics. Eds.: PAPADIMITRIOU, CHRISTOS H. AND STEIGLITZ, ALAN KIRMAN AND MARK SALMON. Oxford: KENNETH. Combinatorial optimization: Algo- Basil Blackwell, 1995. rithms and complexity. Englewood Cliffs, NJ: MARIMON, RAMON; MCGRATTAN, ELLEN AND Prentice-Hall, 1982. SARGENT, THOMAS J. "Money as a Medium of PAYNE, JOHN W.; BETTMAN, JAMES R. AND Exchange in an Economy with Artificially Intel- JOHNSON, ERIC J. "Behavioral Decision Re- ligent Agents," J. Econ. Dynam. Control, May search: A Constructive Processing Perspective," 1990, 14(2), pp. 329-73. Annual Rev. Psych., 1992, 43, pp. 87-131. MARIMON, RAMON AND SUNDER, SHYAM. "Inde- . The adaptive decision ma7er. Cambridge: terminacy of Equilibria in a Hyperinflationary Cambridge U. Press, 1993. World: Experimental Evidence," Econometrica, PEREIRA, ALFREDO M. AND SHOVEN, JOHN B. Sept. 1993, 61(5), 1073-1107. "Survey of Dynamic Computational General MARSCHAK, JACOB AND RADNER, Roy. Economic Equilibrium Models for Tax Policy Evaluation," theory of teams. New Haven: Yale U. Press, J. Policy Modeling, Fall 1988, 10(3), pp. 401- 1972. 36. MARTELLO, SILVANO AND TOTH, PAOLO. Knap- PESARAN, M. HASHEM AND TIMMERMANN, AL- sack problems: Algorithms and computer imple- LAN. "Predictability of Stock Returns: Robust- mentations. New York:Wiley, 1990. ness and Economic Significance," J. Finance, MEADE, JAMES E. Efficiency, equality and the Sept. 1995, 50(4), pp. 120-28. ownership of property. Cambridge: Harvard U. PETERSON, STEVEN P. "Forecasting Dynamics Press, 1964. and Convergence to Market Fundamentals: Evi- MEHRA, RAJNISH AND PRESCOTT, EDWARD C. dence from Experimental Asset Markets," J. "The Equity Premium: A Puzzle," J. Monet. Econ. Behav. Organ., Dec. 1993, 22(3), pp. Econ., Mar. 1985,15(2), pp. 145-61. 269-84. MIDGLEY, DAVID F.; MARKS, ROBERT E. AND PHELPS, EDMUND S. AND WINTER, SIDNEY G. COOPER, LEE G. "Breeding Competitive "Optimal Price Policy under Atomistic Compe- Strategies." Economics Working Paper 95-06- tition," in Microeconomic foundations of em- 052. Santa Fe Institute, 1995. ployment and inflation Theory. Eds.: EDMUND MONGIN, PHILIPPE AND WALLISER, BERNARD. S. PHELPS ET AL. New York: Norton, 1970, pp. "Infinite Regressions in the Optimizing Theory 309-37. of Decision," in Risk, decision and rationality. PINGLE, MARK A. "Costly Optimization: An Ex- Ed.: BERTRAND R. MUNIER. Dordrecht: D. periment," J. Econ. Behav. Organ., Jan. 1992, Reidel, 1988, pp. 435-57. 17(1), pp. 3-30. NAISH, HOWARD F. "The Near Optimality of PITZ, GORDON F. AND SACHS, NATALIE J. "Judg- Adaptive Expectations,"J. Econ. Behav. Organ., ment and Decision: Theory and Application," Jan. 1993, 20(1), pp. 3-22. Annual Rev. Psych., 1984, 35, pp. 139-63. NELSON, RICHARD R. "Recent Evolutionary PLOTT, CHARLES R. AND SUNDER, SHYAM. "Ra- Theorizing about Economic Change," J. Econ. tional Expectations and the Aggregation of Di- Lit., Mar. 1995, 33(1), pp. 48-90. verse Inlformationin Laboratory Security Mar- NELSON, RICHARD R. AND WINTER, SIDNEY G. kets," Econometrica, Sept. 1988, 56(5), pp. An evolutionary theory of economic change. 