INFORMATION AND THE CHANGE IN THE PARADIGM IN ECONOMICS,PART 1

by Joseph E. Stiglitz*

The research for which George Akerlof, Mike Spence, and I are being recognized is part of a largerresearch program which, today, embraces hundred, perhaps thousands, of researchers around the world.In this lecture, I want to set the particular work which was sited within this broader agenda, and thatagenda within the broader perspective of the history of economic thought. I hope to show that Informa-tion Economics represents a fundamental change in the prevailing paradigm within economics. Prob-lems of information are central to understanding not only market economics but also political economy,and in the last section of this lecture, I explore some of the implications of information imperfectionsfor political processes.

INTRODUCTION

Many years ago Keynes wrote:

The ideas of economists and political philoso-phers, both when they are right and when theyare wrong, are more powerful than is com-monly understood. Indeed, the world is ruledby little else. Practical men, who believethemselves quite exempt from any intellectu-al influences, are usually the slaves of somedefunct economist. Madmen in authority, whohear voices in the air, are distilling their fren-zy from some academic scribbler of a fewyears back. Keynes [1936].

Information economics has already had a pro-found effect on how we think about economic poli-cy, and is likely to have an even greater influence inthe future. The world is, of course, more complicat-ed than our simple-or even our more complicatedmodels-would suggest. Many of the major politi-cal debates over the past two decades have centeredaround one key issue: the efficiency of the marketeconomy, and the appropriate relationship betweenthe market and the government. The argument ofAdam Smith [1776], the founder of modem eco-nomics, that free markets led to efficient outcomes,"as if by an invisible hand" has played a central rolein these debates: it suggested that we could, by andlarge, rely on markets without government interven-

tion. There was, at best, a limited role for govern-ment. The set of ideas that I will present here under-mine Smith's theory and the view of governmentthat rested on it. They have suggested that the rea-son that the hand may be invisible is that it is sim-ply not there-or at least that if is there, it is palsied.

When I began the study of economics some fortyone years ago, I was struck by the incongruitybetween the models that I was taught and the worldthat I had seen growing up, in Gary Indiana, a citywhose rise and fall paralleled the rise and fall of theindustrial economy. Founded in 1906 by U.S. Steel,and named after its Chairman of the Board, it haddeclined to but a shadow of its former self by theend of the century. But even in its heyday, it wasmarred by poverty, periodic unemployment, andmassive racial discrimination. Yet the theories thatwe were taught paid little attention to poverty, saidthat all markets cleared-including the labor mar-ket, so unemployment must be nothing more than aphantasm, and that the profit motive ensured thatthere could not be economic discrimination.' If thecentral theorems that argued that the economy wasPareto efficient-that, in some sense, we were liv-ing in the best of all possible worlds-were true, itseemed to me that we should be striving to create adifferent world. As a graduate student, I set out totry to create models with assumptions-and con-clusions-closer to those that accorded with theworld I saw, with all of its imperfections.

* 2001 Nobel Laureate, Columbia Business School, Columbia University, 1022 International Affairs,Building, 420 West 118th Street, New York, NY 10027, USA.

Due to the length of the essay, the first part is published in the Fall issue, and the second part, whichincludes all the notes and references, will be published in the Spring 2004 issue. A complete set of notesand references is also available by contacting Richard Nilsen at e-mail address rn71238@pace.edu

THE AMERICAN ECONOMIST6

My first visits to the developing world in 1967,and a more extensive stay in Kenya in 1969, left anindelible impression on me. Models of perfect mar-kets, as badly flawed as they might seem for Europeor America, seemed truly inappropriate for thesecountries. But while many of the key assumptionsthat went into the competitive equilibrium modelseemed not to fit these economies well, the onesthat attracted my attention were the imperfection ofinformation, the absence of markets, and the perva-siveness and persistence of seeming dysfunctionalinstitutions, like sharecropping. With workers hav-ing to surrender 50% or more of their income tolandlords, surely (if conventional economics werecorrect), incentives were greatly attenuated. Tradi-tional economics said not only that institutions (likesharecropping)2 did not matter, but neither did thedistribution of wealth. But if workers owned theirown land, then they would not face what amountedto a 50% tax. Surely, the distribution of wealth didmatter.

I had seen cyclical unemployment-sometimesquite large-and the hardship it brought as I grewup, but I had not seen the massive unemploymentthat characterized African cities, unemploymentthat could not be explained either by unions or min-imum wage laws (which, even when they existed,were regularly circumvented). Again, there was amassive discrepancy between the models we hadbeen taught and what I saw.

The new ideas and models were not only usefulin addressing broad philosophical questions, suchas the appropriate role of the state, but also in ana-lyzing concrete policy issues. In the 1970s, econo-mists became increasingly critical of traditionalKeynesian ideas, partly because of their assumedlack of micro-foundations. The attempts made toconstruct a new macro-economics based on tradi-tional micro-economics, with its assumptions ofwell functioning markets, were doomed to failure.Recessions and depressions, accompanied by mas-sive unemployment, were symptomatic of massivemarket failures. The market for labor was clearlynot clearing. How could a theory that began withthe assumption that all markets clear ever providean explanation? If individuals could easily smooththeir consumption by borrowing at safe rates ofinterest, then the relatively slight loss of lifetimeincome caused by an interruption of work of sixmonths or a year would hardly be a problem; but theunemployed do not have access to capital markets,

at least not at reasonable terms, and thus unemploy-ment is a cause of enormous stress. If markets wereperfect, individuals could buy private insuranceagainst these risks; yet it is obvious that they can-not. Thus, one of the main developments to followfrom this line of research into the consequences ofinformation imperfections for the functioning ofmarkets is the construction of macro economicmodels that help explain why the economy ampli-fies shocks and makes them persistent, and whythere may be, even in competitive equilibrium,unemployment and credit rationing.

I believe that some of the huge mistakes whichhave been made in policy in the last decade, in forinstance the management of the East Asia crisis orthe transition of the former communist countries tothe market, might have been avoided if there hadbeen a better understanding of issues, like bank-ruptcy and corporate governance, to which the newinformation economics called attention. And the so-called Washington consensus policies3, which havepredominated in the policy advice of the interna-tional financial institutions over the past quartercentury, have been based on market fundamentalistpolicies which ignored the information-theoreticconcerns, and this explains at least in part theirwidespread failures.4

Information affects decision making in everycontext-not just inside firms and households.More recently, I have turned my attention to someaspects of what might be called the political econo-my of information: the role of information in politi-cal processes, in collective decision making. Fortwo hundred years, well before the economics ofinformation became a subdiscipline within econom-ics, Sweden had enacted legislation to increasetransparency. There are asymmetries of informationbetween those governing and those governed, andjust as markets strives to overcome asymmetries ofinformation, we need to look for ways by which thescope for asymmetries of information in politicalprocesses can be limited and their consequencesmitigated.

THE HISTORICAL SETTING

I do not want here to review and describe indetail the models of information asymmetries thathave been constructed in the past thirty years. Inrecent years, there have been a number of survey

Vol. 47, No. 2 (Fall 2003) 7

articles,' even several books,6 and interpretativeessays.'7 I do want to highlight some of the dramat-ic impacts that information economics has had onhow economics is approached today, how it hasprovided explanations to phenomena that were pre-viously unexplained, how it has altered our viewsabout how the economy functions, and, perhapsmost importantly, how it has led to a rethinking ofthe appropriate role for government in our society.In describing the ideas, I want to trace out some oftheir origins: to a large extent, they were responsesto attempts to answer specific policy questions or toexplain specific phenomena to which the standardtheory provided an inadequate explanation. But anydiscipline has a life of its own, a prevailing para-digm, with assumptions and conventions. Much ofthe work was motivated by an attempt to explorethe limits of that paradigm-to see how the stan-dard models could embrace problems of informa-tion imperfections (which turned out not to be verywell.)

For more than a hundred years, formal modelingin economics has focused on models in which infor-mation was perfect. Of course, everyone recognizedthat information was in fact imperfect, but the hope,following Marshall's dictum "Natura non facitsaltum," was that economies in which informationwas not too imperfect would look very much likeeconomies in which information was perfect. Oneof the main results of our research showed that thiswas not true; that even a small amount of informa-tion imperfection could have a profound effect onthe nature of the equilibrium.

