<<

AT THE MILLENNIUM

Robert S. Goldfarb and Thomas C. Leonard

istory teaches that the future will be unkind improve the quality of our forecasts, based on an Hto those foolhardy enough, or well compen- analogy with quantitative forecasting. There, the sated enough, to attempt prediction. The most consensus forecast—which averages individual eminent American of a century ago, forecasters’ predictions—has been shown to be , predicted in October 1929 that “the more accurate than many of its component fore- end of the decline in the stock will … be casts. Many distinguished have since a few more days at the most.” IBM chairman Tho- 1990 risked diagnosing the discipline’s situation mas Watson estimated in 1943 that “there is a or future, including the twenty-two papers in the world market for about five computers.” Charles Economic Journal’s centenary volume (January Duell, of the U.S. Patent Office, asserted in 1899 1991), Solow and Kreps in Daedalus (Winter that “everything that can be invented has been 1997), Lipsey ( Journal of invented.” In 1894, on the cusp of the golden age June 2001), Colander ( Journal of the History of of physics, physicist A.A. Michelson concluded Economic Thought June 2000), and a symposium that “the more important fundamental laws and in the Journal of Economic Perspectives (Winter facts of physical science have all been discovered.” 2000). One remaining bias arises from our profes- Economics also teaches skepticism about pre- sional specialization in , the study diction—forecasts of any real are unlikely of individuals, firms and industries. Macroeconom- to remain valuable for long. Still, the opportunity ics, the study of -wide phenomena like to speculate on the future of economics is irre- , , and , will get short shrift sistible, not least because any guess at the future in this review. must begin with the present, and with the his- Prediction is not used here in the narrow sense tory that brought us here. of forecasting the magnitude or direction of once said, only partly tongue-in- change of a particular variable, such as a stock cheek, that “economics is what economists do.” . Instead, we identify what we—and Economists today do many and diverse things. especially the scholars we cite—see as important Testifying to this diversity, the New Palgrave Dic- tendencies within the profession, and guess which tionary of Economics contains nearly two thou- of these tendencies might strengthen. By virtue sand essay-length entries. How economists do of their broader and vaguer character, our con- economics remains more monolithic. In this sense, jectures on “where economics might be going” neither of your correspondents is a representa- obviously risk less than did Irving Fisher and his tive economist. In a discipline that has adopted client, the Yale University endowment. the techniques and ethos of applied mathemat- ics, we mostly conduct our arguments in prose. SpecieSpeciess of EconomEconomistsists and TTheoTheorheorry vs. EvidenEvidenccee Moreover, our research include the his- Perhaps the hoariest methodological debate in tory and philosophy of economics, fields that to- economics concerns the weight that economics day occupy the periphery of the discipline, so our ought to give to theory versus evidence. Theory views may not be representative. by itself is empty, and of limited value for a disci- To partially compensate for these biases, we pline that aspires at times to be a policy science, present views other than our own. This might also but empirical inquiry uninformed by theory is

24 SOCIETY • NOVEMBER / DECEMBER 2002 likely to be blind. For more than a century, and in spectives Winter 2002) has argued that “(w)hat various guises, economics has revisited, some- really matters in auction design are the same is- times unintentionally, the matter of how to weight sues that any industry regulator would recognize theory versus evidence. Today, what different as key concerns: discouraging collusion, entry- types of (academic) economists do reflects rival deterring and predatory behavior. In short, good views of what the weights ought to be. auction design is mostly good elementary econom- We divide the genus economist into five spe- ics” (pp. 169-170). cies. While there is too much overlap for taxo- Our third species works on pure statistical nomic precision, the different species usefully theory. These “theoretical econometricians” de- stand in for different attitudes on theory versus velop statistical theory motivated, in part, by the evidence. particular theoretical problems that arise in “Pure theorists” are in the business of logically . A fourth species of economist com- deducing the implications of a set of behavioral prises the empirically oriented. One sub-species, axioms taken as fundamental (Hahn in Economic “applied econometricians,” uses sophisticated Journal 1991: p. 47), an enterprise describable as econometric tools—statisti cal techniques devel- Euclidian in spirit (Solow in Daedalus 1997: p. oped for economic applications—to analyze data. 42). Pure theorists prove theorems and lemmas. Though they work with data, their interests are Most of is of the pure theory type. mostly “tool-driven”; they apply new economet- The connection between pure theory and the ric techniques to established areas of empirical economy runs from tenuous to none. At its most inquiry. A second sub-species of empirical econo- abstract and most archetypal (for example, the mists is more data-driven. They use established Nobel-prize-winning 1959 work of Gerard econometric techniques to analyze new (or ex- Debreu), pure economic theory does not even panded) empirical data sets. A third sub-species purport to have empirical consequences. As theo- of empirical economists, still something of a nov- rist puts it (in his 1998 volume elty in economics, uses experimental methods— Modeling Bounded Rationality) , pure theory especially laboratory games—to test a variety of “does not pretend to predict or to advise … the behavioral propositions. most [it] can do is to clarify the concepts we use” A fifth category, “” is not (p. 194). entirely a separate species. These economists are A second species, “applied theorists,” refashion distinguished more by what they study than how pure theory so it can explain or predict real-world they study it. Methodologically eclectic, they use phenomena. One successful example, cited by applied theory and empirical methods to analyze Sutton (in his 2000 book, Marshall’s Tendencies: issues in their field of : for example, labor What Can Economists Know? ), is options pric- economics, , or urban econom- ing. An option is the contractual right to buy or ics. The vast majority of economists can be located to sell an asset, such as common stock, at a speci- in the applied economist category. Indeed, applied fied price and by a future date. Applied theorists econometricians or applied theorists will at times, devised a way to predict an option’s value, given as their research focus shifts, fit the applied eco- only the specified price, the current stock price, nomics category. the time until expiration, a discount rate, and an The different species in our crude taxonomy estimate of the stock-price volatility. Robert roughly proxy different attitudes toward the Merton and won the proper mix of theory and evidence in economics. for their contributions to valuing options. Another As we will see, this question of how to weigh successful example, also from Sutton, is auction theory and evidence is a central concern of the design, which makes use of auction theory, a distinguished scholars we survey. branch of non-cooperative game theory, to design rules that result in successful auctions. As the EconomEconomicsics DefDefineinedd by Its MethodMethod examples suggest, applied theory works best in When outsiders judge modern economics to fairly “controlled” empirica l settings, that is, loca- be monolithic, especially as compared with its tions where the theoretical assumptions are more cognate disciplines, they are referring to eco- likely to obtain. Successful application of theory nomic method—how economics is done. What also often requires a good dollop of elementary contemporary economists do, says , economics. Klemperer (Journal of Economic Per- Nobel Laureate and one of the discipline’s ablest