1085-1118. Cambridge: Harvard U. Press, 1982. RABIN, MATTHEW."Cognitive Dissonance and So- NEWELL, ALLAN AND SIMON, HERBERT A. "Com- cial Change," J. Econ. Behav. Organ., Mar. puter Science as Empirical Inquiry: Symbols 1994,23(2), pp. 177-94. and Search," in Foundations of cognitive sci- RADNER, RoY. "Competitive Equilibrium Under ence: The essential readings. Ed.: JAY L. GAR- Uncertainty," Econometrica, Jan. 1968, 36(1), FIELD. New York:Paragon, 1990, pp. 113-38. pp. 31-58. NISBETT, RICHARD AND Ross, LEE. Human in- . "Satisficing,"J. Math. Econ., June-Sept. ference: Strategies and shortcomings of social 1975,2(2), pp. 253-62. judgment. Englewood Cliffs, NJ: Prentice-Hall, . "Hierarchy:The Economics of Managing," 1980. J. Econ.Lit., Sept. 1992,30(3), pp. 1382-1415. ORCUTT, Guy; MERZ, JOACHIM AND QUINKE, RADNER, ROY AND ROTHSCHILD, MICHAEL. "On HERMANN. Microanalytic simulation models to the Allocation of Effort,"J. Econ. Theory, June support social and financial policy. Amsterdam: 1975,10(3), pp. 358-76. North Holland, 1986. RADZICKI, MICHAEL J. AND STERMAN, JOHN D. 698 Journal of Economic Literature, Vol. XXXIV (June 1996)

"Evolutionary Economics and System Dynam- ory," Philosophy and Public Affairs, Summer ics," in Evolutionary concepts in contemporary 1977, 6(4), pp. 317-44. economics. Ed.: RICHARD W. ENGLAND. Ann . "Rational Behavior," in The new Palgrave: Arbor: U. of Michigan Press, 1994, pp. 61-92. A dictionary of economics. Eds.: JOHN EAT- RAIFFA, HOWARD. Decision analysis: Introduc- WELL, MURRAY MILGATE, AND PETER NEW- tory lectures on choices under uncertainty. MAN. London: Macmillan, 1987, pp. 68-76. Reading, MA: Addison-Wesley, 1968. SEYHUN, H. NEJAT. "Why Does Aggregate Insider RAPOPORT, ANATOL; GUYER, MELVIN J. AND Trading Predict Future Stock Returns?" Quart. GORDON, DAVID G. The 2X2 game. Ann Arbor: J. Econ., Nov. 1992, 107(4), pp. 1303-31. U. of Michigan Press, 1976. SHEA, JOHN. "Union Contracts and the Life- ROSENTHAL, ROBERT W. "Rules of Thumb in Cycle/Permanent-Income Hypothesis," Amer. Games," J. Econ. Behav. Organ., Sept. 1993, Econ. Rev., Mar. 1995, 85(1), pp. 186-200. 22(1), pp. 1-14. SHEFRIN, HERSH M. AND STATMAN, MEIR. "The ROTH, ALVIN E. "Laboratory Experimentation in Disposition to Sell Winners Too Early and Ride Economics: A Methodological Overview," Econ. Losers Too Long: Theory and Evidence," J. Fi- J., Dec. 1988, 98(393), pp. 974-1031. nance, July 1985, 40(3), PP. 777-92. ROTHSCHILD, MICHAEL. "Further Notes on the . "Behavioral Capital Asset Pricing Theory," Allocation of Effort," in Adaptive economic J. Finan. Quant. Anal., Sept. 1994, 29(3), pp. models. Eds.: RICHARD H. DAY AND THEO- 323-49. DORE GROVES. New York: Academic Press, SHEFRIN, HERSH AND THALER, RICHARD H. 1975, pp. 195-220. "The Behavioral Life-Cycle Hypothesis," Econ. RUBINSTEIN, ARIEL. "Finite Automata