The reigning paradigm of the twentieth century,the neoclassical model, ignored the warnings of thenineteenth century and earlier masters on howinformation concerns might alter the analyses, per-haps because they could not see how to embracethem in their seemingly precise models, perhapsbecause doing so would have led to uncomfortableconclusions about the efficiency of markets. Forinstance, Smith, in anticipating later discussions ofadverse selection, wrote that as firms raise interestrates, the best borrowers drop out of the market. Iflenders know perfectly the risks associated witheach borrower, this would matter little; each bor-rower would be charged an appropriate risk premi-um. It is because lenders do not know the defaultprobabilities of borrowers perfectly that thisprocess of adverse selection has such importantconsequences. 8

I have already noted in the introduction thatsomething was wrong-seriously wrong-with thecompetitive equilibrium models that representedthe prevailing paradigm when we went to graduateschool. It seemed to say that unemployment didn'texist, that issues of efficiency and equity could beneatly separated, so that economists could neatly setaside problems of inequality and poverty as theywent about their business of designing more effi-cient economic systems. But there were a host ofother predictions, empirical puzzles, that were hardto reconcile with the standard theory: in micro-eco-nomics, there were tax paradoxes such as why didfirms seemingly not take actions which minimizedtheir tax liabilities, security market paradoxes, suchas why did asset prices seem to exhibit such highvolatility,9 and behavioral puzzles, such as why didfirms respond to risks in ways which were marked-ly different from that predicted by the theory.'° Inmacro-economics, the cyclical movements of manyof the key aggregate variables, such as consump-tion,"1 inventories,'2 real product wages,"3 real con-sumption wages,'4 and interest rates'5 are hard toreconcile with the standard theory, and if the perfectmarket assumptions were even approximately satis-fied, the distress caused by cyclical movements inthe economy would be much less than seems to bethe case.'6

The problems that we saw with the models thatwe were taught was not only that they seemedwrong, but that they left a host of phenomena andinstitutions unexplained-why were IPO's typical-ly sold at a discount? Why did equities, which pro-vided far better risk diversification than debt, playsuch a limited role in financing new investment?' 7

There were, to be sure, some Ptolemaic attemptsto defend and elaborate on the old model. Some,like George Stigler,'8 while recognizing the impor-tance of information, argued that once the real costsof information were taken into account, even withimperfect information, the standard results of eco-nomics would still hold. Information was just atransactions cost. In the approach of many Chicagoeconomists, information economics was like anyother branch of applied economics; one simply ana-lyzed the special factors determining the demandand supply for information, just as agricultural eco-nomics analyzed those factors affecting the marketfor wheat. For the more mathematically inclined,information could be incorporated into productionfunctions of, say, goods by inserting an "I" for the

THE AMERICAN ECONOMIST8

input "information," and "I" itself could be pro-duced by inputs, like labor. Our analysis showedthat that this approach was wrong, as were the con-clusions derived from it.

Practical economists who could not ignore thebouts of unemployment which had plagued capital-ism since its inception talked of the neoclassicalsynthesis: using Keynesian interventions to ensurethat the economy remained at full employment, andonce that was done, the standard neoclassicalpropositions would once again be true. But whilethe neoclassical synthesis"9 had enormous intellec-tual influence, by the 1970s and 80s it came underattack from two sides. It was an assertion, not basedon a coherent view of the economy. One sideattacked the underpinnings of Keynesian econom-ics, its micro-foundations; why would rationalactors be out of equilibrium-with unemploymentpersisting-in the way that Keynes had suggested.This side effectively denied the phenomena whichKeynes was attempting to explain.

Worse still, some saw unemployment as largelyreflecting an interference (e.g. by government insetting minimum wages, or trade unions, in usingtheir monopoly power to set wages too high) withthe free workings of the market, with the obviousimplication: unemployment would be eliminated ifmarkets were made more flexible, that is, if unionsand government interventions were eliminated.Even if wages fell a third in the Great Depression,they should have, in this view, fallen even more.

There was an alternative perspective (articulatedmore fully in Greenwald and Stiglitz, 1987a,1988b): why shouldn't we believe that massiveunemployment was just the tip of the iceberg, ofmore pervasive market efficiencies that were hard-er to detect. If markets seemed to function so badlysome of the time, certainly they must be malper-forming in more subtle ways much of the time. Theeconomics of information bolstered the latter view.

Similarly, given the nature of the debt contracts,the falling prices led to bankruptcy and economicdisruptions, actually exacerbating the economicdownturn. Had there been more wage and priceflexibility, matters might have been even worse.Moreover, neither government nor unions imposedthe limitations on wage and price dynamics in manysectors of the economy; at the very least, those whoargued that the problem was wage and price rigidi-ties had to look for other market imperfections, and

any policy remedy (including a call for greater flex-ibility) had to take those factors into account.

In the next section, I shall explain how it was notjust the discrepancies between the standard compet-itive model and its predictions which lead to itsbeing questioned. The model was not robust-evenslight departures from the underlying assumption ofperfect information had large consequences.

But before turning to those issues, it may be use-ful to describe some of the concrete issues whichunderlay the beginnings of my research program inthis area. Key in my thinking on these issues wasthe time between 1969 and 1971 that I spent at theInstitute for Development Studies at the Universityof Nairobi with the support of the RockefellerFoundation.

Education as a screening device20

The newly independent Kenyan government, asit attempted to forge policies which would promotetheir growth and development, was asking ques-tions that never seemingly been raised by theircolonial masters. How much should it invest in edu-cation? It was clear that a better education got onebetter jobs-the credential put one at the head of thejob queue. Gary Fields, a young scholar working atthe Institute of Development Studies there, devel-oped a simple model2 ' suggesting that the privatereturns to education-the enhanced probability ofgetting a good job-differed from the social return;and that it was possible that as more people get edu-cated, the private returns got higher (it was evenmore necessary to get the credential) even thoughthe social return might decline. Here, education wasperforming a markedly different function than it didin traditional economics literature, where it simplyadded to human capital and improved productivi-ty.22'23 The analysis had important implications forKenya's decision about how much to invest in high-er education. The problem with Fields' work wasthat it did not provide a full equilibrium analysis:wages were fixed, rather than competitively deter-mined.

This led me to ask, what would the market equi-librium look like if wages were set equal to meanmarginal products conditional on the informationthat was available? And this in turn forced me toask: what were the incentives and mechanisms foremployers and employees to acquire or transmitinformation? Within a group of otherwise similar

Vol. 47, No. 2 (Fall 2003) 9

job applicants (who therefore face the same wage),the employer has an incentive to identify who is themost able, to find some way of sorting or screeningamong them, if he could keep that information pri-vate. But he often can't, and if others find out aboutthe true ability, the wage will be bid up, and he willbe unable to appropriate the return to the informa-tion. At the very beginning of this research programwe had thus identified one of the key issues in infor-mation economics, the difficulty of appropriatingthe returns.

On the other hand, the employee, if he knew hisability (that is, if there were asymmetries of infor-mation between the employee and the employer)and he knew that his abilities were above the aver-age of those in the market, he would have an incen-tive to convince the employer of his ability. Butsomeone at the bottom of the ability distributionhad an incentive not have the information revealed.Here was a second principle that was to be exploredin subsequent years: there are incentives on the partof individuals for information not to be revealed,for secrecy, or, in modem parlance, for a lack oftransparency. This raised a question: how did theforces for secrecy and for information disclosureget balanced? What was the equilibrium thatemerged? I will postpone until the next section adescription of that equilibrium.

Efficiency wage theory

That summer in Kenya I began three otherresearch projects related to information imperfec-tions. At the time I was working in Kenya, therewas heavy urban unemployment. My colleagues atthe Institute for Development Studies, MichaelTodaro and John Harris had formulated a simplemodel of labor migration from the rural to the urbansector which accounted for the unemployment.High urban wages attracted workers, and they werewilling to risk unemployment for the chance ofthose higher wages.4 Here was a simple, generalequilibrium model of unemployment, but againthere was one missing piece: how could you explainthe high wages, which were well in excess of theminimum wage? It did not seem as if either gov-emmnent or unions were forcing these high wages.One needed an equilibrium theory of wage determi-nation. I recalled, during an earlier stint at Cam-bridge, discussions with Harvey Leibenstein whohad postulated that in very poor countries, higher

wages lead to higher productivity.25 It might not payfirms to cut wages, if productivity was cut morethan proportionately, even if there was an excesssupply of labor. The key insight was to recognizethat there were a variety of other reasons why, wheninformation and contracting were imperfect, pro-ductivity might depend on wages.29 In that case, itmight pay firms to pay a higher wage than the min-imum necessary to hire labor; such wages I referredto as efficiency wages. With efficiency wages, therecould exist an equilibrium level of unemployment.I explored four explanations for why productivitymight depend on wages (besides nutrition). Thesimplest was that lower wages lead to higherturnover, and therefore the higher turnover costswhich the firm bore.27 It was not until some yearslater that we were able to explain more fully-based on limitations of information-why it wasthat firms had to bear these turnover costs.2 1

But there was another version of the efficiencywage related to the work I was beginning on asym-metric information. Any manager will tell you thatyou attract better workers by paying them higherwages. This was just an application of the generalnotion of adverse selection, which played a centralrole in earlier insurance literature, where firms hadlong recognized that as they charge a higher premi-um, the best risks stopped buying insurance.29 Firmsin a market do not passively have to accept the"market wage." Even in competitive markets, finnscould, if they wanted, offer higher wages than oth-ers. Market clearing was not a constraint on firms.If all finns were paying the market-clearing wage,it might pay a firm to offer a higher wage, to attractmore able workers. The efficiency wage theorymeant that there could exist unemployment in equi-librium.

It was thus clear that the notion that had under-lay much of traditional competitive equilibriumanalysis-that markets had to clear-was simplynot true if information were imperfect.

The formulation of the efficiency wage theorythat has received the most attention over the years,however, has been the one that has focused on prob-lems of incentives. Many firms claim that payinghigh wages induces their workers to work harder.The problem that Carl Shapiro and I [1984] facedwas to try to make sense of this claim. If all work-ers are identical, and were paid the same wage, thenif it paid one firm to pay a high wage, it would payall of them. But if a worker was then fired for shirk-

THE AMERICAN ECONOMIST10

ing, and there was full employment, he couldimmediately get another job, at the same wage. Thehigh wage would provide no incentive. But if therewas unemployment, then there was a price forshirking. We showed that in equilibrium there hadto be unemployment: unemployment was the disci-pline device that forced workers to work?"0 Themodel had strong policy implications, some ofwhich I shall describe below. Our work illustratedthe use of highly simplified models to help clarifythinking about quite complicated matters. In prac-tice, of course, workers are not identical, so prob-lems of adverse selection become intertwined withthose of incentives; being fired does convey infor-mation-there is typically a stigma.

