- Bounded Dunn 01 r i

8/3/2019 - Bounded Dunn 01 r i

1/21

Journal of Post Keynesian Economics /Summer 2001, Vol. 23, No. 4 567

2001 M.E. Sharpe, Inc.

01603477 / 2001 $9.50 + 0.00.

STEPHEN P. DUNN

Bounded rationality is not

fundamental uncertainty:

a Post Keynesian perspective

Economics that assumes transformation, change, can never be as tidy,

secure and elegant as that which assumes and cultivates unchanging ver-

ity. J.K. Galbraith (1991, p. 41)

The analytical consequences of accepting the fact of uncertainty remain,

on the whole, unexplored. Tony Lawson (1985, p. 909)

It is increasingly acknowledged within the economics literature that theconcept of uncertainty emphasized by Keynes and by many PostKeynesians means something more than probabilistic mathematical risk.However, it is far from clear what this something more means. As

Davidson (1988, p. 159) points out, many economists continue to dis-miss fundamental uncertainty as an ill-defined notion which simplymuddies the water of scientific investigationa concept that is anti-theoretical and ultimately results in nihilism. Rosser (1999, pp. 182,185; 2001) has recently alluded to the differing conceptual basis for riskand uncertainty in different theories but does little to clarify the result-

ing confusion.In order to countenance this state of affairs and to promote a broader,

more scientific discussion of Post Keynesian ideas, Paul Davidson (19821983, 1988, 1991, 1994, 1996), in a series of articles and books, has

The author is an Economic Advisor in the Strategy Unit, Department of Health,Whitehall, a Senior Research Fellow at Staffordshire University, and a researcher atthe University of Leeds, United Kingdom. He would like to thank Vicky Chick,Giuseppe Ciccarone, Paul Davidson, Sheila Dow, Peter Earl, Keith Glaister, GeoffHodgson, Andrew Mearman, Barkley Rosser, Malcolm Sawyer, and the participantsof the Sixth Post Keynesian Workshop, Knoxville, Tennessee, 22nd28th June, 2000for theircomments and discussions on earlier versions of this work. He gratefullyacknowledges the financial support of the Centre for Full Employment and Stability,University of Missouri Kansas City (UMKC) and theJournal of Post Keynesian

Economics, which enabled him to attend the Sixth Post Keynesian Workshop. Theusual disclaimer applies.

02 dunn.p65 04/02/2001, 10:53 AM567


2/21

568 JOURNAL OF POST KEYNESIAN ECONOMICS

attempted to precisely define uncertainty. Davidson has technically de-lineated situations of risk and fundamental uncertainty, respectively,in terms of the statistical concepts of ergodic and nonergodic processes,a language that is familiar to economists and underpins their differentmethodologies, in order to promote a wider appreciation of the salienceof this distinction. Recently he has attempted to broaden this definitionby distinguishing between immutable and transmutable economic pro-

cesses (Davidson, 1996).Davidsons concept of fundamental uncertainty is conceptually dis-

tinct from the behavioral uncertainty to which proponents of thebounded rationality approach to the study of economic decision-makingrefer.1 Bounded rationality refers to behavior that is intendedly rational

but only limitedly so (Simon, 1961, p. xxiv). It is used to designaterational choice that takes into account the cognitive limitations of thedecision-makerlimitations of both knowledge and computational ca-pacity (Simon, 1987b, p. 266). Bounded rationality, however, relates to

the behavioral characteristics of agents, whereas fundamental uncer-tainty relates to the essential unknowability of the future, to creativehuman agency and the unique nature of unfolding time. Nevertheless,the importance of this distinction is not generally recognized by econo-mists. In a world in which the present and the future are truly open andtransmutable, no matter how much rationality we impute to agents, the

course of outcomes remains unpredictable. If the world including the

1 It could be argued that in addition to behavioral and fundamental uncertainty,which we discuss in the article, we have omitted to discuss a third type of uncertainty,which we shall label Richardsonian uncertainty after G.B. Richardson (1960).Richardsonian uncertainty refers to the impossibility of forming rational expecta-tions under conditions of perfect competition where actors remain independent intheir choices. The incentive to invest requires in part that either there are limits to thecompetitive supply or that the profit opportunity is not known to anyone else exceptthe investor. However the conclusion that firms have no basis for expectationformation and investment decision, even if consumer preferences are known andstable requires that a full set of Arrow-Debreu (futures) markets exist. One couldargue that futures markets do not work effectively as bounded rationality precludesthe successful global calculations that underpins the effective operation of futuresmarkets (Hodgson, 1988, pp. 189190). Alternatively, futures markets cannot exist atall in the case of not-yet-invented commodities, which do not exist because they havenot been created yet, a consideration, as we shall see, that characterizes transmutablenonergodic environments. As Hodgson (1988, p. 190) notes: Richardsons argumentdepends implicitly upon the assumption that [fundamental] uncertainty is present, orthat rationality is bounded, or both. It is for these reasons that we do not refer toRichardsonian uncertainty in the main text as it is subsumed under the concepts ofeither behavioral and fundamental uncertainty or both.

02 dunn.p65 04/02/2001, 10:53 AM568


3/21

BOUNDED RATIONALITY IS NOT FUNDAMENTAL UNCERTAINTY 569

future is immutable and closed then unless arbitrary and ad hoc con-straints are imposed, agents following rules or satisficing procedurescan come to be substantively successful.

As we shall see confusion exists because of the similar implications offundamental uncertainty and bounded rationality. Both approachesseemingly imply a conception of choice that is meaningful, genuine,and realisticthat is, choice matterscan make a difference to out-

comes, and corresponds to the generalized features of experience. Bothapproaches seemingly imply a rejection of the maximization hypothesis(Rosser, 2001). This has led to fundamental uncertainty being sub-sumed in and conflated with the concept of bounded rationality. Theconsequence is that this distinction has not been appreciated, with the

result that the richness and distinctness of concepts such as fundamen-tal uncertainty are much diminished.

This article argues that the distinction between ergodic and nonergodicprocesses made by Paul Davidson is instructive in providing a technicalapproach that underscores the distinction between deterministic com-plexity and fundamental uncertainty. We shall argue that discussionsof bounded rationality are implicitly conducted in terms of the ergodicaxiomand that in the long run, bounded rationality collapses into thesubstantive rationality of preprogrammed choices. Imputing real andmeaningful creativity to agents implies however there are no pre-

programmed choices to collapse upon and agents are truly uncertain.This distinction is crucial and, as we shall see, yields different conclu-

sions and implies distinct modes of analysis.The article begins by outlining some of the similarities between the

concepts of bounded rationality and fundamental uncertainty that un-derscores this conflation. In the subsequent sections we consider theconcepts of ergodicity and nonergodicity and their nexus to boundedrationality and fundamental uncertainty. We conclude by suggesting thatthose economists interested in the study of decision-making under com-

plexity and uncertainty should recognize this salient distinction in theiranalyses.