Play Re- Inquiry, Oct. 1988, 26(4), pP.609-43. peated Prisoner's Dilemma," J. Econ. Theory, SHILLER, ROBERT J. Market volatility. Cam- June 1986, 39(1), pp. 83-96. bridge: MIT Press, 1989. RUSSELL, THOMAS AND THALER, RICHARD H. SHOGREN, JASON F. ET AL. "Resolving Differ- "The Relevance of Quasi Rationality in Com- ences in Willingness to Pay and Willingness to petitive Markets," Amer. Econ. Rev., Dec. 1985, Accept," Amer. Econ. Rev., Mar. 1994, 84(1), 75(5), pp. 1071-82. pp. 255-70. SAH, RAAJ KUMAR AND STIGLITZ, JOSEPH E. SILVERBERG, GERALD; Dosi, GIOVANNI AND "The Architecture of Economic Systems: Hier- ORSENIGO, LuIGI. "Innovation, Diversity and archies and Polyarchies," Amer. Econ. Rev., Diffusion: A Self-Organization Model," Econ. Sept. 1986, 76(4), pp. 716-27. J., Dec. 1988, 98(393), 1032-54. SAMUELSON,WILLIAM AND ZECKHAUSER,RICH- pp. SIMON, HERBERT A. "A Behavioral Model of Ra- ARD. "Status Quo Bias in Decision Making," J. tional Choice," Quart. J. Econ., Feb. 1955, Risk Uncertainty, Mar. 1988, 1(1), pp. 7-59 69(1), 99-118. SARGENT,THOMAS J. Bounded rationality in mac- pp. . "Rationality in Psychology and Econom- roeconomics. Oxford: Clarendon Press, 1993. S209- "Stochastic Dynamics in Games: ics," J. Bus., Oct. 1986, 59(4, pt. 2), pp. SARIN, RAJIV. 24. Learning and Evolution." Ph.D. dissertation. U. . "Satisficing," in The new A dic- of California, San Diego, 1994. Palgrave: economics. Eds.: JOHN EATWELL, SAVAGE,LEONARD J. Thefoundations of statistics. tionary of New York: Wiley, 1954. MURRAY MILGATE, AND PETER NEWMAN. SCHELLING, THOMAS C. "Altruism, Meanness, London: Macmillan, 1987, pp. 243-45. of and Other Potentially Strategic Behaviors," . "Invariants Human Behavior," Annual Rev. 1-19. Amer. Econ. Rev., May 1978, 68(2), pp. 229-30. Psych., 1990, 41, pp. . "Self-Command in Practice, in Policy, and . "Altruism and Economics," Amer. Econ. in a Theory of Rational Choice," Amer. Econ. Rev., May 1993, 83(2), pp. 156-61. Rev., May 1984, 74(2 pp. 1-11. SIMON, HERBERT A. AND SCHAEFFER, SCHMALENSEE, RICHARD AND WILLIG, ROBERT JONATHAN. "The Game of Chess," in Handbook D., eds. Handbook of industrial organization. of game theory. Eds.: ROBERT J. AUMANN AND Vol. 1 and Vol. 2. Amsterdam: North-Holland, SERGIU HART. Amsterdam: Elsevier, 1992, pp. 1989. 1-17. SELTEN, REINHARD. "The Chain Store Paradox," SINGLETON, KENNETH J. "Specification and Esti- Theory and Decision, Apr. 1978, 9(2), pp. 127- mation of Intertemporal Asset Pricing Models," 59. in Handbook of monetary economics. Vol. I. . "Evolution, Learning, and Economic Be- Eds.: BENJAMIN M. FRIEDMAN AND FRANK H. havior," Games Econ. Behav., Feb. 1991, 3(1), HAHN. Amsterdam: Elsevier, 1990, pp. 583- pp. 3-24. 626. SELTEN, REINHARD AND STOECKER,ROLF. "End SLONIM, ROBERT. "Learning in a Search-for-the- Behavior in Sequences of Finite Prisoner's Di- Best-Alternative Experiment," J. Econ. Behav. lemma Supergames," J. Econ. Behav. Organ., Organ., Oct. 1994, 25(2), pp. 141-65. Mar. 1986, 7(1), pp. 47-70. SLOVIC, PAUL; LICHTENSTEIN, SARAH AND SEN, AMARTYAK. "Rational Fools: A Critique of FISCHHOFF, BARUCH. "Decision Making," in the Behavioral Foundations of Economic The- Stevens' handbook of experimental psychology. Conlisk: Why Bounded Rationality? 699