(There was a fourth version of the efficiencywage, where productivity was related to moraleeffects, perceptions about how fairly they werebeing treated. While I briefly discussed this versionin my earlier work,31 it was not until almost twentyyears later that the idea was fully developed, in theimportant work of Akerlof and Yellen [1990].)

Sharecropping and the general theory ofincentives

This work on the economics of incentives inlabor markets was closely related to the thirdresearch project that I began in Kenya. In tradition-al economic theory, while considerable lip servicewas paid to incentives, there was no real incentiveissue. With perfect information, individuals are paidto perform a particular service; if they perform itthey receive the amount contracted for; if they donot perform, they receive nothing. With imperfectinformation, firms have to motivate and monitor,rewarding them for observed good performance andpunishing them for bad. My interest in the issueswas first aroused by thinking about sharecropping,a common form of land tenancy in a developingcountry, where the worker surrenders half (some-times two thirds) of the produce to the landlord inreturn for the use of his land. At first blush, thisseemed a highly inefficient arrangement; it wasequivalent to a 50% tax on workers' labor. But whatwere the alternatives? The worker could rent theland, but that meant he had to bear all the risk offluctuations in output; and besides, he often did nothave the requisite capital. He could work as wagelabor, but that meant that the landlord would have tomonitor him, to ensure that he worked. Sharecrop-

ping represented a compromise, between risk bear-ing and incentives. The underlying informationproblem was that the input of the worker could notbe observed, but only his output, and his output wasnot perfectly correlated with his input. The share-cropping contract could be thought of as a combi-nation of a rental contract plus an insurance con-tract, in which the landlord "rebates" part of the rentif crops turn out badly. There is not full insurance(which would be equivalent to a wage contract)because such insurance would attenuate all incen-tives. The adverse effect of insurance on incentivesto avoid the insured against contingency is referredto as moral hazard.32 In Stiglitz [1974b] I analyzedthe equilibrium sharecropping contract. In thatpaper, I recognized that the incentive problems Iexplored there were isomorphic to those facingmodem corporations, e.g. in providing incentives totheir managers,33 and there followed a large litera-ture on optimal and equilibrium incentive schemes34

in labor, capital, and insurance markets. Contractshad to be based on observables, like processes (orwhich crops were grown) and observable inputs(like fertilizers). Many of the results obtained earli-er in the work on adverse selection had their paral-lel in this area of "adverse incentives."3 5 Forinstance, with Richard Arnott I analyzed the equi-librium [1988a, 1988b], which entails partial insur-ance.

Equilibrium wage and price distributions

The fourth strand of research looked at the issueof wage differentials that I had observed from a dif-ferent perspective. The work on labor turnover hadsuggested that finns that faced higher turnovercosts might pay higher wages. But one of the rea-sons that individuals quit was to obtain a higherpaying job. The turnover rate depended on the wagedistribution. The challenge was to formulate anequilibrium model, in which there was a wage dis-tribution, which led firms to charge differentwages-the distribution of wages that had original-ly been postulated.

More generally, efficiency wage theory said thatit paid firms to pay a higher wage than necessary toobtain workers; but the level of the efficiency wagecould vary across firms; for instance, finms withhigher turnover costs, or where worker inefficiencycould lead to large losses of capital, or where mon-itoring was more difficult, might find it desirable to

Vol. 47, No. 2 (Fall 2003) 11

pay higher wages. The implication was that similarlabor might receive quite different compensation;wage discrepancies might not be explicable solelyin terms of differences in abilities.

I was to return to these four themes repeatedly inmy research over the following three decades.

FROM THE COMPETITIVEEQUILIBRIUM PARADIGM TO THE

INFORMATION PARADIGM

In the previous section, I described how myexperiences, especially in Kenya-the disparitiesbetween the models used and the world that I saw-had motivated a search for an alternative paradigm.But there was another motivation, driven more bythe intemal logic and structure of the competitivemodel itself, which was the dominant paradigmthirty years ago.

The model virtually made economics a branch ofengineering (with no aspersions to that noble pro-fession), and the participants in the economy betteror worse engineers. Each was solving a maximiza-tion problem, with full information: householdsmaximizing utility subject to budget constraints,firms maximizing profits (market value), and thetwo interacting in competitive product, labor, andcapital markets. One of the peculiar implicationswas that there never were disagreements about whatthe firm should do: alternative management teamswould presumably come up with the same solutionto the maximization problems. Another peculiarimplication was the meaning of risk: when a firmsaid that a project was risky, that (should have)meant that it was highly correlated with the busi-ness cycle, not that it had a high chance of failure.36

I have already described some of the other peculiarimplications of the model: the fact that there was nounemployment or credit rationing, that it focused ononly a limited subset of the information problemsfacing society, that it seemed not to address keyissues-like incentives and motivation.

But much of the research in the profession isdirected not at these big lacunae, but at seeminglymore technical issues-at the mathematical struc-tures. The underlying mathematics requiredassumptions of convexity and continuity, and withthese assumptions one could prove the existence ofequilibrium and its (Pareto) efficiency. The stan-dard proofs of these fundamental theorems of wel-

fare economics did not even list in their enumerat-ed assumptions those concerning information: theperfect information assumption was so ingrained itdid not have to be explicitly stated. The economicassumptions to which the proofs of efficiencycalled attention concemed the absence of externali-ties and public goods. The market failures approachto the economics of the public sector37 discussedalternative approaches by which these market fail-ures could be corrected, but these market failureswere highly circumscribed.

There was, moreover, a curious disjunctionbetween the language economists used to explainmarkets and the models they constructed. Theytalked about the information efficiency of the mar-ket economy, though they focused on a single infor-mation problem, that of scarcity. But there are amyriad of other information problems faced by con-sumers and firms every day, concerning for instancethe prices and qualities of the various objects thatare for sale in the market, the quality and efforts ofthe workers they hire, the returns of investment pro-jects. In the standard paradigm, the competitivegeneral equilibrium model,3" there were no shocks,no unanticipated events: at the beginning of time,the full equilibrium was solved, and everythingfrom then on was an unfolding over time of whathad been planned in each of the contingencies. Inthe real world, the critical question was how, andhow well, do markets handle these informationproblems?

There were other aspects of the standard para-digm which seemed hard to accept. It argued thatinstitutions did not matter-markets could seethrough them, and equilibrium was simply deter-mined by the laws of supply and demand. It saidthat the distribution of wealth did not matter.39 Andit said that (by and large) history did not matter-knowing preferences and technology and initialendowments, one could describe the time path ofthe economy.40

Work on the economics of information began byquestioning each of the underlying premises, eachof the central theorems. Consider, to begin with, themathematical structures that had underlay somemuch of the formalization of economics of the lat-ter half of the twentieth century, the convexityassumptions which corresponded to long standingprinciples of diminishing returns. With imperfectinformation (and the costs of acquiring it) theseassumptions were no longer plausible. It was not

THE AMERICAN ECONOMIST12

just that the cost of acquiring information could beviewed as fixed costs. 41 Work with Roy Radner(Radner and Stiglitz [1984]) showed that there wasa fundamental non-concavity in the value of infor-mation, that is, under quite general conditions, itnever paid to buy just a little bit of information.Work with Richard Arnott (Arnott and Stiglitz[1988a]) showed that such problems were pervasivein even the simplest of moral hazard problems(where individuals had a choice of alternativeactions, e.g. the amount of risk to undertake.) Whilewe had not repealed the law of diminishing returns,we had shown its domain to be more limited thanhad previously been realized.42

Michael Rothschild and I showed that under nat-ural formulations of what might be meant by a com-petitive market with imperfect information, equilib-rium often did not exist43-even when there was anarbitrarily small amount of information imperfec-tion.4 While subsequent research has looked foralternative definitions of equilibrium,45 we remainunconvinced; most of them violate the naturalmeaning of competition, i.e. where each participantin the market is so small that he believes that he willhave no effect on the behavior of others. (Roth-schild and Stiglitz [1997]).

The new information paradigm went further inundermining the foundations of competitive equi-librium analysis, the basic "laws" of economics,which include: the law of demand and supply (hold-ing that market equilibrium was characterized bymarket clearing), the law of the single price, hold-ing that the same good sold for a single pricethroughout the market, the law of the competitiveprice, holding that in equilibrium price equaledmarginal cost, the efficient markets hypothesis,holding that in stock markets prices convey all therelevant information from the informed to the unin-formed. Each of these cornerstones was rejected, orwas shown to hold under much more restrictiveconditions.

We have shown how, when prices affect "qual-ity"-either because of incentive or selectioneffects-equilibrium may be characterized bydemand not equaling supply; firms will not paylower wages to workers, even when they canobtain such workers, because doing so willraise their labor costs; fimns will not chargehigher interest rates, even when they can do so,because of an excess demand for credit,

because doing so will increase the averagedefault rate, and thus lower expected returns.

* We have shown that the market will be charac-terized by wage and price distributions, evenwhen there is no exogenous source of "noise"in the economy, even when all finns and work-ers are (otherwise) identical.

* We have shown that in equilibrium, firms willcharge a price in excess of the marginal costs,or workers are paid a wage in excess of theirreservation wage. The "surplus" is required toprovide the incentive for maintaining a reputa-tion.46 Even in situations where reputation rentswere not required, information imperfectionsgave rise to market power-there is imperfectcompetition-which results in firns chargingprices in excess of marginal cost.47

* The efficient markets hypothesis43 held thatprices in the stock market fully reflected allinformation. But if that were the case, thenthere would be no incentive for anyone toexpend money to collect information. Workwith Sanford Grossman [1976, 1980a] showedthat the price system both imperfectly aggre-gated information and that there was an equi-librium amount of "disequilibrium."