Bounded rationality and fundamental uncertainty compared

Fundamental uncertainty and the concept of bounded rationality, asdeveloped by Herbert Simon (1957, 1959, 1976), appear to hold quitesimilar implications. Simon (1987a, p. 222, emphasis added) argues thatthe term bounded rationality has been proposed to denote the wholerange of limitations on human knowledge and human computation that

02 dunn.p65 04/02/2001, 10:53 AM569


4/21


preventeconomic actors in the real world from behaving in ways thatapproximate the predictions of classical and neoclassical theory: includ-ing the absence of a complete and consistent utility function for order-ing all possible choices, inability to generate more than a small fractionof the potentially relevant alternatives, and inability to foresee the con-sequences of choosing the alternatives. Similarly, the notion of funda-mental uncertainty has been advanced by Post Keynesians as a direct

challenge to the predictions of classical and neoclassical theory and theability to form rational expectations, or form a complete transitive or-dering over all possible alternative future outcomes (Davidson, 1996).

In those situations in which either bounded rationality or fundamen-tal uncertainty are present then decisions have to be made; the past,

either because it cannot be fully comprehended or because it may sub-stantially differ from a yet to be originated future, can only provide alimited guide to future events. Maximization or optimization is simplynot possible in the short run if the world is either complex or open and

fundamentally uncertain.2 The implications should be immediatelyapparent, both under fundamental uncertainty and bounded rationalitydecisions are taken in light of the constraints of the past but are not deter-mined by it (Garner, 1982). Both imply a conception of decision-makingthat is rooted in the actions of agents, which relates to their computa-tional limitations, or their capacity for creative and imaginative thought.

It is for such reasons that some have felt compelled to argue that thebounded rationality hypothesis of Herbert Simon accords well with

George Shackles views [on fundamental uncertainty] (Ford, 1993, p.692). Simon (1972, p. 170), for example, has suggested that the distinc-tion between deterministic complexity and uncertainty is inessential.3

He notes that uncertainty about the consequences that would follow

2 This discussion is also suggestive that both the fundamental uncertainty andbounded rationality imply that the competitive processes may be evolutionary incharacter (cf. Hodgson, 1993, 1994).

3 Rosser (1999), in a recent article in theJournal of Economic Perspectives, surveysrecent developments in complexity theory as it pertains to economics. This is quite adistinct reformulation of complexity than that understood by Simon (1972). Rosser(1998, 1999; Hayek, 1967; Leijonhufvud, 1993) notes that deterministic complexitymodels impute epistemological problems to agents on account of computationallimitations. Omnipotent parties are not confronted with epistemological problems,however. As Davidson (1991, 1996) has pointed out, deterministic models ofdecision-making (stochastic or otherwise) are conducted in logical time and requireSavages ordering axiom for decision-making, that is, the presumption, at least inprinciple, that each agent can make a transitive ordering of all possible alternativeoutcomes. Of course if the system is too complex for the agent to have complete

02 dunn.p65 04/02/2001, 10:53 AM570


5/21


from each alternative, incomplete information about the set of alterna-tives, and complexity preventing the necessary computations from be-ing carried out . . . tend to merge: . . . uncertainty, whatever its source, isthe same (Simon, 1972, pp. 169170).4 Radner (1970, p. 457) has simi-larly conflated problems of creativity and imagination with issues ofcomputation and adopted the assumption of bounded rationality givensubstantive rationality requires that economic agents possess capabili-

ties of imagination and calculation that exceed reality by many orders ofmagnitude. Clearly if the environment is complex, yet closed and de-terministic, and decision-makers are limited in their information-han-dling capacities, then it may be unlikely that agents will have, or be ableto process, all the information they require to make a globally optimal

decision (Simon, 1972; Rosser, 1999, 2001).However, it is perhaps because of the similarities in the conclusions

reached using these theoretically distinct concepts that they are typi-cally integrated and fused together in a theoretically confusing manner

with the salience and distinctiveness ofboth approaches not recognized.Since both bounded rationality and fundamental uncertainty imply, atleast in certain areas of decision-making, agents are unable to makeoptimizing choices whose payoffs are expected in the future, it is easyto conflate under bounded rationality with an analysis of decision-mak-ing under uncertainty. Bounded rationality, however, relates principally

to the epistemological problems that confront agents in an ontologicallyclosed world. It implies that decisions (including the deferring of choice)

need to be made. If the environment cannot be fully understood and itsimplications fully computed then decisions have to be made, it is forthis reason that agents engage in satisficing behavior.5 Maximization, ofutility, profit, or whatever, is simply not possible in the short run (Loasby,

information then they can only transitively order those outcomes they have informa-tion about. However, unless ad hoc and arbitrary epistemological bounds are placedupon agents then they can in principle learn and discover the ontology that governsthe system. See also footnotes 8 and 20.

4 One can find many instances of this conflation in Simon. For example, withRichard Cyert he has suggested that, Since uncertainty is a time-honored word ineconomics, let us gather all of these imperfections and limits upon rationality

incomplete knowledge, inadequate means of calculationunder the umbrella ofuncertainty (Cyert and Simon, 1983, p. 104).

5 Simon (1959, p. 263) notes that Models of satisficing behavior are richer thanmodels of maximizing behavior, because they treat not only of equilibrium but of themethod of reaching it as well. Psychological studies of the formation and change ofaspiration levels support propositions of the following kinds. (a) When performance

02 dunn.p65 04/02/2001, 10:53 AM571


6/21


1967). Choice and decision for Simon is identified with the informa-tional processing abilities of agents and the shortfall with respect todeterministic omniscience. Agents cannot study the data and in light ofthe structure of the problem make the substantively rational decisionthat is embodied in the closed deterministic equations that characterizethe system (Rosser, 1999).