ECONOMICS AT THE MILLENNIUM 25 ambassadors, is build models: “modern mainstream ing current method as “fact-driven model build- economics consists of little else but examples of ing.” Thus does wryly describe an this process” (Solow in Daedalus 1997: p. 43). In economist as “someone who sees that something a typical economics paper, the section immedi- works in practice and wonders if it also works in ately following the introduction is invariably en- theory” (in the , May titled “The Model.” A model is “a deliberately sim- 1988, p. 7). What most unifies economics is how plified [mathematical] representation of a much economics is done. Colander (in Journal of the more complicated situation.… The idea is to fo- History of Economic Thought 2000) concurs. cus on one or two causal or conditioning vari- “Modern economics” has become “enormously ables, excluding everything else, and hope to un- broad in its acceptance of various assumptions derstand how these aspects of reality work and and content.” But it is “extremely narrow when it interact” (ibid). comes to method.… The modeling approach to The determined historian can locate unexpect- problems is the central element of modern eco- edly venerable quantitative, mathematical and sta- nomics” (p. 137). tistical precursors to modern economics: there is, for example, the “Political Arithmetick” of the EconomEconomicsics Def ined by its CanoniCanonicalcal IdIdeaseas 1660s, and Augustin Cournot’s (1838) prescient What ingredients have been used to construct and sophisticated mathematical treatments of economic models? Two ideas have typically monopoly and duopoly. But scholarly economics formed the heart of the structure—maximization journals before World War II are virtually free of and equilibrium . Maximization says individua l even rudimentary mathematical notation. In agents make optimal choices consiste nt with a today’s journals, every other page is “pockmarked completely specified objective or “maximand”: by algebra,” to use Baumol’s phrase (in the Eco- “” for consumers and profit for firms. Equi- nomic Journal 1991, p. 2). librium says that the aggregate consequences of But while mathematics is obviously useful for these individual choices are equilibria stable and economic modeling, it is not sufficient. Pure theo- unique enough to permit prediction. We can, af- rists, who are occupied with formal consider- ter Kreps (in Daedalus 1997), call maximization ations, do not bother with models, because mod- and equilibrium the canonical principles. els are meant to isolate and refer to observable If agents are to maximize, they must not only empirical phenomena. Models are also routinely be purposeful and forward looking, they must also designed for special cases, involve fairly basic have a detailed probabilistic picture of the future mathematics, and are therefore neither general nor (Kreps 1997: p. 71). Maximization also requires fancy enough for the pure theorist. that agents’ preferences over possible states of the Because economic models are empirically ori- future be “well-behaved,” that is, amenable to a single, ented, and because they can be stated (if less com- complete, and transitive rank ordering. Rationality pactly) in non-mathematical terms, Solow regards consists only in doing what one most prefers. as misplaced the common criticism that econom- Preferences are typically treated as primitive ics is excessively formal. Model builders, says concepts—given, and beyond evaluation or analy- Solow, are “obsessed with data” (1997: p. 57). sis. The economist’s pared-down treatment of ra- Solow even explains the post-war transition in tional choice is sometimes called “thin rational- American economics from prose to algebraic ex- ity,” because “it leaves unexamined the beliefs and pression as the product of more and better statis- desires that form the reasons for the actions” (John tical data. “Technique and model-building,” says Elster, Sour Grapes , 1985). More broadly, eco- Solow, “came along with the expanding availabil- nomic rationality embeds Lionel Robbins’ concep- ity of data, and each reinforces the other. Facts tion of economics, where the fundamental eco- ask for explanations, and explanations ask for new nomic problem is resource allocation under facts” (ibid, p. 47). In fact, “the modern approach . Economics is the science, Robbins said to economics is mostly about accounting for data” famously in 1935, “which studies human behav- (ibid: p. 53). ior as a relationship between ends and scarce We might quarrel with Solow’s historical the- means which have alternative uses.” sis that better data explains the ascendancy of The term equilibrium most commonly refers model building technique, but there is no disput- to an aggregate outcome that is stable in the sense ing his characterization of the discipline’s prevail- that none of the maximizing agents can change