The most fundamental reason why markets withimperfect information differ from the perfect onesis that actions (including choices) convey informa-tion, market participants know this, and this affectstheir behavior.

A decision by a firm to provide a guarantee is notjust a matter that the firm is better able to absorb therisk of a product failure; his willingness to providea guarantee conveys information about his confi-dence in the product. An insured is willing to take apolicy with a large deductible not because he is notrisk averse, but because this action conveys infor-mation to the insurance company, that he is willingto bear the risk because he thinks the likelihood ofan accident is low. At the same time, a firm may notassign an employee to a highly visible job becauseit knows that the assignment will be interpreted asan indication that the employee is good, making itmore likely that a rival will try to hire the personaway. Even if he fails, the current employer willhave to pay a higher salary.

One of the early insights (Akerlof, 1971) wasthat markets may be thin or absent. One of the stan-dard assumptions of the old paradigm was that therewas a complete set of markets-including intertem-

Vol. 47, No. 2 (Fall 2003) 13

poral markets (capital markets) and risk markets.The absence of particular markets, e.g. for risk, hasprofound implications for how other markets func-tion. The fact that workers and firns cannot buyinsurance against many of the risks which they faceaffects labor and capital markets; it leads, forinstance, to labor contracts in which the employerprovides some insurance. But the design of thesemore complicated, but still imperfect and incom-plete, contracts, affects the efficiency, and overallperformance, of the economy.

Perhaps most importantly, under the standardparadigm, markets are Pareto efficient, except whenone of a limited number of market failures occurs.Under the imperfect information paradigm, marketsare almost never Pareto efficient.

While information economics thus underminedthese long standing principles of economics, it alsoprovided explanations for many phenomena thathad long been unexplained. Earlier, I mentioned theseemingly inefficient institution of sharecropping,for which information economics provided anexplanation. Before turning to these applications, Iwant to present a somewhat more systematicaccount of the principles of the economics of infor-mation.

Some problems in constructing analternative paradigm

The fact that information was imperfect was, ofcourse, well recognized by all economists. Whilethey may have hoped that economies with imper-fect information behaved much like economies withperfect information, the real reason that modelswith imperfect information were not developed wasthat it was not obvious how to do so. There wereseveral problems that had to be overcome: whilethere was a single way in which information is per-fect, there are an infinite number of ways in whichinformation can be imperfect. One of the keys tosuccess was formulating simple models in whichthe set of relevant information could be fully speci-fied-and so the precise ways in which informationwas imperfect could also be fully specified. Butthere was a danger in this methodology, as useful asit was: in these over simplistic models, there weresometimes ways in which there could be full infor-mation revelation; the information problems couldbe fully resolved. In the real world, of course, thisnever happens, which is why in some of the later

work (e.g. Grossman and Stiglitz [1976, 1980a], weworked with models with an infinite number ofstates.49

Perhaps the hardest problem was modeling equi-librium. It was important to think about both sidesof the market-employers and employees, theinsurance company and the insured, lender and bor-rower. Each had to be modeled as "rational," insome sense, making inferences on the basis ofavailable information. Each side's behavior too hadto be rational, based on beliefs about the conse-quences of their actions; and those consequences inturn depended on what inferences others woulddraw from those actions. I wanted to model com-petitive behavior, where each actor in the economywas small, and believed he was small-and so hisactions could not or would not affect the equilibri-um (though others' inferences about himself mightbe affected). Finally, one had to think carefullyabout what was the feasible set of actions: whatmight each side do to extract or convey informationto others.

As we shall see, the variety of results obtained(and much of the confusion in the early literature)arose partly from a failure to be as clear as onemight about the assumptions. For instance, the stan-dard adverse selection model had the quality of thegood offered in the market (say of used cars, or risk-iness of the insured) depending on price. The carbuyer (the seller of insurance) knows the statisticalrelationship between price and quality, and thisaffects his demand. The market equilibrium is theprice at which demand equals supply. But that is anequilibrium if and only if there is no way by whichthe seller of a good car can convey that informationto the buyer-so that he can earn a quality premi-um-and if there is no way by which the buyer cansort out good cars from bad cars. Typically, thereare such ways, and it is the attempt to elicit thatinformation which has profound effects on howmarkets function. To develop a new paradigm, wehad to break out from long established premises, toask what should be taken as assumptions and whatshould be derived from the analysis. Market clear-ing could not be taken as an assumption; neithercould the premise that a firm sells a good at a par-ticular price to all customers. One could not beginthe analysis even by assuming that in competitiveequilibrium there would be zero profits. In the stan-dard theory, if there were positive profits, a firmmight enter, bidding away existing customers. In

THE AMERICAN ECONOMIST14

the new theory, the attempt to bid away new cus-tomers by slightly lowering prices might lead tomarked changes in their behavior or in the mix ofcustomers, in such a way that the profits of thenew entrant actually became negative. One had torethink all the conclusions from first premises.

We made progress in our analyses because webegan with highly simplified models of particularmarkets, that allowed us to think through careful-ly each of the assumptions and conclusions. Fromthe analysis of particular markets (whether theinsurance market, as in Rothschild Stiglitz, theeducation market, the labor market, or the landtenancy/sharecropping market), we attempted toidentify general principles, to explore how theseprinciples operated in each of the other markets. Indoing so, we identified particular features, partic-ular informational assumptions, which seemed tobe more relevant in one market or another. Thenature of competition in the labor market is differ-ent than that in the insurance market or the capitalmarket, though they have much in common. Thisinterplay, between looking at the ways in whichsuch markets are similar and dissimilar, proved tobe a fruitful research strategy.50

SOURCES OF ASYMMETRIES OFINFORMATION

Information imperfections are pervasive in theeconomy: indeed, it is hard to imagine what aworld with perfect information would be like.Much of the research I will describe below focus-es on asymmetries of information, the fact that dif-ferent people know different things: workers knowmore about their ability than does the finn; theperson buying insurance knows more about hishealth, whether he smokes and drinks immoder-ately, than the insurance firm; the owner of a carknows more about the car than potential buyers;the owner of a firm knows more about the firmthat a potential investor; the borrower knows moreabout his risk and risk taking than the lender.

The essential feature of a decentralized marketeconomy is that different people know differentthings; in this sense, economists had long beenthinking of markets with information asymme-tries. But the earlier literature had neither thoughtabout how they were created, nor what their con-sequences might be. Moreover, while much of the

earlier literature focused on simple situations ofinformation asymmetry-such as those describedin the previous paragraphs, the problems of infor-mation imperfections run deeper, and the researchdescribed below discusses some of these moregeneral results. The individual may know littleabout his true health condition; the insurance com-pany, through a simple examination, might evenbecome more informed (at least concerning rele-vant aspects, e.g. implications for life expectancy).

Some of these information asymmetries areinherent: the individual naturally knows moreabout himself than does anyone else. Some of theasymmetries arise naturally out of economicprocesses. The current employer knows moreabout the employee than other potential employ-ers; a firm may find out a great deal of informationin the process of dealing with suppliers that othersmay not know; the owner of a car naturally knowsthe faults of the car better than others-and in par-ticular, he knows whether or not he has a lemon.While such information asymmetries inevitablyarise, the extent to which they do so and their con-sequences depend on how the market is structured,and the recognition that they will arise affectsmarket behavior. For instance, one of the impor-tant insights of work in this area is to show howinformation asymmetries lead to thin or non-exis-tent markets (Akerlof [1970]). But this means thateven if an individual has no more informationabout his ability than potential employers, themoment he goes to work for an employer, an infor-mation asymmetry has been created-the employ-er may know more about the individual's abilitythan others. The consequence is that the "usedlabor" market does not work well. Others will bemore tame in bidding for his services, knowingthat they will succeed in luring him away from hiscurrent employer only if they bid too much. If theybid less than his productivity, his current employ-er will match. Labor mobility is impeded. But thatgives market power to the first employer, which hewill be tempted to exercise. The recognition of thisnaturally affects even the "new labor" market.Because an individual is locked into a job, he willbe more risk averse in accepting an offer. Theterms of the initial contract have to be designed toreflect the diminution of the workers' bargainingpower and his reduced labor mobility that occursimmediately after signing.5 ' 52

Vol. 47, No. 2 (Fall 2003) 15

To take another example, it is natural that in theprocess of oil exploration, a company finds outinformation that is relevant for the likelihood thatthere will be oil in a neighboring tract. There is aninformational externality. 53 The existence of thisasymmetric information affects the nature of thebidding for oil rights on the neighboring tract.Bidding where there is known to be asymmetriesof information will be markedly different fromthat where such asymmetries do not exist.54 Thosewho are uninformed will presume that they willwin only if they bid too much-information asym-metries exacerbate the problem of the winners'curse.55 The government (or other owners of largetracts to be developed) should take this intoaccount in its leasing strategy. And the bidders inthe initial leases too will take this into account:part of the value of winning in the initial auction isthe information rent that will accrue in laterrounds.

Creating asymmetries and imperfections ofinformation

While early work in the economics of informa-tion dealt with how markets overcame problems ofinformation asymmetries, and information imper-fections more generally, later work turned to howmarkets create information problems, partly in anattempt to exploit market power. Managers offinns attempt to entrench themselves, increasingtheir bargaining power, e.g. vis-a-vis alternativemanagement teams,56 and one of the ways that theydo this is to take actions which increase informa-tion asymmetries. (Edlin and Stiglitz [1995]).Doing so effectively reduces competition in themarket for management. This is an example of thegeneral problem of corporate governance, towhich I alluded earlier, and to which I will returnlater.