Eds.: RICHARD ATKINSON ET AL. New York: 1992; also available in paperback, Princeton U. Wiley, 1988, pp. 673-738. Press, 1994. SMALLWOOD, DENNIS E. AND CONLISK, JOHN. THALER, RICHARD H. AND SHEFRIN, HERSH. "Product Quality in Markets Where Consumers "An Economic Theory of Self-Control,"J. Polit. Are Imperfectly Informed," Quart. J. Econ., Econ., Apr. 1981, 89(2), pp. 392-406. Feb. 1979, 93(1), pp. 1-23. TIMMERMANN, ALLAN. "Can Agents Learn to SMITH, HOLLY. "Deciding How to Decide: Is Form Rational Expectations? Some Results on There a Regress Problem?" in Foundations of Convergence and Stability of Learning in the decision theory. Eds.: MICHAEL BACHARACH UK Stock Market," Econ. J., July 1994, AND SUSAN HURLEY. London: Basil Blackwell, 104(425), pp. 777-97. 1991, pp. 194-219. . "VolatilityClustering and Mean Reversion SMITH, VERNON L. "Theory, Experiment and of Stock Returns in an Asset Pricing Model wth Economics,"J. Econ. Perspectives, Winter 1989, Incomplete Learning."Discussion paper 95-23. 3(1), pp. 151-69. U. of California, San Diego, 1995. in The state of "Rational Choice: The Contrast between TOBIN, JAMES. Untitled cha pter Economics and Psychology" J. Polit. Econ., economic science: Views of six Nobel laureates. Aug. 1991, 90(4), pp. 877- 7. Ed.: WERNER SICHEL. Kalamazoo, Michigan: SMITH, VERNON L.; SUCHANEK, GERRY L. AND Upjohn Institute, 1989, pp. 61-76. WILLIAMS, ARLINGTON W. "Bubbles, Crashes, TVERSKY,AMOS. "Elimination by Aspects: A The- and Endogenous Expectations in Experimental ory of Choice," Psych. Rev., July 1972, 79(4), Spot Asset Markets,"Econometrica, Sept. 1988, pp. 281-99. 56(5), pp. 1119-51. TVERSKY, AMOS AND KAHNEMAN, DANIEL. "Loss Choice: A Reference- L. AND WALKER, JAMES M. Aversion in Riskless SMITH, VERNON Econ., Nov. 1991, "Monetary Rewards and Decision Cost in Ex- Dependent Model," Quart. J. 1039-61. perimental Economics," Econ. Inquiry, Apr. 106(4), pp. H. 1993, 31(2), pp. 245-61. TVERSKY, AMOS AND THALER, RICHARD "Anomalies: Preference Reversals," J. Econ. SOBEL, JOEL. "It's Not What You Know, It's Who Spring 1990, 4(2), pp. 201-11. You Play," in Game equilibrium models II: Perspectives, VAN HUYCK, JOHN B.; COOK, JOSEPH P. AND morals, and markets. Ed.: REINHARD Methods, BATTALIO,RAYMOND C. "Selection Dynamics, 71- SELTEN. Berlin: Springer-Verlag, 1991, pp. Asymptotic Stability, and Adaptive Behavior,"J. 84. Polit. Econ., Oct. 1994, 102(5), pp. 975-1005. . "How to Count to One Thousand," Econ. VEGA-REDONDo, FERNANDO. "Competition and J., Jan. 1992,102(410), pp. 1-8. Culture in an Evolutionary Process of Equilib- SOLOW, ROBERT M. Untitled chapter in The state rium Selection: A Simple Example," Games of economic science: Views of six Nobel laure- Econ. Behav., Oct. 1993, 5(4), pp. 618-31. ates. Ed.: WERNER SICHEL. Kalamazoo, Michi- Boundedly Rational Agents Play Upjohn Institute, 1989, pp. 25-39. . "Bayesian gan: the Finitely Repeated Prisoner's Dilemma," STAHL, DALE 0. "Evolution of Smartn Players, Theory and Dectsion, Mar. 1994, 36(2), pp. Games Econ. Behav., Oct. 1993, 5(4), pp. 604- 187-206. 