In sharp contrast, Post Keynesians argue that in an uncertain environ-

ment, where the past presents a limited and narrow guide to a future thathas yet to be created, crucial decisions can and have to be made(Davidson, 19821983, 1988, 1991, 1996). In this view, however, agentsare truly uncertain in that there are no deterministic laws or equations todiscover. The ontology of the system is open and problems of acting

cannot be reduced solely to problems of epistemology (Lawson, 1997).It is thus perhaps timely that we turn to the distinction between ergodicand nonergodic processes made by Paul Davidson, which will allow usto elaborate the reasons for this conflation and highlight the salience of

the distinction between bounded rationality and complexity and funda-mental uncertainty.6

Immutability

Ergodic theory has been explicitly developed in the theory of stochasticprocesses although the term was borrowed from statistical mechanics(Parry, 1987). Samuelson (1969) has argued that economics claim to be

scientific rests on the acceptance of the ergodic hypothesis. GivenSamuelsons stature within the economics profession it is not unreason-able to suggest that the conventional wisdom concerning economicknowledge about the future relationship among observable economic

falls short of the level of aspiration, search behavior (particularly search for newalternatives of action) is induced. (b) At the same time, the level of aspiration beginsto adjust itself downward until goals reach levels that are practically attainable. (c) Ifthe two mechanisms just listed operate too slowly to adapt aspirations to performance,emotional behaviorapathy or aggression, for examplewill replace rationaladaptive behavior.

6 Lawsons (1997) distinction between open and closed systems is broadly similarto Davidsons distinction between immutable (ergodic) systems and transmutable(nonergodic) systems. Both approaches can be marshaled to elaborate the distinctionbetween bounded rationality and fundamental uncertainty. However, in order tofacilitate debate, Davidsons approach, which is grounded in more familiar stochasticconcepts, is drawn on upon here.

02 dunn.p65 04/02/2001, 10:53 AM572


7/21


variables is defined in terms of the probability distribution functionsand statistical averages generated by ergodic process (Davidson, 1988,p. 331). Davidson has concluded that the implication of this is that theinverse of knowledge, unknowledge, that is (fundamental) uncertainty,must be expounded in terms of the ergodic theory.

All stochastic processes yield time series data (realizations) that allowthe construction of averages such as the mean or standard deviation.

These averages constitute our empirical knowledge about past and cur-rent real world relationships. Time averages are calculated from time-series data, that is, observations that relate to a period of calendar time,while space averages are computed from cross-sectional data, that is,observations that relate to a given point in time across realizations.

Davidson (1988) notes that we are dealing with an ergodic stochasticprocess if: (a) for infinite realizations the time and space averages coin-cide, or (b) for finite realizations the time and space averages converge(with a probability of one) as the number of observations increases.

This means that space or time averages calculated from past realiza-tions collapse onto the objective probability distribution that describesall possible (past, present, and future) realizations and that these timeand space averages form reliable estimates of (and govern) both currentand future events (Davidson, 1988, p. 331).7 The past represents a samplefrom the futureor to put it another way, the past reveals the future.However, while the concept of ergodicity is generally understood to re-late to stochastic processes, in a wider sense it implies, as Davidson

(1996, pp. 480481) points out, the presumption of a programmed sys-tem where the past, present, and future reality are predetermined whetherthe system is stochastic or not.8

If the world confronted by decision-makers is ergodic then to ascribeany salience to history is fallacious. The future is merely the statisticalreflection of the past. Economic activities are timeless and immutable(Davidson, 1994, p. 90). If relationships between economic variables

7 One must be careful not to conflate the concepts of stationarity and ergodicity. Astochastic process is stationary if the estimates of time averages do not vary with theperiod under observation. Since some stationary stochastic processes are nonergodic,

that is, limit cycles, nonstationarity is not necessary for nonergodicity. But since allnonstationary processes are nonergodic nonstationarity is a sufficient condition.

8 Davidson (1991, p. 178) has explicitly expanded the concept of nonergodicity toincorporate situations that are not defined with respect to stochastic processes. Asnoted in footnote 4, the ordering axiom plays the same role in deterministic models asthe ergodic axiom in stochastic models.

02 dunn.p65 04/02/2001, 10:53 AM573


8/21


are ergodic, then time in any substantive sense is unimportant.

9

IndeedSamuelsons (1969) stated aim in introducing the ergodic hypothesiswas to remove a concern for path dependence and history and placeeconomics firmly in the scientific realm. However, the presumption ofan immutable system in which history is predetermined, stochasticallyor otherwise, as both Shackle (1972) and Loasby (1976) repeatedly pointout, makes a mockery out of any notion of choice, especially choice

regarding the future. If the economic world is immutable and ahistoric,then, in the long run, outcomes are preprogrammed and independent ofthe decisions made by economic agents.10 Choice is neither genuine normatters. It cannot make a difference in the long run. Consequently, anyattempt to assign any uniqueness and importance to human agency and

history is fallacious. The implication of this is that there can be no igno-rance (in the long run at least, see below) as to the (probable) futureoutcomes of an ergodic system as long as one is willing to allocate re-sources so as to ascertain knowledge regarding the processes (stochastic

or not) that govern the system. Ignorance can ultimately yield to omni-science. Maximization is theoretically possible in the long run. It is tothis that we now turn.

Immutability and bounded rationality

The reason for this near universal acceptance of the importance of theconcept of bounded rationality among economists relates to the conclu-

sion reached by Loasby (1976) and Davidson (1977) that there is noneed for a theory of money, or any other salient institutions for thatmatter, in general equilibrium models. Even probabilistic versions ofgeneral equilibrium theory, which allow informational problems of astylized and restricted kind, provide no reason for the institution of money

given the existence of a competitive general equilibrium price vector

9 Expanding on this, Davidson (1988, p. 332) refers to Billingsley (1978, p. 1)[who] states, if the laws governing . . . change remain fixed as time passes [then]ergodic theory is a key to understanding these fluctuations. Whenever the passage oftime does not affect the set of joint probability laws governing experimentation(outcomes), then the assumption of ergodicity permits regularities to be perceived

from what might at first sight be patternless fluctuations (Billingsley, 1978, p. 2, seealso pp. 6065). This pattern of regularities can then be reliably projected into thefuture.

10 Ergodic theory, as Samuelson (1969, p. 184) points out, implies that if the stateredivided income each morning, by night the rich would also be sleeping in their bedsand the poor under the bridges.

02 dunn.p65 04/02/2001, 10:53 AM574


9/21


(Arrow and Hahn, 1971). The conclusion drawn, perhaps incorrectly, isthat because the probabilistic version of general equilibrium theory, whichcannot describe institutions such as money (and firms), relies heavily onthe assumption that agents are unboundedly rational, ergo any modelthat attempts to explain them must relax this assumption.11 It is at thispoint that bounded rationality enters into economists theorizing (Sargent,1993; cf. Sent, 1997).