26 SOCIETY • NOVEMBER / DECEMBER 2002 it, or, if they can change it, none would want to. main: (1) has the extent of mathematical expres- In so-called “general equilibrium” accounts, agents sion in economics gone too far—in the sense that are ordinarily powerless to affect outcomes, some- too much of a good thing has created real intel- times by virtue of their large numbers and small lectual costs, and (2) is the economic method of size relative to the market. They are called “price fact-driven model building scientifically success- takers.” In game-theoretic settings, where agents ful? can influence outcomes (“price makers”), an equi- Richard Lipsey (in Journal of Economic Meth- librium that no (maximizing) agent wishes to de- odology 2001, p. 84) thinks that the mathemati- viate from is, roughly speaking, a Nash equilib- zation of economics has had several adverse ef- rium. Economic method focuses on equilibria fects: (1) a tendency for “(g)enerality to be desired more than on the out-of-equilibrium processes by for its own sake, even when it obscures the sim- which economic agents arrive at these equilibria. plicity of solutions to some problems”; (2) obscu- If there are multiple equilibria, as there often are, rantism: using mathematics “even if it adds noth- especially in game-theoretic settings, then predic- ing to your verbal analysis”; (3) intellectual tion requires an additional theoretical explanation : “the high cost of learning advanced of how economic agents come to settle on a mathematics [pushes] more descriptive and fac- unique equilibrium outcome. tual material out of the curriculum”; and (4), per- In sum, economics uses its canonical behavioral haps most dangerous of all, the confusion of va- principles, maximization and equilibrium, as a lidity and truth: “the implicit assumption that if source of ideas and in order to organize its char- some result is derived from a complex model con- acteristic way of doing business, the enterprise taining all the OK assumptions … it must be true.” of writing down, testing, and refining models. (in Economic Journal 1991) seconds Lipsey in the concern that excessive fo- SSccholarhhoolalarrs onon the Cur rent State ooff the Discipline cus on mathematical expression works to crowd How well does the method of fact-driven model out “other lines of [scholarly] attack.” “(T)hese days building actually work? Later we will take up pre- few specialized students are allowed to proceed dictions about the future of the canonical prin- without devoting a very considerable portion of ciples, and about the scope of economics. Take their time to the acquisition of mathematical tools, first pure theory, which is unconcerned with evi- and they often come away feeling that any … dence. Pure theory occupies a curious place in writing they produce will automatically be re- contemporary economics. On the one hand, its jected … if it is not liberally sprinkled with … practitioners are regarded as among the most able algebraic symbols.” This, Baumol says, has led to a of economists, and mathematical talent is dispro- misallocation of intellectual resources within eco- portionately admired and rewarded in the profes- nomics. “Mathematics,” says general equilibrium sion. On the other hand, said the eminent theo- theorist “has gone too far, lead- rist in 1984 (in Equilibrium and ing theorists to have an inadequate concern for ) there is something unworldly actuality” (in the Economic Journal 1991: pp. 73- about pure theorizing. “It cannot be denied,” Hahn 74). Nobel Laureate in the same said, “that there is something scandalous in the symposium also worries that economists use spectacle of so many people refining the analysis mathematics for obscurantist ends. He argues that of economic states which they have no reason to contemporary economists rely “on mathematics suppose will ever, or have ever, come about” (p. and econometrics beyond the point of vanishing 88). This non-empirical orientation may be why, returns.” The cost, says Friedman, is under-invest- as Solow claims, the vast majority of economists ment in empirical work. Friedman’s explanation pay almost no attention to pure theory (Solow is economic. It is cheaper per publication to pro- 1997: p. 43), even as they honor the pure theo- duce theorems than it is “to gather original data rists. … to explore their reliability and accuracy … and Solow wants to distinguish workaday model- derive a full understanding of the historical and building, which employs mathematics, from the institutional circumstances in which they were formalism of pure theory. His argument is that the generated” ( Economic Journal 1991: 37). discipline’s use of mathematics, by itself, is nei- Friedman’s last point, which returns to the re- ther formalism nor any other methodological sin. lationship between theory and evidence, is ech- We mostly agree, but two related questions re- oed in one form or another by many scholars in

ECONOMICS AT THE MILLENNIUM 27 our sample. Many agree that theory, which here cial and physical sciences—politic s, , means analytical puzzle-solving, is cheap and that chemistry and physics—for the 1982-86 period. data are expensive , so economic and statistical Economics is more mathematical than political theory tends to outrun the evidence for it (Solow science and sociology, which produce strictly 1997: p. 57). “Without a close relation between mathematical papers only in 18 and one percent evolving theory and empirical observation,” says of all cases, respectively. Economics is also less Lipsey, “new theory tends to be developed in un- empirical than its sister fields—non-em pirical constrained ways that are empirically relevant work in political science and sociology is 42 and only by accident” (2001: p. 174). Pencavel (in the 22 percent, respectively, well under the 50-60 Economic Journal 1991) concurs, asserting that percent in Leontief’s and Morgan’s surveys. Papers economists often avoid meaningfully confronting without empirical work are unheard of in chem- theoretical hypotheses with empirical evidence, istry (zero percent) and rare in physics (12 per- in favor of theorizing alone, which offers quicker cent). or more certain returns. One standard defense, that economics lacks The harder question, on which scholars dis- recourse to the controlled experimental tech- agree, is to what extent theory should outrun the niques characteristic of the natural sciences, is as evidence for it. The critique of modern econom- old as the critique. The Nobel laureate and pure ics as insufficiently empirical is a venerable one. theorist Gerard Debreu argues, for example, that In his version, Andrew Oswald (in the Economic economics’ inherent empirical disadvantage re- Journal, 1991) cites the grumbles of two Nobel quires relatively greater in pure theory. Laureates, and a former editor of The Journal of “Being denied a sufficiently secure experimental Economic Literature : “ (1982) has base, economic theory has had to adhere to the argued that our discipline has deteriorated into a rules of logical discourse and must renounce the second-rate branch of applied mathematics in facility of internal inconsistency … [on pain of] which, unscientifically, researchers eschew em- being useless…” ( AER, 1991, p. 3). While changes pirical investigation. (1986, p. are afoot with respect to experimental econom- 384) says that the subject is ‘widely perceived to ics, most empirical work in economics still con- be discredited because it has so little empirical sists of statistical inference—ec onometric at- content and cares so little about developing it.’ tempts to find law-like regularities in historical John Pencavel (1989, p. 1) concludes that econo- observations. mists do not want applied work to be done, be- Inference from historical observations presents cause it is likely to reveal the irrelevance of their hazards. Unlike physical phenomena, which are hypotheses and undermine their ability to derive ontologically well behaved—that is, relatively in- sweeping implications from theoretical models” variant over (human-scale) time and place—eco- (Oswald, p. 75). nomic phenomena are not. Economics, says Lipsey, Leontief’s critique (in Science, 1982) derived “has nothing like the theories of physics that pre- from an analysis of The American Economic Re- dict specific quantitative outcomes.” view (AER) from March 1972 to December 1981, The first hazard of this kind of historical infer- which found that more than 50 percent of the ence, argues Solow, is assuming too much: econo- papers contained mathematical argument with no mists’ reliance on physics-like modeling tech- empirical data; in total, about two-thirds contained niques can lead them to downplay the ontological no empirical work. Theodore Morgan ( Journal of difference between physical and social phenom- Economic Perspectives 1988) updates Leontief by ena. There is, says Solow, “the temptation to be- examining the AER from March 1982 to Decem- lieve that laws of economics are like laws of phys- ber 1986. He finds an increase in empirical analy- ics: exactly the same everywhere on earth and ses, rising from roughly one third in the Leontief forever. But the part of economics that is inde- period up to one half of all papers in the later pendent of history and social context is not only period. Morgan also surveyed Britain’s leading small but dull” (1997: p. 36). journal, the Economic Journal , over the entire The second hazard is assuming too little: econo- 1972-1986 period, finding a roughly constant 58 mists can invoke the relative inferiority of social percent of papers with empirical work. Morgan data as a crutch for ignoring empirical anomalies. also compared economics (as represented by the In natural science, Lipsey says, because there are AER and the EJ) to a sample from four other so- “tight theories” based on stable quantitative rela-