Similarly, the presence of information imper-fections give rise to market power; and firms canexploit this market power through "sales" andother ways of differentiating among individualswho have different search costs. (Salop, 1977,Salop and Stiglitz, 1976, 1982, Stiglitz 1979a).The price dispersions which exist in the market arecreated by the market-they are not just the fail-ure of markets to arbitrage fully price differencescaused by shocks that affect different markets dif-ferently.

OVERCOMING INFORMATIONASYMMETRIES

I now want to discuss briefly the ways by whichinformation asymmetries are dealt with, how theycan be (partially) overcome.

Incentives for gathering and disclosinginformation

There are two key issues: what are the incentivesfor obtaining information, and what are the mecha-nisms. My brief discussion of the analysis of educa-tion as a screening device suggested the fundamen-tal incentives: more able individuals (lower riskindividuals, firms with better products) will receivea higher wage (will have to pay a lower premium,will receive a higher price for their products) if theycan establish that they are more productive (lowerrisk, higher quality).

We noted earlier that while some individualshave an incentive to disclose information, thosewho are less able have an incentive not to have theinformation disclosed. Was it possible that in mar-ket equilibrium, only some of the informationwould be revealed? One of the early importantresults was that, if the more able succeed (costless-ly) to establish that they are more able, then themarket will be fully revealing, even though all ofthose who are below average would prefer that noinformation be revealed. In the simplest models, Idescribed a process of unraveling: if the most ablecould establish his ability, he would; but then all butthe most able would be grouped together, receivingthe mean marginal product of that group; and themost able of that group would have an incentive toreveal his ability. And so on down the line, untilthere was full revelation.5 7

What happens if those who are more able cannotcredibly convince potential employers of their abil-ity (or if those who are low risk cannot convincepotential insurance companies)? The other side ofthe market has an incentive too to gather informa-tion: an employer that can find a worker who is bet-ter than is recognized by others will have found abargain; his wage will be determined by what oth-ers think of him. The problem, as we noted, is thatif what he knows becomes known to others, thewage will be bid up, and he will be unable to appro-priate the returns on his investment in informationacquisition.

THE AMERICAN ECONOMIST16

The fact that if there was competition, it wouldbe difficult for the screener to appropriate thereturns had an important implication: in marketswhere, for one reason or another, the more able (thefirms with the better investment projects, the moreable workers) cannot (fully) convey their attributes,if there is to be investment in screening there mustbe imperfect competition in screening. The econo-my, in effect, has to choose between two differentimperfections: imperfections of information orimperfections of competition. Of course, in the end,there will be both forms of imperfection.58' 59

This is but one of many examples of the inter-play between market imperfections. Earlier, forinstance, we discussed the incentive problems asso-ciated with sharecropping, which arise when work-ers do not own the land that they till. This problemcould be overcome if individuals could borrowmoney, to buy their land. But capital market imper-fections-limitations on the ability to borrow,which themselves arise from information imperfec-tions-explain why this "solution" does not work.

There is another important consequence: if mar-kets were fully informationally efficient-that is, ifinformation disseminated instantaneously and per-fectly throughout the economy-then no one wouldhave any incentive to gather information, so long asthere was any cost of doing so. That is why marketscannot be fully informationally efficient. (SeeGrossman and Stiglitz, 1976, 1980a.)

Mechanisms for elimination or reducinginformation asymmetries

In simple models where individuals know theirown ability, or the insured knows his own risk, orthe borrower knows his own likelihood of repaying,there might seem an easy way to resolve the prob-lem of information asymmetry: let each person tellhis true characteristic. The underlying problemarose from the fact that individuals did not neces-sarily have the incentive to tell the truth. Assumeemployees knew their abilities. An employer mightask, what is your ability? The more able mightanswer honestly. As we have seen, the least ablewould have an incentive to lie, to say that he wasmore able than he was. Talk is cheap. There had tobe some other ways by which information could becredibly conveyed.

Screening by examination

The simplest way by which that could be donewas an exam. As I constructed a simple competitiveequilibrium model,60 two further general principlesbecame apparent: the gains of the more able werelargely at the expense of the less able; by establish-ing that an individual is of higher ability, therebyleading, in equilibrium, to higher wages, he simul-taneously establishes that others are of lower abili-ty. The private returns to expenditures on educationexceed the social returns. It was clear that therewere important externalities associated with infor-mation, a theme which was to recur in later work.

But a more striking result emerged: there couldexist multiple equilibria, one in which informationwas fully revealed (the market identified the highand low ability people) and the other of which itwas not (called a pooling equilibrium). The poolingequilibrium Pareto dominated the equilibrium withfull revelation. This work, done some thirty yearsago, established two results of important policy,which remarkably have not been fully absorbed intopolicy discussions even today. First, markets do notprovide appropriate incentives for infonnation dis-closure. There is, in principle a role for government.And secondly, expenditures on information may betoo great.6"

The simplest adverse selection model

But much of the information firns glean abouttheir employees, banks about their borrowers,insurance companies about their insured comes notfrom examinations but from making inferencesbased on their behavior. This is a common practicein life-but not in our economic models. As I havealready noted, the early discussions of adverseselection in insurance markets recognized that as aninsurance company raised its premiums, those whowere least likely to have an accident decided not topurchase the insurance; the willingness to purchaseinsurance at a particular price conveyed informa-tion to the insurance company.62 George Akerlofrecognized that this phenomenon was far more gen-eral: the willingness to sell a used car, for instance,conveyed information about whether the car was orwas not a lemon.

Bruce Greenwald [1979, 1986] took the idea oneimportant step further, showing how adverse selec-tion applied to labor and capital markets:63 The will-

Vol. 47, No. 2 (Fall 2003) 17

ingness of an employer not to match the bid of acompetitor conveyed information about the currentemployer's judgment of that individual's ability;the willingness of insiders in a firm to sell stock ata particular price conveyed information about theinsider's view of the price relative to the expectedreturn. Akerlof's claim that the result of these infor-mation asymmetries was that markets would be thinor absent helped explain why labor and capital mar-kets often did not function well. It provided part ofthe explanation for why firms raised so little of theirfunds through equity (Mayer, 1990). Stigler waswrong: imperfect information was not just liketransactions costs.

The consequences go well beyond just an absentor missing market. Weak equity markets meant thatrisks could not be divested, leading firms to act in arisk averse manner, explaining some of what wouldotherwise seem to be anomalous aspects of firmbehavior.' These capital market imperfections, inturn, played a central role in the macro-economictheories to be described below. We have alreadydescribed how the labor market imperfections-thelimited mobility of labor and the firm's marketpower that results-affects the labor market, bothbefore the asymmetry of information is created inthe process of hiring and after.

The simplest adverse incentive model

In the adverse selection model, individuals dif-fered. There was a single action which conveyedinformation: they either entered or did not enter theparticular market. But information imperfectionsalso relate to what people do. A worker can workharder, a borrower can undertake greater risk andthe insured can undertake greater care. The employ-er would like to know how hard his employeeworks; if he could, he would specify that in the con-tract; the lender would like to know the actionswhich borrower will undertake; if he could, hewould specify that in the contract. These asymme-tries of information about actions are as importantas the earlier discussed asymmetries. Just as in theadverse selection model, the seller of insurance maytry to overcome the problems posed by informationasymmetries by examination, so too in the adverseincentive model, he may try to monitor the actionsof the insured. But examinations and monitoring arecostly and, while they yield some information, typ-ically there remains a high level of residual infor-

mation imperfection. Just as in the adverse selectionmodel, the seller of insurance recognizes that theaverage riskiness of the insurance applicants isaffected by the terms of the insurance contract, sotoo the level of risk taking can be affected. And sim-ilar results hold in other markets. Borrowers' risktaking is affected by the interest rate charged.(Stiglitz and Weiss [1981]).

Efficiency wage theory, credit rationing

While the early work in adverse selectionexplored the equilibrium in markets where the sell-er of insurance (the employer, the buyer of usedcars, the lender) was rational enough to recognizethe dependence of quality on price, he was notrational enough to exploit as fully as he could theinformation. While the law of supply and demandhad been assumed to be a law of economics, thereis in fact no law that requires the insurance firn tosell to all who apply at the premium he announces,the lender to lend to all who apply at the interestrate he announces, the employer to employ all thosewho apply at the wage he announces. So ingrainedwas the competitive equilibrium model in the mind-set of economists that they simply assumed price-taking behavior. With perfect information and per-fect competition, any firm that charged a pricehigher than the others would lose all of his cus-tomers; and at the going price, one faced a perfect-ly elastic supply of customers. In adverse selectionand incentive models, what mattered was not justthe supply of customers or employees or borrowers,but also their "quality"-the riskiness of the insuredor the borrower, the retums on the investment, theproductivity of the worker.

Since "quality" may increase with price, it maypay to offer a higher wage than the market clearingwage, for the lender to lend at an interest rate whichexceeds the market clearing interest rate. This istrue whether the dependence on quality arises fromadverse selection or adverse incentive effects (or, inthe labor market, because of morale or nutritionaleffects). And what matters is that there be imperfectinformation, not asymmetries of information. Thehealthy who decide not to buy insurance at a highpremium do not need to know that they are healthy;they could be as uninformed as the insurance com-pany, but simply-perhaps because of theirhealth-have different preferences, e.g. they preferto spend more of their money on recreational sports.