Bounded rationality makes provision for an analysis that is more con-gruent with experience and, as noted above, the conclusions appear, su-perficially at least, to yield conclusions that are consistent with a stresson fundamental uncertainty. Bounded rationality, when introduced intothe analysis, describes situations whereby agents are unable to provide

for an exhaustive list of (future) states of the world and possible coursesof action that relate to the problem at hand because of limitations intheir computational (and linguistic) ability. According to bounded ration-ality proponents, individuals making decisions in a complex world do

not possess the computational skills necessary to process all of the rel-evant information available to themit is for this reason that they satisficebecause they do not possess the wits to maximize (Simon, 1957).However, while bounded rationality proponents discard the strong as-sumption retained by rational expectation and general equilibrium theo-rists, that there are no bounds to agents computational (or linguistic)

capacities, they implicitly retain the assumption that the objective con-ditions that describe economic processes are ergodic and deterministic.

The conventional approach to bounded rationality is conducted implic-itly in terms of ergodic theory, or more specifically under the assump-tion of an immutable environment.

To suggest that the concept of bounded rationality is generally linked toergodic immutable systems is not unreasonable. As Loasby (1989, p. 143)

11 Simon (1959, p. 254) has revealingly commented that: Economists have beenrelatively uninterested in descriptive microeconomicsunderstanding the behavior ofindividual economic agentsexcept as this is necessary to provide a foundation formacroeconomics. The normative microeconomist obviously doesnt need a theory ofhuman behavior: he wants to know how people ought to behave, not how they dobehave. On the other hand, the macroeconomists lack of concern with individualbehavior stems from different considerations. First, he assumes that the economicactor is rational, and hence he makes strong predictions about human behaviorwithout performing the hard work of observing people. Second, he often assumescompetition, which carries with it the implication that only the rational survive. Thus,the classical economic theory of markets with perfect competition and rational agentsis deductive theory that requires almost no contact with empirical data once itsassumptions are accepted.

02 dunn.p65 04/02/2001, 10:53 AM575


10/21


has remarked, the writings of Herbert Simonthe progenitor of the con-cept of bounded rationalitysometimes give the impression that we dolive in a fully defined [immutable] system, if we only had the wit to under-stand it.12 This is clearly suggestive of an ergodic (immutable) world inwhich the future ispotentially knowable in the sense that the realizationsof the past are a reliable guide to future events and these may, if resourcesare allocated to their understanding, form the basis for decision-making in

the long run. The future already existsit is embedded in the past.The assumption of bounded rationality means that agents are unable

to engage in maximizing behavior in the short run. Agents may be un-able to form (rational) expectations about the future that are efficient,unbiased and without persistent errors in the short run because of their

limitations in ability, that is, as a result of constraints in agents comput-ing power. Agents are prevented from using existing market data to ob-tain short-run reliable knowledge regarding the future nature of alleconomic variables. However, over time as the ergodic nature of the

environment is more fully understood, that is, as the potential for sur-prise diminishes as one learns about the objective probability distribu-tions that describe the ergodic environment, or in a broader sense as onelearns about the immutable laws that govern reality,13 then expectationsabout the future, in the Marshallian long rundefined as the asymp-totic end state of a process of learningwill facilitate more efficient

decisions to be made about the future.14 As Cyert and Simon note: un-

12 Loasby (1989, p. 146) provides further support for this assertion when he statesthat Simons own analysis no doubt helps buttress his belief in our ability to solveproblems at least as fast as they are created; but although he recognizes the practicalimportance of the varying speeds with which different interactions take effect,nevertheless, for one who prefers process to equilibrium as an organizing principle, heseems to have very little sense of the significance of time in human affairs. Comparehis writings with those of Marshall. Shackle, or the later Hicks, and one is immedi-ately conscious of a great difference in attitude and style. In the sense alluded toabove, time or history has no substantive role in an immutable world.

13 As Davidson (1996, p. 486) notes typically in bounded rationality models agentsform subjective expectations (usually, but not necessarily in the form of Bayesiansubjective probabilities). In the short-run, subjective probabilities need not coincidewith the presumed immutable objective probabilities. Todays decision-makers,therefore, can make short-run errors regarding the uncertain (i.e., probabilistic risky)

future. Agents learn from these short-run mistakes so that subjective probabilities ordecision weights tend to converge onto an accurate description of the programmedexternal reality.

14 Many parallels could be drawn here with Days (1967) discussion of theconvergence of satisficing to marginalism given a stable (ergodic) environment, whichagents can learn about in the light of outcomes and feedback from the environment

02 dunn.p65 04/02/2001, 10:53 AM576


11/21


certainty gives way to certainty and as knowledge increases, the behav-ioral rules change and move closer to those derived by a priori reason-ing (Cyert and Simon, 1983, p. 105).

However, this need not imply that decision-making becomes inevita-bly more complex. If the environment is one in which the potential forsystematic error is diminishing, that is, as the nature of the ergodic pro-cess becomes more fully understood, then it is also likely to be one in

which behavior becomes increasingly routine. As Davidson (1996) pointsout, routine, repetitious activities are the hallmark of ergodic environ-ments.15 In the long run, activities become increasingly routineevery-thing is done the way it was yesterday. The assumption of boundedrationality cannot mean in any substantive sense that systematic mis-

takes will continue to be made in the long run if based on an ergodicconception of economic processes.16Ex ante maximization ultimatelyreplaces satisficing, resulting in the concept of the decision being emp-tied of all its substantive content in the long run. Boundedly rational

decisions, ultimately, are neither crucial nor novel.17

and subsequently revise their decision rules. Proponents of Simons approachpresumably would counter this argument by suggesting that bounded rationality refersto computational capacity and not to the learning process. However the rationalaspect to the concept of bounded rationality is suggestive that decision-makers willattempt to revise and devise more successful decision rules, especially if they find

themselves making persistent mistakes, that, if the environment is stable (ergodic),will converge on omniscience in the long run (as defined above). To deny this woulddeny the rational aspect of bounded rationality that is highly coveted.

15 One should distinguish between routine that is embedded in the preprogrammeddeterministic models and convention that emerges as a response and strategy forcoping with the uncertainty. The concept of convention implied in the latter iscontingent upon the agency of individuals and not system characteristics.

16 In the presence of a lack of full information and an inability to process all theavailable information, agents would not know what a maximizing decision wouldbein the short run. While agents can judge whether the existing outcome issatisfactory, they are precluded in the short-term of knowing whether it is a maximiz-ing one. However, as further information is acquired and as processing capacities aremodified and advanced, agents may learn that an outcome previously judged assatisfactory can be improved upon. This feedback loop underpins the rationaldimension to the concept which is so cherishedin Simons (1957, 1959) nomencla-

ture the aspiration level adjusts to new information gleaned resulting in an improvedbehavioral response to a complex environment.