28 SOCIETY • NOVEMBER / DECEMBER 2002 tions, “conflicts between theory and new evidence tives. A deeper problem, emphasized by Lipsey … anomalies … are immediately obvious. They (2001) and other scholars, arises in the modeling typically encourage research until either the new approach itself. Economic modeling entails regard- evidence is proven to be erroneous or the theory ing all differences between the actual model in is amended to accommodate it” (2001: pp. 172- use and the “true” model as “noise;” that is, the 173). In economics, in contrast, “anomalies, par- modeling approach entails seeing all omitted ex- ticularly those that cut across the sub-disciplines planatory variables as having no non-random in- and that can be studied with various levels of tech- fluence on the variable being explained. nical sophistication, are tolerated on a scale that John Sutton’s cogent monogra ph (2000) ex- would be impossible in most natural sciences— plains this problem using Alfred Marshall’s anal- and would be regarded as a scandal if they were.” ogy of the tides. Tides are affected by two forces, And, since economics lacks truly stable quantita- the gravitational force of the sun and moon, which tive relations—widely accepted, precise empiri- can be modeled with precision, and meteorologi- cal regularities—the discipline instead unifies cal factors, which are famously difficult to pre- around theoretical commonalities. This, in turn, dict. Fortunately, the meteorological factors are says Lipsey, fosters a “lack of communication of secondary importance, so tides can be pre- among economists operating at various levels of dicted using only the gravitational forces as the theoretical abstraction and empirical sophistica- explanatory variables, and treating local weather tion,” which “causes them to be unaware of many as truly random influences, having no systematic of these anomalies, so their existence often does effect upon the tides (Sutton 2000: pp. 4-5). Eco- not induce research to resolve them” (2001: p. nomic modeling, which proceeds on Marshall’s 173). analogy, likewise assumes that any omitted vari- The disadvantages of empirical work in social ables only randomly perturb the model’s predic- science compared to the natural sciences are real tions. The worry, which is more acute in complex enough. But even if we do not require of econom- settings, is that the omitted variables do system- ics the kind of scientific success demanded in the atically influence the variable being explained, but natural sciences, when do we judge an economic that they are “unobservable.” That is, we cannot model successful? Solow proposes that success measure or proxy or otherwise control for them consists in explaining what the data show. He (Sutton 2000: p. 8). When the omitted variables means “explain” not fundamentally but pragmati- do matter but cannot be measured or proxied or cally, in the sense of capturing a relationship be- controlled for, Marshall’s tides analogy does not tween economic variables to “a fair degree of ap- hold, and more data will not be sufficient for sci- proximation” (1997: p. 49). entific success. This definition of success leads to what can be called the problem of model choice. Since all Prredictedictioededictiictionsionsonsns a bout Methods in EconomEconomicsics models are radical simplifications, even the most What does the future hold for “fact-driven successful models will not fit all the known facts model building”? And what will be the fate of pure of a given situation. Because there is no ideal theory? In 1991, pure theorists were not sanguine model, there often are, instead, two or more mod- about the prospects for pure theory. The distin- els—often quite different—that fit the facts guished theorist Frank Hahn predicted the demise equally well. When the evidence does not choose of deductive, axiomatic theorizing as currently between rival models, “it can become very diffi- practiced. “Theorizing of the ‘pure’ sort will be- cult to ever displace an entrenched model by a come both less enjoyable and less and less pos- better one.” “Clever and motivated … people can sible … It is not my contention that it will wither fight a rear-guard battle that would make Robert under the scorn of practical men or women. The E. Lee look like an amateur,” quips Solow (1997: reasons for the demise are all ‘internal’ to the p.50). The result is that “old models never die, they theory itself.… [T]here will be an increasing re- just fade away” (ibid). alization by theorists that rather radical changes In principle, more data can make the problem in questions and methods are required if they are of model choice less acute. Then the question is to deliver, not practical, but theoretically useful “merely” whether data good enough to meaning- results” (Economic Journal 1991: p. 47). fully discriminate among rival models can be had, Why? Hahn argues that none of the emerging given resource constraints and disciplinary incen- crucial questions “can be answered by the old