THE AMERICAN ECONOMIST18

The consequence, as we have noted, is that mar-ket equilibrium may be characterized by demandnot equaling supply: in equilibrium, the interest rateis lower than that at which the demand for loansequals the supply-there is credit rationing (Stiglitzand Weiss [1981], Keeton [1979]); the wage rate ishigher than that at which the demand for laborequals the supply-there is unemployment.65

Conveying information through actions

There is a much richer set of actions which con-vey information beyond those on which traditionaladverse selection models have focused. An insur-ance company wants to attract healthy applicants. Itmight realize that by locating itself on the fifth floorof a walk up building, only those with a strong heartwould apply. The willingness or ability to walk upfive floors conveys information. More subtly, itmight recognize that how far up it needs to locateitself, if it only wants to get healthy applicants,depends both on the premium charged and howhigh it locates itself. Or it may decide to throw infor free a membership in a health club, but charge ahigher premium. Those who value a health club-because they will use it-willingly pay the higherpremium. But these individuals are likely to behealthier.

There are a host of other actions which conveyinformation. The quality of the guarantee offered bya firm can convey information about the quality ofthe product; only firms that believe that their prod-uct is reliable will be willing to offer a good guar-antee. The guarantee is desirable not just because itreduces risk, but because it conveys information.The number of years of schooling may conveyinformation about the ability of an individual. Moreable individuals may go to school longer, in whichcase the increase in wages associated with anincrease in schooling may not be a consequence ofthe human capital that has been added, but rathersimply be a result of the sorting that occurs.66 Thesize of the deductible that an individual chooses inan insurance policy may convey information abouthis view about the likelihood of an accident or thesize of the accidents he anticipates-on average,those who are less likely to have an accident may bemore willing to accept high deductibles. The will-ingness of an entrepreneur to hold large fractions ofhis wealth in a firm (or to retain large fractions ofthe shares of the firm) conveys information about

Vol. 47, No. 2 (Fall 2003)

his beliefs in the finn's future performance. If afirm promotes an individual to a particular job, itmay convey information about the firn's assess-ment of his ability.

The fact that these actions may convey informa-tion affects behavior. In some cases, the action willbe designed to obfuscate, to limit information dis-closure. The firm that knows that others are lookingat who it promotes, and that it will compete morevigorously for those, may affect the willingness ofthe firm to promote some individuals or assignthem to particular jobs. (Waldman, 1984.) In others,the action will be designed to convey information ina credible way, to alter beliefs. The fact that cus-tomers will treat a firm that issues a better guaran-tee as if its product is better-and therefore be will-ing to pay a higher price-may affect the guaranteethat the firm is willing to issue. Knowing that hisselling his shares will convey a negative signal con-cerning his views of the future prospects of his firm,an entrepreneur may retain more of the shares of thefirm; he will be less diversified than he otherwisewould have been (and accordingly, he may act in amore risk averse manner).

A simple lesson emerges: some individuals wishto convey information; some individuals wish notto have information conveyed (either because suchinformation might lead others to think less well ofthem, or because conveying information may inter-fere with their ability to appropriate rents). In eithercase, the fact that actions convey information leadspeople to alter their behavior, and changes howmarkets function. This is why information imper-fections have such profound effects.

Once one recognizes that actions convey infor-mation, two results follow. First, in making deci-sions about what to do, individuals will not onlythink about what they like (as in traditional eco-nomics) but how it will affect others' beliefs aboutthem. If I choose to go to school longer, it may leadothers to believe that I am more able, and I willtherefore decide to stay in school longer, notbecause I value what is being taught, but because Ivalue how it changes others' beliefs concerning myability. This means, of course, that we have torethink completely firm and household decisionmaking.

Secondly, we noted earlier that individuals havean incentive to "lie"-the less able to say that theyare more able. Similarly, if it becomes recognizedthat those who walk up to the fifth floor to apply for

19

insurance are more healthy, then I might be willingto do so even if I am not so healthy, simply to foolthe insurance company. If it becomes recognizedthat those who stay in school longer are more able,then I might be willing to do so, even if I am lessable, simply to fool the employers. Recognizingthis, one needs to look for ways by which informa-tion is conveyed in equilibrium. The critical insightas to how that could occur was provided in a paperwith Michael Rothschild [1976]. If those who weremore able, less risk prone, more credit worthy actedin some observable way (had different preferences)than those who were less able, less risk prone, lesscredit worthy, then it might be possible to design aset of choices, which would result in those with dif-ferent characteristics in effect identifying them-selves through their self-selection. One of the rea-sons that they might behave differently is that theyknow they are more able, less risk prone, more cred-itworthy-that is there is asymmetric information.But it is only one of the bases for self-selection.

The particular mechanism which we explored inour insurance model illustrates how self-selectionmechanisms work. People who know they are lesslikely to have an accident will be more willing toaccept an insurance policy with a high deductible,so that an insurance company that offered two poli-cies, one at a high premium and no deductible, onewith a low premium and high deductible, would beable to sort out who were high risk and who low. Itis an easy matter to construct choices which thusseparate.

Monopoly and self-selection

Analyzing the choices which arise in full equi-librium behavior turned out, however, to be a diffi-cult task. The easiest situation to analyze was thatof a monopolist.5' He could construct a set of choic-es that would differentiate among different types ofindividuals, and analyzed whether it was profitmaximizing for him to do so fully, or to (partially)"pool"-that is, offer a set of contracts such thatseveral types might choose the same one. This worklaid the foundations of a general theory of price dis-crimination. Under standard theories of monopoly,with perfect information, firms would have anincentive to price discriminate perfectly (extractingthe full consumer surplus from each). If they didthis, then monopoly would in fact be non-distor-tionary. Yet most models assumed no price discrim-

ination (that is, the monopolist offered the sameprice to all customers), without explaining why theydid not do so, and argued that monopoly was dis-tortionary. Our work showed how, given limitedinformation, firms could price discriminate, butcould do so only imperfectly. Subsequent work by avariety of authors (such as Salop [1977] and Adamsand Yellen [1976]) explored a variety of ways bywhich a monopolist might find out relevant charac-teristics of his customers-the extent of discrimina-tion limited by his ability to identify each person's"surplus," the maximum they would be willing topay. (For an insurance company, the relevant char-acteristics are not only the likelihood of having anaccident, but also the degree of risk averse, the pre-mium that an individual would be willing to pay todivest himself of risk.) The economics of informa-tion thus provided the first coherent theory ofmonopoly.

Self-selection and competitive equilibrium

The reason that analyzing monopoly was easy isthat the monopolist could structure the entire choiceset facing his customers. The hard question is todescribe the full competitive equilibrium, that is aset of insurance contracts such that no one can offeran alternative set which would be profitable. Eachfirm could control the choices that he offered, butnot the choices offered by others; and the decisionsmade by customers depended on the entire set ofchoices available. In our 1976 paper, Rothschildand I succeeded in analyzing this case.

Three striking results emerged from this analy-sis. The first I have already mentioned: under plau-sible conditions, given the natural definition ofequilibrium, equilibrium might not exist. Therewere two possible forms of equilibria, pooling equi-libria, in which the market is not able to distinguishamong the types, and separating equilibria, inwhich it is. The different groups "separate out" bytaking different actions. We showed that there nevercould be a pooling equilibrium-if there were a sin-gle contract that everyone bought, there was anoth-er contract that another firm could offer whichwould "break" the pooling equilibrium. On theother hand, there might not exist a separating equi-librium. The cost of separation was too great. Anyputative separating equilibrium could be broken bya profitable pooling contract, a contract which

THE AMERICAN ECONOMIST20

would be bought by both low risk and high risktypes.

Second, even small amounts of imperfections ofinformation can change the standard results, con-cerning the existence and characterization of equi-librium. Equilibrium, for instance, never existswhen the two types are very near each other. As wehave seen, the competitive equilibrium model issimply not robust.

Thirdly, and relatedly, we now can see how thefact that actions convey information affects theequilibrium. In particular, our analysis here rein-forced the earlier analysis of adverse selectionabout markets not functioning well. In perfect infor-mation models, individuals would fully divestthemselves of the risks which they face, and accord-ingly would act in a risk neutral manner. Weexplained why insurance markets would not workwell-why most risk averse individuals would buyonly partial insurance. There were numerous subse-quent applications to other markets, reinforcing, forinstance, the earlier conclusions concerning the lim-itations on equity markets. (The reason that theoriginal owners of a firm might want to sell hisshares was to "insure" himself against the risk of abad outcome; an owner that believed that there wasa smaller probability of a bad outcome would bewilling to buy less insurance, i.e. to divest himselfof fewer of his shares. Retention of shares can thusbe thought of as a market sorting mechanism, thewillingness to' keep these shares a "signal" of theowners' confidence.)68

The result was important not only for the insightsit provided in the workings of an important set ofmarkets in the economy, but because there areimportant elements of insurance in many transac-tions and the general principle that actions conveyinformation, and that market transactions are great-ly affected by this fact, has implications in a stillwider variety of contexts. The relationship betweenthe landlord and his tenant, or the employer and hisemployee, can be viewed as containing in it aninsurance component; limitations on the ability todivest oneself of risk are important in explaining ahost of contractual relationships.