17 As Shackle (1972, p. 426) notes: The most dramatic and spectacular secret ofsuccess is novelty, and novelty is that which an infallible algorithm must, by defini-tion, exclude. This also has concomitant implications for theories of competition,bargaining, games, models of dynamic learning, and the like (see also Loasby, 1989).

02 dunn.p65 04/02/2001, 10:53 AM577


12/21


What is the mechanism that assures the learning process occurs? Con-sistent with conventional wisdom, some form of market-based Darwin-ian natural selection is invoked to suggest that those decision rules androutines that persist are in some sense optimal (cf. Friedman, 1953).18

In this ergodic Darwinian world, agents learn from their short-runmistakes and make provision for them so that their subjective assess-ment of the environment reveals (converges on) the objective ergodic

environment (Davidson, 1996, p. 486). Those that do not learn the rel-evant routines or probability functions that govern the immutable pro-cess will fall foul of the market process.19

Transmutability

According to Davidson, the world inhabited by economic agents is a

world in which history need notbe governed by stochastic processes orimmutable laws (cf. Shackle, 1972). Davidson defines fundamentaluncertainty about the future course of events in terms of the absence ofgoverning ergodic processes, that is, in terms of a creative transmut-able environment. Davidson has labeled this situation nonergodic. Innonergodic environments even ifagents have the ability to collect and

successfully process all the information relating to past and current out-comes, then this existing (market) information does not, and can not,provide reliable data for forecasting future outcomes and learning aboutthe future (Davidson, 1996, p. 482). There are no social and economic

laws to learn.20 As Davidson (1991, p. 133) remarks:

If, however, true uncertainty conditions prevail [i.e., nonergodicity] in

certain decision making areas, then at least some economic processes are

18 Hodgson (1993, 1994) points out appeals by economists to simplistic evolution-ary arguments in extolling the virtues of the competitive process are common.Hodgson demonstrates, however, the view that evolutionary processes lead generallyin the direction of optimality and efficiency is in fact erroneous.

19 Davidson (1996, p. 486) notes: Those agents whose subjective probabilities donot converge on the objective probabilities will make persistent systematic forecastingerrors. The market embodies some form of a Darwinian process of natural selectionthat weeds out the persistent error-makers who make inefficient choices until, in the

long-run, only agents who do not make systematic errors remain.20 The discussion is underpinned by a distinction between ontologically certain

(closed immutable systems) and ontologically uncertain (open transmutablesystems) environments. Models based on an ontologically certain view of theworld, such as rational expectations models (both New Classical and New Keynesian)make knowledge (epistemological) claims regarding the informational sets that agents

02 dunn.p65 04/02/2001, 10:53 AM578


13/21


such that expectations based on past probability distribution functions

can differ persistently from the time averages that will be generated as

the future unfolds and becomes historical fact. In these circumstances,

sensible economic agents will not rely on available market information

regarding relative frequencies, for the future is not statistically calculable

from past data and is truly uncertain.

In an open nonergodic world sensible agents will recognize that the

future is one that can significantly differ from past experience and presentexpectations.21 In a nonergodic world, where statistical distributions ofthe past provide a limited guide to the course of future events, agents aretruly uncertain, as there currently does not exist information that willenable them to discover the future. Decisions have to be made, including

the maintenance of convention, and choice is genuine. As Robinson (1980,p. 219) pointed out Today is influenced, but not completely bound bythe past. Any action or decision taken today is either the result of blindhabit and convention or it is directed towards its future, which cannot yet

be fully known. As a result agents will have to invent or create thefuture by themselves by their actions within evolving and existingorganisations (Davidson and Davidson, 1984, pp. 329330, emphasisadded). Agents recognize that the environment in which they make deci-sions is in some dimensions characterized by the absence of governingergodic processes, that is, it is uncertain, and thus transmutable or cre-

ative. This creative economic reality involves an uncertain future thatcan be enduringly transformedby the purposeful and intentional actions

of individuals, groups such as unions and cartels or governments, oftenin ways not completely foreseeable by the creators of change (Davidson,1996, p. 482, emphasis added; see also Shackle, 1955; Lawson, 1997).

Transmutability and creative crucial decisions

Decision-making in situations where information gathered from the pastprovides an insufficient basis in which to assess future outcomes is of

are assumed to possess in order to obtain their short-run policy implications. Modelsthat suggest the future is not completely known but can be learned, that is, agents are

subject to bounded rationality, invoke epistemological uncertainty but eschew acreative and transmutable ontology. They are not however, founded on a creative andontological open-system view of uncertainty.

21 This forms the basis for Davidsons (19821983) assertion that the rationalexpectations hypothesis is a fallacious foundation for the study of crucial decision-making processes. (See also Davidson, 1996, p. 493.)

02 dunn.p65 04/02/2001, 10:53 AM579


14/21


necessity creative. In those situations where a decision cannot be un-done, where the economic landscape is forever changed, when the cir-cumstances of the decision are non-repeatable, then the decision is crucial(Shackle, 1955). Crucial decisions refer to those non-routine situationsthat take place in historical time, circumstances in which re-contractingis expensive (unlike in ArrowDebreu type constructions).22 Crucialdecisions made today, that turn out to be wrong tomorrow, cannot be

remade without cost. In these (crucial) situations decision-making agentsare tied to the choices they make. The future does not wait for its con-tents to be discovered, but for that content to be originated (Shackle,1980 quoted in Davidson 19821983, p. 192). The nonroutine, non-re-peatable nature of such crucial choices does not imply however that

such choices are rare. Davidson (1996, pp. 500501) notes that:Crucial choices are more common than one might expect; where there are

transactions costs: no decision is fully reversible. . . . Because of the sub-

stantial transactions costs involved in investment, production, and (at least)

big ticket consumption decisions, in these areas, agents are necessarily

married to their choices; decisions in these areas are normally crucial and

once an action is made, the possible future path is changed. In such a trans-

mutable world, he who hesitates regarding choices in these areas and de-

cides to remain liquid is saved to make a crucial decision another day.

As the discussion above intimates, the creativeness associated withcrucial decision-making is inextricably linked to the concept ofnonergodicity. The existence of crucial decisions represents a sufficientcondition for the existence of non-deterministic nonergodic processes.Accordingly, the demand for not spending ones income on todays prod-

ucts of industry, that is, the decision to remain liquid implies that in-come earners have a choice between employment-inducing demandand non-employment inducing demand (Hahn, 1977, p. 39). In otherwords, in a world of crucial decisions and fundamental uncertainty, SaysLaw is inapplicable and an involuntary unemployment equilibrium canexist, even in the long run.23 Agents possess the capacity to effect real

22 As noted by Davidson (1996, pp. 500501): Mainstream micro as well as macrotheorists ignore this element of cruciality in almost all decisions. Orthodox theorists

assume the ability to re-contract without costs if one does not initially trade at thegeneral equilibrium prices that embody the objective reality governed by the realparameters of a predetermined economic system.