ECONOMICS AT THE MILLENNIUM 29 30 SOCIETY • NOVEMBER / DECEMBER 2002 procedures.” Complexity and multiple equilibria frequently be validated through mathematics” are the culprits. Because questions have become (2000: p. 144). increasingly complex, computer simulations will The scholars in our sample frequently believe be needed instead of theorems. Increasingly, “his- that a firmer empirical foundation for economics torical modes of analysis will eventually seem to is desirable, and many, including Solow and Lipsey, be unavoidable”, to pin down which path a sys- argue that this requires a better dialogue between tem with multiple equilibria will actually follow. models and the “facts” they are built to explain. “Instead of simple transparent axioms there But how will this empirical dialogue be mediated? looms the likelihood of psychological, sociologi- Some predict that relatively new empirical tech- cal and historical postulates” (1991, p. 47). Our niques—simulatio n and experiment especially— successors, says Hahn, “will have to bring to the will become as common as today’s statistical in- particular problems they will study particular his- ference from historical data. tories and methods capable of dealing with the Wiseman (Economic Journal 1991) predicts complexity of the particular, such as computer that current econometric techniques will disap- simulation. Not for them the grand unifying theory pear “or become marginalised, being displaced by of particle physics … or the pleasures of theo- the more general use of experimental methods” rem and proof. Instead the uncertain embrace of (1991: p. 153). Colander ( Journal of Economic history and sociology and biology” (p. 50) . It is no Perspectives 2000) foresees economics in 2050 small irony that purely theoretical developments as more plural in its empirical method: “In 2050 have underscored the importance of history to … simulation models … form the core of what economic processes . students are taught.… Economis ts … do empiri- Decision theorist Peter Fishburn agrees with cal work in a wider variety of ways.… The y both Hahn that behavioral axioms will be brought un- create data and analyze it. Experimental econom- der greater scrutiny: “researchers will continue to ics is now an extremely important way of creat- axiomatise new models … but the status of ing data; interestingly, it only began in the late axiomatising will diminish. At the same time, ex- 20th century. Economists today also use natural perimental research on decision behaviour in experiments and randomized field trials to create laboratory and field will flourish. A much better data much more than they did earlier” (pp. 128- understanding of risk typology, attitudes toward 129). ambiguity, and the effects of time on preferences Schmalensee’s ( Economic Journal 1991) mea- will emerge” ( Economic Journal 1991: p. 29). sured discussion sees economists increasingly The distinguished econometrician E. Malinvaud relying on laboratory experiments as a means for sees the decline of very general theories in favor empirically reducing the theoretical indeter- of “a richer system of theoretical models.” He sees minancy of multiple equilibria, especially in stra- such systems as “constellations of specific mod- tegic settings. “The experimental approach offers els … for dealing with some particular aspects of a way to circumvent serious limitations on the phenomena.… Very general models no longer suf- availability of micro-data and seems particularly fice for the more specific questions we have to well suited for testing the implications of strate- consider” ( Economic Journal 1991: p. 67). gic behavior. Progress in computer software and Not every economist views as desirable a re- hardware seems likely to reduce the cost of ex- treat from the mathematical virtues of generality, periments. Unless future research reveals that abstraction, and logical coherence. The theorist laboratory experiments have fatal flaws, I would Beth Allen ( Journal of Economic Perspectives expect them to be routinely used in a number of 2000) laments that “many theorists are now back- fields of economics” (1991: pp. 116-117). tracking from rigor in their work.” Allen argues Alvin Roth ( Economic Journal 1991), an ex- that “economics would be better served if theo- perimentalist, makes an even stronger case for the rists would more often deliberately move in the empirical virtues of laboratory games. Roth warns direction of abstraction and generality, which is that game theory will become purely scholastic where theory can most effectively contribute to without experiments “directed primarily at test- economic science” (2000: p. 145). Allen also seems ing and developing economic theory” (1991: unconcerned with improving the dialogue be- p.108). “If we do not take steps in the direction tween theoretical and empirical work: “much out- of adding a solid empirical base to game theory,” standing theory is inherently untestable, but it can argues Roth, “but instead continue to rely on game

ECONOMICS AT THE MILLENNIUM 31 theory primarily for conceptual insights … it is to plant crops that won’t burn, or to relocate the likely that … game theory will have experienced crops. sharply ” (ibid). Whatever its influence on legal scholarship, economics did not intend to colonize Law. If the Predictiedictionsons a bout the ScopeScope of Economics movement is evidence of eco- Having considered the future of economic nomics imperialism, it is a peculiar kind—acci- method, what can we forecast about the scope of dental imperialism. Whether or not economics has economics, the breadth of its topical interests? been imperialist, accidentally or otherwise, we Prewar economics was best defined not by its predict that the discipline will not abandon its method but by its interests, a loose collection of traditional (prewar) areas of interest, nor will it fields within its purview—governmen t finance, retreat from those sister disciplines it has more railroads, utility regulation, money and banking, recently influenced. At the same time, we think industrial relations, etc. Postwar economics, in that economics will come to be seen as less im- contrast, is better defined by its method and ideas, perialist. models built upon maximization and equilibrium. The first reason for this forecast is mere trend At the same time, the very generality of modern extrapolation. The last century has witnessed dra- economics has not only unified procedures in the matic changes in economic method and ideas, but sub-fields of economics, it has also influenced exhibits a striking continuity of research interests. adjoining disciplines. The spread of economic “There has been,” claims Milton Friedman, “little method and ideas has led to the charge that eco- change in the major issues occupying the atten- nomics has imperial designs on adjoining disci- tion of economists; they are much the same as … plines. dealt with” ( Economic Journal 1991: The imperialism charge has some merit, at least p. 37). Leonard (in Roger Backhouse and Jeff circumstantially. In the last 30-40 years, econom- Biddle (eds), Toward a History of Applied Eco- ics has influenced, sometimes considerably, Law nomics, History of , Supple- (Law & Economics), History (), Poli- ment to Vol. 32, 2000) finds that the debate over tics ( and positive political legal minimum , a topic in Anglo-American economy) and Sociology (), economics for over 150 years, shows a striking to say nothing of Finance. But in other respects, continuity in the issues of interest and the policy the charge does not hold up. positions taken. The future is likely to resemble Law & Economics, and one its founders, Nobel the past in this respect: there will be changes in Laureate , provides an illustration. the tools of economic analysis and the nature and Coase has proved extremely influential in Law. But quality of empirical evidence, but the issues will Coase claims he never meant to influence legal remain much the same. scholarship. In making an argument aimed at What is likely to change are the motives be- economists, he ultimately changed the way legal hind the interest of economics in adjoining disci- scholars regard the notion of cause. He argued plines. Some of the past generation’s outward (in “The Problem of ” Journal of Law expansion in economic methods and ideas can and Economics 1960) that external costs—harms rightly be regarded as analytical tools in search incurred by parties external to an economic ex- of fresh research applications. We predict that the change—are jointly caused. The common law tra- relationship between economics and its sister ditionally assigned legal liability for harms by as- disciplines will move closer to one of mutually suming causation is one-way—railroad sparks beneficial trade. Economics will look to nearby cause fires in nearby crops, so railroads should disciplines less for new applications and more for be made liable. Coase insisted that, in the absence new ideas. And, as with all mutually beneficial of flammable crops planted close to the rail bed, exchange, the idea of foreignness will be eroded: there is no harm. Thus, farmers also cause the some of the current disciplinary boundaries will harm—it takes two to tort. And if “cause” in the be effaced and redrawn. traditional, one-way sense is no longer legally de- Part of the reason economics is likely to increas- cisive, the door is open to other adjudicative cri- ingly look outside the discipline for inspiration is teria for deciding who should be made to bear sociological: like other invaders, economists social costs. Economists of course suggest effi- abroad are likely to go native, as Schmalensee ciency as a criterion: it may be socially cheaper (Economic Journal 1991) notes. But much of the