17. VRIEND, NICOLAS J. "Self-Organized Markets in a STAHL, DALE 0. AND WILSON, PAUL W. "Exeri- Decentralized Economy." Economics Working mental Evidence on Players' Models of Other Paper 94-03-013. Santa Fe Institute, 1994. Players," J. Econ. Behav. Organ., Dec. 1994, WALDMAN, MICHAEL. "Systematic Errors and the 25(3), pp. 309-27. Theory of Natural Selection," Amer. Econ. Rev., STERMAN, JOHN D. "Misperceptions of Feedback June 1994, 84(3), pp. 482-97. in Dynamic Decision Making," Organ. Behav. WATSON, JOEL. "Strategy Perturbations in Re- Human Decision Processes, June 1989, 43(3), peated Games as Rules of Thumb." Discussion pp. 301-35. Paper 94-20. U. of California, San Diego, 1994. SUGDEN, ROBERT. "Rational Choice: A Survey of WELCH, Ivo. "Sequential Sales, Learning, and Contributions from Economics and Philoso- Cascades," J. Finance, June 1992, 47(2), pp. phy," Econ. J., July 1991, 101(407), pp. 751-85. 695-732. THALER, RICHARD H. "Seasonal Movements in WILLIAMSON, OLIVER E. The economic institu- Security Prices. II. Weekend, Holiday, Turn of tions of capitalism: Firms, markets, relational the Month, and Intraday Effects," J. Econ. Per- contracting. New York:Free Press, 1985. spectives, Fall 1987, 1(2), pp. 169-77. a Economic organization: Firms, markets, . "Anomalies: The Winner's Curse," J. and policy control. Brighton: Wheatsheaf, 1986. Econ. Perspectives, Winter 1988, 2(1), pp. 191- WINSTON, GORDON C. "Imperfectly Rational 202. Choice: Rationality as the Result of a Costly Ac- . "Saving, Fungibility, and Mental Ac- tivity." J. Econ. Behav. Organ., Aug. 1989, counts," J. Econ. Perspectives, Winter 1990, 12(1), pp. 67-86. 4(1), pp. 193-205. WINTER, SIDNEY G. "Economic 'Natural Selec- - The winner's curse. Paradoxes and anoma- tion' and the Theory of the Firm," Yale Econ. lies of economic life. New York: Free Press, Essays, Spring 1964, 4(1), pp. 225-72. 700 Journal of Economic Literature, Vol. XXXIV (June 1996)

. "Satisficing, Selection, and the Innovatin ory," J. Econ. Behav. Organ., Dec. 1982, 3(4), Remnant," Quart. J. Econ., May 1971, 85(2) pp. 277-321. pp. 237-61. WITT, ULRICH."Firms' Market Behavior Under . "Optimization and Evolution in the The- Imperfect Information and Economic Natural ory of the Firm," in Adaptive economic models. Selection," J. Econ. Behav. Organ., Sept. 1986, Eds.: RICHARD H. DAY AND THEODORE 7(3), pp. 265-90. GROVES. New York: Academic Press, 1975, pp. YOUNG,H. PEYTON."The Evolution of Conven- 73-118. tions," Econometrica, Jan. 1993, 61(1), pp. 57- . "Binary Choice and the Supply of Mem- 84.