23 As Arrow and Hahn (1971, p. 361) point out, if money (and therefore the demandfor liquidity) can affect real outcomes, then all general equilibrium existence theoremsare jeopardized.

02 dunn.p65 04/02/2001, 10:53 AM580


15/21


change and determine historical outcomes in a manner that is not preor-dained by the system dynamics. This transmutable principle that under-lies and binds the concepts of cruciality and nonergodicity.

Under conditions of nonergodicity or fundamental uncertainty futurestates of the world cannot be fully specified because they areyet to becreated. This is suggestive of the fact that future states of the worldcannot be anticipated. What has occurred in the past and the present

need not occur again in the future. Individual agents are ignorant of theavailable courses of action or of the extent of future states of the worldbecause of the irreversible and open-ended nature of time, because thefuture is transmutable, and not because of limitations in the processingabilities of economic agents (Dunn, 2000). Agents are architects of the

future. In a nonergodic world the future cannot be known prior to itscreation regardless of the processing powers we impute to agents. Timediffers from complexity in an essential manner. We can never know exante historys direction, and thus no matter how much information and

computational capacity a decision-maker has they can never ex ante pre-dict with (probabilistic) certainty the future. Agents, and groups of agents,make their own history. But the history created is one that was not nec-essarily intended (Shackle, 1972).

This discussion suggests an important clarification to how we under-stand and conceptualize decision-making. We may accept that it may be

reasonable to assume an unchanging (ergodic) economic reality in situ-ations of routine, repeatable decision-making (indeed this is a key as-

pect of learning by doing and bounded rationality approaches). In somesituations it might be reasonable to assume an unchanging economic(immutable) reality and that the future will mirror the past and the cur-rent situation.24

However relying on the past may be inappropriate and quite rigidapproach to theorizing in dealing with surprise and change, that is, un-der conditions of transmutable nonergodicity, even in the long run.25

24 It is often ignored that even Davidson himself entertains this possibility. He hasnoted that In the real world, some economic processes may be ergodic, at least forshort sub-periods of calendar time, while others are not. The problem facing every

economic decision-maker is to determine whether (a) the phenomenon involved iscurrently being governed by distribution functions which are sufficiently timeinvariant as to be presumed ergodicat least for the relevant future, or (b) nonergodiccircumstances are involved (1988, p. 163).

25 There is a sense in which the definition of the long run changes upon recognitionof the transmutable nature of economic processes. We return to this below. Clearly the

02 dunn.p65 04/02/2001, 10:53 AM581


16/21


Uncertainty resulting from situating decision-making into a flow ofhistorical time, constructed as a sequence of unique events, cannot leadto ex ante maximization even in the long run (Shackle, 1972). We cannever know the end or conclusion to which our decisions will actuallytake us and we have nothing to compare it against. In an ergodic worldex ante decisions about the future made by boundedly rational agentsmay be mistaken vis--vis omniscient parties, but only in the short run.

As noted above those agents that continue to make mistakes vis--visomniscient parties in the long run will be weeded out by some form ofDarwinian competitive market process, that is, there can be no mis-takes in the long run.

The concepts of novelty and creativity, however, undermine the no-

tion of rational maximization even in the long run.26

Although agentsmay strive for certain objectives when making decisions, such as profitmaximization, if the outcome is unsatisfactory in the sense that theirexpectations are disappointed and unfulfilled, it does not logically fol-

low that the initial decision was wrong. Considering agents informa-tional set at the time when the initial decision was made, one cannot apriori or a posteriori say whether the decision is wrong, or how theworld would have been different if another decision had been made.One might be disappointed, in terms of history, afterit has been created,but one cannot be mistaken vis--vis any omniscient parties because

there are no omniscient parties in an open and transmutable nonergodicworld to judge decisions. As Robinson (1962, p. 75) has pointed out, In

history, every event has its consequence, and the question, What wouldhave happened it that event had not occurred? is only idle speculation;in theory there is one position of equilibrium the system will arrive at,no matter where it starts. Similarly Shackle (1972, pp. 245246) notes:

Rational choice, choice which can demonstrate its own attainment of

maximum objectively possible advantage, must be fully informed

choice. . . . The paradox of rationality is that it must concern itself with

choosing amongst things fully known; but in the world of time, only this

is fully known which is already beyond the reach of choice, having al-

ready become actual and thus knowable. Rational choice, it seems, must

be confined to timeless matters.

26 It is for this reason that Davidson advances the notion of sensible expectations.Sensible agents recognize that the future substantially differs from the past andentertain this possibility in decision-makingrational Walrasian intertemporaloptimization is not possible and it is for this reason that it becomes reasonable to holdnoninterest-bearing assets such as money as a hedge against uncertainty.

02 dunn.p65 04/02/2001, 10:53 AM582


17/21


One cannot make a maximizing decision, that is, one that is demon-strably superior, in a nonergodic world (i.e., in history) (see Shackle,1972, pp. 84, 229230). How can choice be based on foreknowledge ofwhat that choice is called on to create? (Shackle, 1979, p. 58). If theactual consequences of choice in the present depend on the unknownand indeed unknowablefuture crucial decisions of others, economicagents can never be certain that they have adopted the best means to the

achievement of their ends. One does not know what the maximizingalternative is, rather one can only observe that the hypothesis that thedecision chosen was expected to be the maximizing decision. In whatsense can the superiority of choice be asserted or demonstrated? In anuncertain world the notion of maximizing is bereft of substantive con-

tentit is vacuous.Moreover, it should be apparent that the definition of the long run as

an asymptotic end state of a process of learning requires revision in aworld in which emergent novelty usurps the process of learning. It would

perhaps be more precise given a transmutable conception of economicprocesses to jettison the notion of a long run, as an end of an adjust-ment process construct, as a misleading guide to the study of open-ended processes. However, there is a sense in which the long run can beredefined a la Kalecki (1968, p. 263) as a slowly changing componentof a chain of short period situations; it has no independent entity and

thus linked to a transmutable conception of economic process.27 Indeedas Keynes (1982, CW, XXVIII, pp. 6264, quoted in Skidelsky, 2000, p.