32 SOCIETY • NOVEMBER / DECEMBER 2002 outward-looking impetus comes from within eco- tion of how real individuals actually do go about nomics itself: the canonical ideas of late 20th-cen- making decisions. tury economics have, particularly in the last 10- Half a century after Simon first made this argu- 15 years, come under increasing theoretical and ment, economics has begun to take notice. Begin- empirical strain. ning a few years after Becker’s manifesto, a small group of economists—who have adopted the Predictiedictionsons a boboutut EconomEconomics’ics’ CCanoniCanoanonicanicalcall Ideas name of “”— became con- Nobel laureate wrote in 1976 (in vinced that the weight of experimental and other The Economic Approach to Human Behavior ) empirical evidence was disconfirming of the maxi- that “the combined assumptions of maximizing mization hypothesis. Psychologists Amos Tversky, behavior, market equilibrium and stable prefer- Daniel Kahnemann and others offered evidence ences, used relentlessly and unflinchingly, form that homo sapien s does not much resemble his the heart of the economic approach as I see it” idealized maximizing homo economicus cousin— (1976: p. 5). Becker adds stable preferences to homo sapiens makes systemic errors in his deci- maximization and equilibrium because it facili- sions, relies on rules of thumb instead of calcula- tates dynamic maximization, determining an op- tion, reverses choices in response to different timal series of choices over time. All three assump- framing of the same question, and discounts the tions are being weakened, a trend that we are sure near future more steeply than the remote future. will continue. Behavioral economics tries to capture these Criticism of the maximization hypothesis as and other empirically documented departures unrealistic is longstanding. Nobody believes that from idealized rationality, what (in ordinary human beings are computational prodi- Journal of Economic Perspectives 2000), godfa- gies who routinely use sophisticated mathemat- ther of behavioral economics, characterizes as ics to make decisions. Economists traditionally homo economicus losing IQ. The overarching reply that it is as if agents decide by setting up modeling principle is what Simon called bounded and solving constrained maximization problems. rationality: the view that cognitive and time scar- Imagine a consumer who operates in a world with city make solving for optima impossible or too two , x1 and x2, who has a budget of I, and costly, so that agents have incentives to look for who faces of p1 and p2. Assume that the alternative methods of making choices—“heuris- consumer spends fraction a of her budget on the tics”, rules of thumb, etc. A major theoretical fo- first good, and fraction (1- a) on the second good. cus is on how cognitively constrained individu- The consumer’s decision can be seen as apply- als decide how to decide (Plott, Economic ing a rule of thumb—always spend aI on good 1. Journal 1991: p. 91). Her decision can also be presented as if it were One ironic unexpected outcome is that eco- the solution to a constrained optimization prob- nomic theorists, currently remote from behavioral a 1-a lem: maximizing the utility function x1 x2 sub- economics, may increasingly turn to the more ject to the constraints presented by her limited practical but mathematically demanding problems budget and prices, I = p1 x1 + p2 x2 . It is as if the of bounded rationality. One example involves consumer who applies a spending rule of thumb methods for determining good choices where the solves a constrained utility maximization problem. best-possible (that is, “optimal”) choice is not com- (The example is taken from Rubinstein 1998: p. putable, as in traveling-salesman-type problems. 10). Since both decision rules yield identical re- For concreteness, consider a traveling salesman sults, maximization can be seen as a useful fic- who must map out a route which includes visits tion. to N cities. The optimal solution involves mini- Herbert Simon, a Nobel Laureate, has long ar- mizing travel time. But the number of possible gued that the maximization hypothesis, and the routes grows so quickly as N increases that the as if defense, present two problems. First, there optimization problem becomes effectively unsolv- are good economic reasons against the maximi- able. With N cities to visit, there will be N! dis- zation hypothesis: cognitive resources are scarce, tinct possible alternative routes. For instance, if as is decision-making time. Economists, who lo- N =30, N! is approximately 2.65*10 32. Evaluating cate scarcity at the very center of economic rea- that number of alternatives would keep any cur- soning, should not assume cognitive free lunches. rent supercomputer busy for eons. (The example Second, the as if method begs the scientific ques- is from a 2001 paper by Xavier Gabaix and David