Sorting, screening, and signaling

In equilibrium, both buyers and sellers, employ-ers and employees, insurance company and insured,lender and creditor are aware of the informational

consequences of their actions. Each side of the mar-ket needs to consider the consequences, e.g. of act-ing in a different way, or of confronting the otherside of the market with different choices. In the casewhere say, the insurance company or employer oremployee takes the initiative for sorting out appli-cants, self-selection is an alternative to examina-tions as a sorting device. In the case where theinsured, or employee, or borrower, takes the initia-tive for identifying himself as a better risk, a betteremployee, a borrower more likely to repay, then wesay he is signaling.6 9 But of course, in equilibriumboth sides are aware of the consequences of alter-native actions, and the differences between signal-ing and self-selection screening models lie in thetechnicalities of game theory, and in particularwhether the informed or uninformed (employee oremployer, insured or insurance company) movesfirst.70

Still, some of the seeming differences betweensignaling and screening models arise because of afailure to specify a full equilibrium We noted earli-er that there were many separating contracts, but aunique separating equilibrium. We argued that ifone considered any other separating set of con-tracts, then, say, in the insurance market, a firmcould come in and offer an alternative set of con-tracts and make a profit; the original set of separat-ing contracts could not have been an equilibriumThe same is true in, say, the education signalingmodel. There are many,educational systems which"separate"-that is, the more able choose to go toschool longer, and the wages at each level of edu-cation correspond to the productivity of those whogo to school for that length of time. But all exceptone are not in full equilibrium. Assume, forinstance, there were two types of individuals, a lowability and a high ability. Then if the low abilitygoes to 12 years of schooling, then any educationsystem in which the high ability went sufficientlylong-say more than 14 years-might separate. Butthe low ability would recognize that if it went toschool for 11 years, it would still be treated as lowability. The unique equilibrium level of educationfor the low ability is that which maximizes his netincome (taking into account the productivity gainsand costs of education); and the unique equilibriumlevel of education for the high ability is the lowestlevel of education such that, if his wage corre-sponds to his productivity at that level of education,the low ability will still prefer to remain at his low

Vol. 47, No. 2 (Fall 2003) 21

level of education rather than pretend to be moreable by staying in school longer.7 '

The education system, of course, was particular-ly infelicitous for studying market equilibrium. Thestructure of the education system is largely a matterof public choice, not of market processes. Differentcountries have chosen markedly different systems.The minimum level of education is typically not amatter of choice, but set by the government. Withineducational systems, examinations play as impor-tant a role as self-selection or signaling, thoughgiven a certain standard of testing, there is a processof self-selection involved in deciding whether tostay in school, to try to pass the examination. 72 Forthe same reason, the problems of existence whicharise in the insurance market are not relevant in theeducation market-the "competitive" supply sideof the market is simply absent. But when the sig-naling concepts are translated into contexts inwhich there is a robust competitive market, theproblems of existence cannot be so easily ignored.73

Existence of equilibrium

What are we to make of the problem of exis-tence? Clearly, insurance markets exist, even if theyare far from complete. To some extent, the marketdoes exhibit instability. Rates vacillate enormously;as rates sometimes skyrocket and coverage is cur-tailed, the public clamors for reforms. Such periodsare often followed by periods of relative stability, tobe followed by another "crisis" in the market. Moststates regulate rate setting, though at least partly forprudential reasons and this may help stabilize themarket.

Moreover, though there is considerable evidencefor the kinds of selection processes discussedabove, there is also considerable evidence that themarket is far from as rational as the theory wouldsuggest. Most health insurance policies do not basepremia on the number of children, though that is aneasily observable variable which clearly affects therisk exposure. Many insurance companies do notuse past experience as heavily as one would haveexpected in setting premia, i.e. there is less experi-ence rating.

My own suspicion, however, is that the majorlimitation of Rothschild-Stiglitz is its assumption ofperfect competition; competition is far more limitedthan we postulated; there are, for instance, signifi-cant search costs, and considerable uncertainty

about how easy it is to get the insurance firm to payon a claim. Self-selection is still relevant, but themodel of monopoly, or some version of monopolis-tic competition, may be more relevant than themodel of perfect competition.

THEORY OF CONTRACTS ANDINCENTIVES

The work with Rothschild was related to the ear-lier work that I had done on incentives (sharecrop-ping), besides the obvious way that both were con-cemed with problems of limited information, onefocusing on selection effects and one on incentives.Both entailed equilibrium in "contracts." The con-tracts that had characterized economic relations inthe standard competitive model were extraordinari-ly simple: I will pay you a certain amount if you dosuch and such. If you did not perform as promised,the pay was not given. But with perfect informa-tion, individuals simply didn't sign contracts thatthey did not intend to fulfill. Insurance contractswere similarly simple: a payment occurred if andonly if particular specified events occurred.

The work on sharecropping and on equilibriumwith competitive insurance markets showed thatwith imperfect information, a far richer set of con-tracts would be employed, 74 and thus began a largeliterature on the theory of contracting. 7"

In the simple sharecropping contracts of Stiglitz[1974b], the contracts involved shares, fixed pay-ments, and plot sizes, and76 more generally, optimalpayment structures related payments to observ-ables, inputs, processes, outputs. 77' 76 Because whatwent on in one market affect others, the credit,labor, and land markets were interlinked; one couldnot decentralize in the way hypothesized by thestandard perfect information model.79 The theorythus served as the basis of the rnral organization indeveloping countries.Y0

The basic principles were subsequently appliedin a variety of other market contexts. The mostobvious was the design of labor contracts.8!' 82

Payments, too, can depend on relative perfor-mance; relative performance may convey more rel-evant information than absolute performance. If aparticular company's stock goes up when all othercompanies' stock goes up, it may say very littleabout the performance of the manager. In Nalebuffand Stiglitz [1983a, 1983b] we analyzed the design

THE AMERICAN ECONOMIST22

of these relative performance compensationschemes (contests). One of the strong arguments forcompetitive, decentralized structures is that theyprovide information on the basis of which one candesign better incentive pay structures than thosewhich rely on the performance of a single individ-ual only.

Credit markets too are characterized by compli-cated contracts. Lenders would specify not only aninterest rate, but also impose other conditions (col-lateral requirements, equity requirements) whichwould have both incentive and selection effects." 3

Indeed, the simultaneous presence of both selectionand incentive effects was important: in the absenceof the former, it might be possible to increase thecollateral requirement and raise interest rates, stillensuring that the borrower undertook the safe pro-ject.84

Incentives in market equilibrium

Incentives are based on rewards and punish-ments. In modern economies, the most severe pun-ishment that one can impose is to fire an individ-ual.85 But if the individual could get a job just likehis current one, then there would be no cost. Goodbehavior is driven by eaming a surplus over whatone could get elsewhere. Thus, in labor markets, thewage must be higher than what the worker could getelsewhere (which may be zero, if there is unem-ployment); in the goods market, firms must feel aloss when they lose a customer because of a shod-dy product, so the price must exceed the marginalcost of production. Thus, the long standing pre-sumption that in competitive equilibrium priceequals marginal cost cannot be true in markets withimperfect information. (See Shapiro and Stiglitz[1984], Shapiro [1983] and Klein and Leffler[1981].)

EQUILIBRIUM WAGE AND PRICEDISTRIBUTIONS

One of the most obvious differences between thepredictions of the model with perfect informationand what we see in every day life is the conclusionthat the same good sell for the same price every-where. We all spend a considerable amount of timeshopping for good bargains. The differences inprices represent more than just differences in quali-

ty (service). There are real price differences. SinceStigler's classic paper [1961], there has been a largeliterature exploring optimal search behavior.Stigler, and most of the search literature, took, how-ever, the price or wage distribution as given. Theydid not ask how did it arise. Given the search costs,could it be sustained? For instance, if search costsare relatively low, one might have thought (if onebought the older theories) that markets would lookvery much like they would with zero search costs,in which case there would be no price or wage dis-tribution. It is not surprising that given that infor-mation is costly, if there are shocks to the econo-my-the demand for a good goes up in some locale,so price there rises-that prices are not fully arbi-traged instantaneously. But much of the wage andprice dispersion cannot be related to such "shocks."

Our analysis of efficiency wage theory providedan alternative explanation. We showed that it paidfirms to pay more than they had to, e.g. to reducelabor turnover costs. But it might pay some firms topay higher wages than others.

As I began to analyze these models, an importantinsight occurred: there could be a wage distributioneven if all firms were identical, e.g. faced the samesearch costs. It was clear that even small searchcosts could make a large difference to the behaviorof product and labor markets. This was a point thatDiamond [1971] had independently made in a high-ly influential paper, which serves to illustrate pow-erfully the lack of robustness of the competitiveequilibrium theory. Assume, as in the standard the-ory, all firms were charging the competitive price,but there were an epsilon cost of searching, of goingto another store. Then any firm which chargedepsilon/2 greater would lose no customers. It wouldthus pay him to increase his price. And it wouldsimilarly pay all other firms to increase their prices.But at the higher price, it would again pay each toincrease his price. And price increases until theprice charged is the monopoly price. Even smallsearch costs thus lead even a market with manyfirms to charge monopoly prices. Work with Salop(Salop and Stiglitz [1977, 1982, 1987], Stiglitz[1979b, 1989c]), showed that in situations wherethere were even small search costs, markets wouldbe characterized by a price distribution. If everyonewere charging the same price, it would pay somefirm either to raise his price, to exploit the highsearch costs customers who he would not lose, or tolower his price, to steal customers away from his

Vol. 47, No. 2 (Fall 2003) 23

rivals. The standard wisdom that said that noteveryone had to be informed to ensure that the mar-ket acted in a perfectly competitive manner wassimply not in general true.86

EFFICIENCY OF THE MARKETEQUILIBRIUM AND THE ROLE OF

THE STATE

Perhaps the most important single idea in eco-nomics is that competitive economies lead, as if byan invisible hand, to a (Pareto) efficient allocationof resources, and that every Pareto efficientresource allocation can be achieved through a com-petitive mechanism, provided only that the appro-priate lump sum redistributions are undertaken. It isthese fundamental theorems of welfare economicswhich provide both the rationale for the reliance onfree markets, and the belief that issues of distribu-tion can be separated from issues of efficiency,allowing the economist the freedom to push forreforms which increase efficiency, regardless oftheir seeming impact on distribution; if society doesnot like the distributional consequences. it shouldsimply redistribute income.