33) put it: Life and history are made up of short runs. That is to say thelong run can only be defined in relation to the passing of actual histori-cal time as specified by the analyst after consideration of the processbeing investigated. Under this conception, the long run is nothing butthe historical accumulation of successive short runs. Choice under un-certainty, however, acquires substantive meaning, is genuine and doesmatter in the long run.28

27 As Sawyer (2000, p. xv) notes this laconic statement, on which Kalecki did notelaborate, can be interpreted as undermining the predominant [ergodic, closed system]equilibrium approach to economic analysis whereby there is a long-period equilib-rium around which the economy fluctuates or towards which the economy tends andwhich is unaffected by the short period movements of the economy.

28 Although strictly speaking one might accept that ultimately the concept of thelong run appears a misleading foci for historical analysis in which uncertainty andexpectations matter and should be rejected for a historical and processual approach tothe study of economic phenomenon. However, as noted above, such a view is impliedby Kaleckis reformulation. Nevertheless, the notion of a long run may be retainedto promote dialogue across different research traditions.

02 dunn.p65 04/02/2001, 10:53 AM583


18/21


Conclusions

In this article we have drawn out the differences between the concept of

bounded rationality and a technically defined concept of fundamentaluncertainty. As we have demonstrated uncertainty is not bounded ratio-nalitythey are not synonymous! Bounded rationality might appear amore realistic characterization of choice, especially as it seeminglyswitches the focus toward the actual procedures used in decision-mak-ing. However, such a methodological approach necessitates an implicittheoretical trade-off. Bounded rationality directs attention to the short-fall with respect to omniscience and the incremental efforts to movetoward it, and away from the problems of time. Bounded rationalityrefers to a specific view of behavior. Fundamental uncertainty refers to

a specific view of time. Economists forget this at their peril. FollowingPaul Davidson we would suggest that the impossibility of foreseeingfuture knowledge arises in contemplation of a nonergodic environmenteven ifagents could make full use of present knowledge. In a nonergodicenvironment the processing capacities of agents will affect the nature of

choice but not thefactof choice. An emphasis on choice under uncer-tainty and a stress on the dynamic evolutionary nature of the competi-tive process is not, therefore, tied to any (behavioral) conception of theprocessing abilities of agents (Dunn, 1999).

In an ergodic, immutable world, in the long run boundedly rationalagents would, or at least have the potential to, learn all the relevant future

intertemporal contingencies and incorporate them into maximizing be-havior so as to obviate the need for crucial creative decision-making inthe (very) long run. In an ergodic world agents who are boundedly ratio-

nal can learn all the relevant future intertemporal contingencies throughtime. The past provides a guide to the futurealbeit a complex one!Conversely, in nonergodic, transmutable environments all future statesof the world cannot be anticipated because they are yet to be created.

The distinction between bounded rationality and fundamental un-certainty suggests that one need not make any claims about the rational-

ity of agents vis--vis their ability to process information in rejectingthe axiom of maximization.29 In nonergodic transmutable environmentsthere is a fundamental asymmetry between the past and the future. Agents

cannot learn all the relevant future intertemporal contingencies throughtimethe past does not provide a guide to the future in any substantive

29 This does not entail that one must reject the concept of bounded rationality ratherthat the distinction be observed.

02 dunn.p65 04/02/2001, 10:53 AM584


19/21


manner. Agents can learn, however, that the past does not provide a reli-able guide to the future course of events, and thus recognize that theywill have to create, but cannot determine, the future by their actionswithin evolving and existing organizations and institutions. It is thisconclusion, that the future is substantially different from the past, thattime and history matter, which has generally been conflated with thebehavioral assumption of bounded rationality.

The result has been that the salience of this concept, not least therecognition of the fact the future can never be known in advance of itscreation, which has often been overlooked by economists. This distinc-tion is not a sterile methodological debate about epistemological ver-sus ontological uncertainty but rather it has important practical and

policy implications regarding the role of liquidity and twenty-first cen-tury international financial markets in determining the level of(un)employment. In an ergodic, immutable world in which agents areimputed with shortfalls in processing powers there is no need for li-

quidity or, as Hahn has described it in the aforementioned quote, a choicebetween employment-inducing demand and non-employment induc-ing demand. In such a world, Keyness theory of involuntary unem-ployment equilibrium is inapplicable. There is no role, in the long runat least, for government in providing a guiding influence on privatespending decisions (Keynes, 1936, p. 378). Conversely, in a world of

(nonergodic) fundamental uncertainty, agents can divert portions of theirincome into liquid assets, or nonemployment inducing demands in

Hahns terms, even in the long run, and Keyness General Theory ofEmploymentis relevant and applicable.

REFERENCES

Arrow, K.J., and Hahn, F.H. General Competitive Analysis. Edinburgh, UK: Oliverand Boyd, 1971.

Cyert. R.M., and Simon, H.A. The Behavioral Approach: With Emphasis onEconomics.Behavioral Science, 1983, 28, 95108. [Reprinted in P.E. Earl (ed.),

Behavioral Economics: Volume 1. Aldershot, UK: Edward Elgar, 1988.]

Davidson, P. Money and General Equilibrium. 1977. [Reprinted inMoney andEmployment, The Collected Writings of Paul Davidson, Volume 1. London:Macmillan, 1991.]

. Rational Expectations, A Fallacious Foundation for Studying Crucial

Decision-Making Processes.Journal of Post Keynesian Economics, Winter 19821983, 5 (2), 182197.

. A Technical Definition of Uncertainty and the Long-Run Non-Neutrality ofMoney. Cambridge Journal of Economics, 1988, 12 (3), 329337.

. Is Probability Theory Relevant for Uncertainty? A Post Keynesian Perspec-tive.Journal of Economic Perspectives, 1991, 5 (1), 129143. [Reprinted in

02 dunn.p65 04/02/2001, 10:53 AM585


20/21


Inflation, Open Economics and Resources: The Collected Writings of Paul Davidson,

Volume 2. London, Macmillan, 1991, pp 124138.]. Post Keynesian Macroeconomic Theory: A Foundation for Successful

Economic Policies for the Twenty First Century. Aldershot, UK: Edward Elgar, 1994.

. Reality and Economic Theory.Journal of Post Keynesian Economics,Summer 1996, 18 (4), 479508.

Davidson, P., and Davidson, G.S. Financial Markets and Williamsons Theory ofGovernance: Efficiency versus Concentration versus Power. Quarterly Review of

Economics and Business, Winter1984, 24. [Reprinted inMoney and Employment,The Collected Writings of Paul Davidson, Volume 1. London, Macmillan, 1991.]

Day, R.D. Profits, Learning and the Convergence of Satisficing to Marginalism.Quarterly Journal of Economics, 1967, 81, 302311. [Reprinted in P.E. Earl (ed.)