ECONOMICS AT THE MILLENNIUM 33 Laibson, forthcoming in I. Broca and J. Carillo eds., butions, the relevant characteristics of which Collected Essays in Psychology and Economics ). (means, variances, etc.) are assumed to be known A second prong of the behavioral economics to agents in the economy. By any standard this is program concerns maximizing over time—a set- a restrictive representation of the issue” ( Eco- ting where choice and its consequences may be nomic Journal, 1991: p. 145). separate in time. The issues here are the stability The future thus presents two forms of uncer- of one’s preferences as one matures, and uncer- tainty—uncertai nty regarding what will happen, tainty about how to characterize an unknown and uncertainty as to how one’s future welfare future. The Beckerian program assumes stable might be affected by each possible outcome. The preferences, which, among other things, is tanta- two forms of uncertainty are complementary and mount to assuming unbounded willpower. On the reinforcing. Consider the choice to smoke. First, behavioral view, even when real human beings there is uncertainty about what will happen—for know what is best, they will sometimes fail to example, the likelihood and timing of negative choose it, for lack of sufficient willpower. health consequences. Second, there may be un- If, in the morning, you prefer not to eat dessert certainty as to how one’s future welfare will be after dinner, then, with stable preferences, you will affected by these possible future events. One still prefer no dessert rather than ice cream after might, for example, suspect that one’s own pref- dinner. You will not be tempted. How then to ex- erences are likely to change with age: the typical plain the enormous self-control industry? Billions 40-year-old does not value activities, goods and are spent on fat farms, diet plans, smoke-ending experiences the same way he did at age 20. Or clinics, drug rehabilitation, and fitness trainers. future preferences might be affected by previous Moreover, there are less formal self-control de- of a habit-forming good, such as to- vices, such as buying cigarettes by the pack rather bacco. When there is uncertainty as to how one’s than the carton, avoiding streets with taverns, future self will be affected by future events which keeping sweets out of the house, and locating the are themselves uncertain, the theoretical tempta- alarm clock across the room. Behavioral econo- tion to revert to stable preferences is understand- mists explore theoretical assumptions consistent able. The experimentalist Charles Plott agrees that with temptation. Most current work explains “the nature of individual choice of process, a set temptation via the assumption that rates of impa- of rules and institutions that will operate when tience (time preference) are not constant but de- one’s preferences are different from those that crease with the duration of the wait. Many alter- exist at the time of choice, will … be a perplex- native ways of weakening the stable-preferences ing challenge” (Plott, Economic Journal, 1991, p. assumption remain to be explored. 91). Because maximization generally requires it, The challenge is more than one of theoretical economists ordinarily treat uncertainty only as tractability. Mutable preferences go to the heart risk. In risky situations, agents are presumed to of the policy enterprise of comparing the wel- know the relative likelihoods and payoffs of all fare of individuals before and after a policy change possible outcomes, as with a roulette wheel. In (such as, say, an increase in cigarette excise taxes). risky settings, economic agents don’t know what If self “one” has different preferences than future will happen, but they do know everything that self “two,” which self’s welfare should policy rec- could happen. True uncertainty, a more serious ognize, and on what basis? We hazard no predic- kind of indeterminacy, arises when agents lack a tion on how the analysis of this crucial problem complete probabilistic picture of the future. will evolve, but we will wager that, in the next The analysis of the unknown remains an im- 10-20 years, the Nobel Prize will be awarded to a portant frontier in economics. Some regard the behavioral economist. roulette-wheel conception—which tames uncer- Finally, the trajectory of behavioral economics tainty into risk—as misbegotten. Wiseman, for tells us something about the relationship between example defines the future as “unknowledge,” that theory and evidence in economics. For years, which “contain events which we cannot foresee” economists ignored the psychologists’ experimen- (Economic Journal 1991: p. 152). Less critical tal findings of departures from the maximizing observers, like Turnovsky, agree that “our analyti- paradigm. Richard Thaler’s work, which empha- cal treatment of uncertainty is pretty primitive. sized these empirical anomalies, was once dis- Typically it is represented by probability distri- missed as mere anecdote. Abandoning the canoni-