The economics of information showed that nei-ther of these results was, in general, true. To besure, economists over the preceding three decadeshad identified important market failures-such asthe externalities associated with pollution-whichrequired government intervention.87 But the scopefor market failures was limited, and thus the arenasin which government intervention was requiredwere limited.

Early work, already referred to, had laid thefoundations for the idea that economies with infor-mation imperfections would not be Pareto efficient,even taking into account the costs of obtaininginformation. There were interventions in the marketthat could make all parties better off. We hadshown, for instance, that incentives for the disclo-sure and acquisition of information were far fromperfect; imperfect appropriability meant that theremight be insufficient incentives, but the fact thatmuch of the gains were "rents," gains by some atthe expense of others, suggested that there might beexcessive expenditures on information. One of thearguments for unfettered capital markets was thatthere were strong incentives to gather information;if one discovered that some stock was more valu-

able than others thought, and if you bought it beforethey discovered the information, then you wouldmake a capital gain. This price discovery functionof capital markets was often advertised as one of itsstrengths. But the issue was, while the individualwho discovered the information a nano-secondbefore any one else might be better off, was societyas a whole better off: if having the information anano-second earlier did not lead to a change in realdecisions (e.g. concerning investment), then it waslargely redistributive, with the gains of thoseobtaining the information occurring at the expenseof others. Another example illustrates what is atissue. Assume hundred dollar bills were to fall, oneeach at the left foot of each student in my class.They could wait to the end of the lecture, then pickup the money; but that is not a Nash equilibrium. Ifall students were to do that, it would pay any one tobend down and quickly scoop up what he could.Each realizing that immediately picks up the dollarbill at his foot. The equilibrium leaves each no bet-ter off than if he had waited-and there was a greatsocial cost, the interruption of the lecture.88

There are potentially other inefficiencies associ-ated with information acquisition. Information canhave adverse effects on volatility.89 And informationcan lead to the destruction of markets, in wayswhich lead to adverse effects on welfare. Wedescribed earlier how the existence of asymmetriesof information can destroy markets. Individualssometimes have incentives to obtain information(creating an asymmetry of information), which thenleads to the destruction of insurance markets, andan overall lowering of welfare. Welfare might beincreased if the acquisition of this kind of informa-tion could be proscribed. Recently, such issues havebecome sources of real policy concern, in the arenaof genetic testing. Even when information is avail-able, there are issues concerning its use, with theuse of certain kinds of information having either adiscriminatory intent or effect, in circumstances inwhich such direct discrimination itself would beprohibited.9 0

Moreover, asymmetries of information wereshown to be related to absent or imperfect markets.They help explain why the market for lemons, orthe credit or labor or equity markets worked imper-fectly. The fact that markets with imperfect infor-mation worked differently-and less well-thanmarkets with perfect information was not, by itself,a damning criticism of markets. After all, informa-

THE AMERICAN ECONOMIST24

tion is costly, and taking into account the costs ofinformation, markets might be fully efficient.Stigler had essentially argued for this perspective,but without proof. Our research showed that thisassertion-or hope-was simply not correct. Earli-er work had established that when markets areabsent or imperfect, market equilibrium might beconstrained Pareto inefficient, that is, taking intoaccount the absence of the market, everyone couldbe made better off.9 ' Moreover, since asymmetriesof information give rise to market power, and per-fect competition is required if markets are to be effi-cient, it is perhaps not surprising that markets withinformation asymmetries and other informationimperfections are far from efficient.

But while it was thus not surprising that marketsmight not provide appropriate incentives for theacquisition and dissemination of information, themarket failures associated with imperfect informa-tion are far more profound. The intuition can beseen most simply in the case of models with moralhazard. There, the premium charged is associatedwith the average risk, and, therefore, fthe averagecare taken by seemingly similar individuals. Themoral hazard problem arises because the level ofcare cannot be observed. Each individual ignoresthe effect of his actions on the premium; but whenthey all take less care, the premium increases. Thelack of care by each exerts a negative externality onothers.

The essential insight of Greenwald and Stiglitz[1986192 was to recognize that such externality-likeeffects are pervasive whenever information isimperfect or markets incomplete-that is always-and as a result, markets are essentially never con-strained Pareto efficient. 93 In short, market failuresare pervasive.

There were two other implications. The first wasthe non-decentralizability of efficient market solu-tions. The notion that one could decentralize deci-sion making to obtain (Pareto) efficient resourceallocation is one of the fundamental ideas in eco-nomics. Greenwald and Stiglitz showed that thatwas not in general possible. Again, a simple exam-ple illustrates what is at issue. An insurance compa-ny cannot monitor the extent of smoking, which hasan adverse effect on health. The government cannotmonitor smoking any better than the insurancecompany, but it can impose taxes, not only on ciga-rettes, but also on other commodities which arecomplements to smoking (and subsidies on substi-

tutes which have less adverse effects.)94 Earlierwork with Bravernan [1982] had shown the conse-quences of this non-decentralizability, the interlink-age of land, labor, and credit markets in agrarianmarkets of developing countries.

Markets are also interlinked over time. Intertem-poral linkages impair the efficacy of competitiveprocesses, as we have already noted. Standard the-ory stated that if an employer does not treat anemployee well, he simply moyes to another firm.But informational asymmetries impair labor mobil-ity, partially locking the employee into his employ-er, or the borrower into his creditor.95 While withperfect information and perfect markets, some ofthe consequences of this reduction in ex post com-petition could be corrected by the intensity of exante competition, there is little reason to believethat is in fact the case.96

One of the sources of the market failures isagency problems, such as those which arise whenthe owner of land is different from the person work-ing the land. The extent of agency problems-andtherefore of market failures-thus depends on thedistribution of wealth, as we noted earlier in ourdiscussion of sharecropping. It is simply not thecase that one can separate out issues of equity andefficiency.97

Moreover, the notion that one could separate outissues of equity and efficiency also rested on theability to engage in lump sum redistributions. Butas Mirrlees [1971] had earlier pointed out, withimperfect information, this was not possible; allredistributive taxation was distortionary. But thishad important implications for a wider range ofpolicies beyond simply the design of tax structures.It meant that interventions in the market whichchanged the before tax distribution of income couldbe desirable, because they lessened the burden onredistributive taxation.9" Again, the conclusion: thesecond welfare theorem, effectively asserting theability to separate issues of distribution and effi-ciency, was not true.99

In effect, the Arrow Debreu model had identifiedthe single set of assumptions under which marketswere (Pareto) efficient. There had to be perfectinformation, or, more accurately, i-nformation(beliefs) could not be endogenous, they could notchange either as a result of the actions of any indi-vidual or fimn, including investments in informa-tion. But in an information economy, a model

Vol. 47, No. 2 (Fall 2003) 25

which assumes that information is fixed seemsincreasingly irrelevant.

Dysfunctional institutions

As the theoretical case that markets in whichinformation was imperfect were not efficientbecame increasingly clear, several arguments wereput forward against government intervention. Onewe have already dealt with: the government toofaces informational imperfections. But our analysishad shown that the incentives and constraints facinggovernment differed from those facing the privatesector, so that even when government faced exactlythe same informational constraints, welfare couldbe improved upon."°

There was another argument, which held up nobetter. The existence of market failures-absent orimperfect markets-does give rise to non-marketinstitutions. The absence of death insurance gaverise to burial societies. Families provide insuranceto their members against a host of risks for whichthey either cannot buy insurance, or for which theinsurance premium is viewed to be too high. But in

what I call the functionalist fallacy, it is easy to gofrom the observation that an institution arises to ful-fill a function to the conclusion that actually, inequilibrium, it serves that function. Those who suc-cumbed to this fallacy seemed to argue that therewas no need for government intervention becausethese nonmarket institutions would "solve" themarket failure, or at least do as well as any govern-ment. Richard Arnott and I [199la] showed that, tothe contrary, non-market institutions could actuallymake matters worse. Insurance provided by thefamily could crowd out market insurance; insurancecompanies would recognize that the insured wouldtake more risk because they had obtained insurancefrom others, and accordingly cut back on theamount of insurance that they offered. But since thenon-market (family) institutions did a poor job ofdivesting risk, welfare was decreased."0 '

The Amott-Stiglitz analysis reemphasized thebasic point made at the end of the last subsection: itwas only under very special circumstances thatmarkets could be shown to be efficient. Why thenshould we expect an equilibrium involving non-market institutions and markets to be efficient?

THE AMERICAN ECONOMIST26

COPYRIGHT INFORMATION

TITLE: Information and the Change in the Paradigm inEconomics, Part 1

SOURCE: Am Econ 47 no2 Fall 2003WN: 0328804680002

The magazine publisher is the copyright holder of this article and itis reproduced with permission. Further reproduction of this article inviolation of the copyright is prohibited. To contact the publisher:http://www.cba.ua.edu/~ode/

Copyright 1982-2004 The H.W. Wilson Company. All rights reserved.