Behavioral Economics: Volume 1. Aldershot, UK: Edward Elgar, 1988.]

Dunn, S.P. Bounded Rationality, Fundamental Uncertainty and the Firm in theLong Run. In S.C. Dow and P. Earl (eds.), Economic Organisation and EconomicKnowledge: Essays in Honour of Brian Loasby, Volume Two. Cheltenham, UK:

Edward Elgar, 1999.. Fundamental Uncertainty and the Firm in the Long Run.Review ofPolitical Economy, 2000, 12 (4), 419433.

Ford, J.L. G.L.S. Shackle (19031992): A Life with Uncertainty. The EconomicJournal, May 1993, 103, 683697.

Friedman, M.Essays in Positive Economics. Chicago: University of Chicago Press,1953.

Galbraith, J.K. (1991) Economics in the Century Ahead. The Economic Journal,1991, 101 (404), 4146.

Garner, A.C. Uncertainty, Human Judgement, and Economic Decisions.Journal ofPost Keynesian Economics, 4 (3), 1982, 413424.

Hahn, F.H. Keynesian Economics and General Equilibrium Theory. In G.C.Harcourt (ed.), The Microfoundations of Macroeconomics. London: Macmillan, 1977.

Hayek, F.A. (1967) The Theory of Complex Phenomena. In F.A. Hayek (ed.),Studies in Philosophy, Politics, and Economics. London: Routledge & Kegan Paul,1967, pp. 2242.

Hodgson, G.Economics and Institutions: A Manifesto for Modern InstitutionalEconomics. Cambridge: Polity Press, 1988.

.Economics and Evolution: Bringing Life Back Into Economics. Cambridge,UK: Polity Press, 1993.

. Optimisation and Evolution: Winters Critique of Friedman Revisited.Cambridge Journal of Economics, 1994, 18 (4), 413430.

Kalecki, M. Trend and the Business Cycle.Economic Journal, 1968, 78, 263276.

Keynes, J.M. The General Theory of Employment, Interest and Money, London:Macmillan, 1936. [Reprinted in J.M. Keynes (1973). The Collected Writings of J.M.Keynes, Vol. VII, The General Theory of Employment, Interest and Money. London:Macmillan for the Royal Economic Society.]

. The Collected Writings of J.M. Keynes, Vol. XXVIII, Social, Political andLiterary Writings. London: Macmillan for the Royal Economic Society, 1982.

Lawson, T. Uncertainty and Economic Analysis.Economic Journal, 1985, 95, 909927.

.Economics and Reality. London: Routledge, 1997.

Leijonhufvud, A. Towards a Not-Too-Rational Macroeconomics. SouthernEconomic Journal, July 1993, 60 (1), 113.

02 dunn.p65 04/02/2001, 10:53 AM586


21/21


Loasby, B. Management Economics and the Theory of the Firm.Journal of

Industrial Economics, 1967, 15, 165176. [Reprinted in P.E. Earl (ed.),BehavioralEconomics: Volume 1. Aldershot, UK: Edward Elgar, 1988.]

. Choice, Complexity and Ignorance: An Enquiry into Economic Theory andPractise of Decision Making. Cambridge: Cambridge University Press, 1976.

. The Mind and Method of the Economist. Aldershot, UK: Edward Elgar, 1989.

Parry, W. Ergodic Theory. In J. Eatwell, M. Milgate, and P. Newman (eds.), TheNew Palgrave. London: Macmillan, 1987.

Radner, R. New Ideas in Pure Theory: Problems in the Theory of Markets UnderUncertainty.American Economic Review, 1970, 50, 454460.

Richardson, G.B.Information and Investment. Oxford: Oxford University Press,1960.

Robinson, J.Economic Philosophy. Harmondsworth, UK: Penguin, 1962.

. Time in Economic Theory. Kylos, 1980, 33, 219229.

Rosser, J.B., Jr. Complex Dynamics in New Keynesian and Post Keynesian Models.

In Roy J. Rotheim (ed.),New Keynesian Economics/Post Keynesian Alternatives.London: Routledge, 1998, pp. 288302.

. (1999) On the Complexities of Complex Economic Dynamics.Journal of Economic Perspectives, 1999, 13 (3), 169192.

. Alternative Keynesian and Post Keynesian Perspectives on Uncertainty andExpectations.Journal of Post Keynesian Economics, 2001, this issue.

Samuelson, P.A. Classical and Neo-Classical Monetary Theory. In R. Clower (ed.),Monetary Theory. Harmondsworth: Penguin, 1969, 170190.

Sargent, T.J.Bounded Rationality in Macroeconomics. Oxford: Clarendon Press, 1993.

Sawyer, M.C. (ed.). The Intellectual Legacy of Michal Kalecki. Aldershot, UK:Edward Elgar, 2000.

Sent, E.-M. Sargent versus Simon: Bounded Rationality Unbound. CambridgeJournal of Economics, 1997, 21 (3), 323338.

Shackle, G.L.S. Uncertainty in Economics: And Other Reflections. Cambridge:

Cambridge University Press, 1955..Epistemics and Economics. Cambridge: Cambridge University Press, 1972.

.Imagination and the Nature of Choice. Edinburgh: Edinburgh UniversityPress, 1979.

. Imagination, Unknowledge and Choice. Greek Economic Review, 2, 1980,95100.

Simon, H.A.Models of Man: Social and Rational. New York: Wiley, 1957.

. Theories of Decision-Making in Economics.American Economic Review,1959, 49, 253283. [Reprinted in P.E. Earl (ed.),Behavioral Economics: Volume 1.Aldershot, UK: Edward Elgar, 1988.]

. Administrative Behavior,2d ed. New York: The Free Press, 1961.

. Theories of Bounded Rationality. In C.B. McGuire and R. Radner (eds.),Decision and Organisation. London: North-Holland, 1972.

. From Substantive to Procedural Rationality. In S. Latis (ed.),Method andAppraisal in Economics. Cambridge: Cambridge University Press, 1976.

. Behavioral Economics. In J. Eatwell, M. Milgate, and P. Newman (eds.),The New Palgrave. London: Macmillan, 1987a.

. Bounded Rationality. In J. Eatwell, M. Milgate, and P. Newman, (eds.),The New Palgrave. London: Macmillan, 1987b.

Skidelsky, R.John Maynard Keynes, Volume 3: Fighting for Britain, 19371946.London: Macmillan, 2000.

02 dunn.p65 04/02/2001, 10:53 AM587

Documents

- Bounded Dunn 01 r i