34 SOCIETY • NOVEMBER / DECEMBER 2002 cal principles was criticized as entailing a descent have little or no use for laws, norms and conven- into chaotic ad-hockery; a fearful if non-specific tions. The upshot is that economics came to be disease that one might find in, say, the salons of seen as somehow opposed to explaining legal sociology (Kreps Daedalus 1997: p. 74). Eventu- rules, social norms and conventions. ally though, the weight of the evidence proved But then a funny thing happened. Institutions disconfirming enough to induce some economists reappeared. The causes are multifarious and un- to risk ad hockery, and explore theoretical alter- coordinated. We suggest that the different ways natives. That Thaler’s work also turned out to have in which researchers have relaxed the canonical concrete testable implications in finance also principles has, intentionally and as a byproduct, helped. Ten years later, Alvin Roth’s forecast looks helped revive the importance of laws, norms and prescient: “ … will play conventions in economic analysis. We offer four an important role in helping game theory bridge examples. the gap between the study of ideally rational be- New Institutional economists, for example, havior and the study of actual behavior” ( Eco- explicitly model legal property and contract nomic Journal 1991: p. 107). rights. Because there is uncertainty, contracts that Thaler and confreres do not intend to over- specify every possible future contingency are throw economic method. The behavioral econo- impossible, and it is this incompleteness of con- mists believe only that their behavioral postulates tracts that creates a rationale for contract law offer a better explanation for the data than does enforcement. In that other former world of maxi- the maximization hypothesis, which Thaler pre- mizing agents with complete information, con- dicts will eventually come to be seen as a special tracts will be perfectly complete, and contract law case. Behavioral economists believe that people is superfluous. are purposive and forward looking, but imperfect, and that their departures from perfection lead to interesting and sometimes unexpected conse- American economiceconomicss and Americanican quences. In our final section, we consider one sociology sharsharee common descent, such consequence, what we call “revenge of the the fields divererged aroundound norms.” the FFiirst WWorldorld WWaarr.. RReevengeenngge of the NorNor msms American economics and American sociology Laboratory games show that experimental sub- share common descent. When the fields diverged jects are concerned with fairness—who gets around the First World War, among the several what—and are willing to spend their own money causes was a different conception of human ac- to enforce norms of fairness. While there is noth- tion. Jon Elster ( Journal of Economic Perspectives ing in the canonical principles that requires ego- Fall 1989) calls it one of the deepest conceptual ism, self-interested action has nonetheless been a cleavages in social science: that between homo commonplace assumption of late 20th-century economicus, a creature who is individualistic, economics, and fairness, or “bounded self-interest,” purposeful and forward looking, and homo has emerged as one of the research areas of be- sociologus, who is social, conventional and some- havioral economics. times myopic. Thus could say and others study common-prop- 40 years ago: “Economics is all about how people erty resource settings which have a prisoner’s- make choices; sociology is all about why they dilemma structure, that is, payoff-maximizing in- don’t have any choices to make” (in Demographic dividual choices lead to collectively inferior and Economic Change , National Bureau of Eco- results (a.k.a. the ). nomic Research, 1960, p. 233). Ostrom finds in experiments and in field stud- As 20th-century economic theory converged ies that some communities successfully evolve on the maximization-equilibrium-stable prefer- and enforce norms against opportunistic behav- ences paradigm, it became increasingly silent on ior. social structures. Institutions such as laws, norms, Game theory has proven to be an unexpected and conventions were ordinarily treated as exog- source of insight into the nature and function of enous, beyond analysis. The reason was simple: conventions. It is no coincidence that game theory maximizing agents with complete information is a place where the weaknesses of both canoni-

ECONOMICS AT THE MILLENNIUM 35 cal assumptions—maximization and equilib- ver), drive on the right (or left) side of the road, rium—have been exposed. The key problem is determine price by haggling (or by posting or by that of multiple equilibria. auction), use standard-form contracts, etc. Conven- John Nash shared a Nobel Prize for his tions enable coordination of expectations when dissertation’s proof that in non-cooperative games deduction by itself is insufficient. an equilibrium exists. But even if we can expect It has long been known that the coordinating players to select Nash equilibria, which of the function of conventions is valuable. What we now multiple Nash equilibria will they pick? In the better understand is that there is necessity be- 1970s and 1980s, an equilibrium selection litera- hind their virtue. It is because real people are ture sought ways to reduce the number of Nash cognitively constrained, that they, unlike their equilibria. But this Nash refinements literature still idealized cousins, have incentives to look for hewed closely to game theory’s traditional em- and learn conventions that can make them bet- phasis on wholly deductive approaches: what ide- ter off. ally rational players, given only a complete de- The revival of economics’ interest in laws, scription of the game, must deduce. norms and conventions is by itself not enough, of Paradoxically, this Nash refinements literature course, to reunite homo economicus and homo produced too many solution concepts —how is sociologus, who were separated at birth. But the the agent to choose among multiple theories of revenge of the norms in economics is a scientifi- equilibrium selection? In response, in the 1990s cally welcome effect of the ongoing process some game theorists moved away from these whereby economics reduces the IQ of homo wholly deductive approaches, instead going to- economicus, thereby sometimes making him ward evolutionary models. In evolutionary mod- smarter. els, adapted from theoretical biology, players are boundedly rational: they have incomplete infor- mation, limited memory and simple conceptions SUGGESTED FURTHER READINGS of how others are likely to behave. The process of to an equilibrium is dynamic. Play- The Economic Journal January 1991 contains short ers grope towards an equilibrium, more pushed articles by twenty-two economists discussing the by a changing environment, than pulled by their future of economics. Among the especially inter- own deductive prowess. esting ones are those by Baumol, Buchanan, Fried- One upshot is that social conventions have man, Pencavel, Plott, Roth and Schmalensee. become relevant again. Traditional game theory The Journal of Economic Perspectives Winter 2000 ignored conventions as superfluous: a uniquely ra- contains a Symposium “Forecasts for the Future tional solution obviates the need for conventions, of Economics.” Of particular interest are the ar- so players who observe conventions cannot be ticles by Colander and Thaler. ideally rational, and, conversely, should it prove Lipsey, Richard. “Successes and Failures in the Trans- rational to follow a convention, then the claim formation of Economics,” Journal of Economic of a uniquely rational solution must be false. Methodology June 2001, pp. 169-201. The hyper-rational agent of traditional game Solow, Robert. “How Did Economics Get That Way, and theory—who can rely upon deduction alone— What Way Did It Get?” Daedalus , 126, Winter 1997, fails at some rudimentary tasks of coordination pp. 39-58. that homo sapiens manages rather well. Consider Sutton, John. Marshall’s Tendencies: What Can Econo- a coordination game, where two randomly mists Know? Cambridge, Mass.: MIT Press, 2000. paired traders can use one of two , gold or silver. If both play gold, each gets a Robert S. Goldfarb teaches in the Economics De- payoff of one; if both play silver, each gets a partment at George Washington University. Tho- payoff of one. If they fail to coordinate, each mas C. Leonard teaches in the Economics Depart- gets zero. Because there are multiple Nash equi- ment at Princeton University. In addition to their libria (e.g., both play gold, both play silver), the research on economic methodology and the his- superhuman agent of classical game theory can- tory of economic thought, Goldfarb and Leonard, not decide what to do. along with their co-author Stephen Suranovic, But real people can. They coordinate by re- have also published several articles on the econom- course to convention. They trade in gold (or sil- ics of addiction.

36 SOCIETY • NOVEMBER / DECEMBER 2002