The Economic Agent: Not Human, But Important
Don Ross
Department of Philosophy and Department of Finance, Economics and Quantitative Methods, University of Alabama at Birmingham
School of Economics, University of Cape Town
1 Introduction
Critics of mainstream economics typically rest important weight on the differences between people and the 'agents' that populate economic theory and economic models. Hollis and Nell (1975) is both representative of and ancestral to many more recent variations on the theme. Lately, the upgraded status of behavioral economics (BE) within the discipline's mainstream has encouraged a number of writers to use revolutionary rhetoric in promotion of a 'paradigm shift' that includes the rejection of 'rational economic man' (Ormerod 1994, Heilbroner and Milburg 1995, Fullbrook 2003). The current leading developers of BE are generally more circumspect, claiming that their approach complements standard theory rather than promising to supplant it (Camerer and Loewenstein 2004, Angner and Loewenstein this volume). However, they generally join the more florid critics in supposing that microeconomics is bound to improve its empirical relevance to the extent that it substitutes the study of people for that of abstract economic agents. Another body of thought that promotes this view stems from Sen's (1977) attack on standard economic agents as 'rational fools', amplified in Davis's (2003) argument that since economic agents lack some essential properties of human individuals, economic theory requires fundamental reform if it is to make progress in explaining human behavior.
That economic agents and people have different properties
should strike no one as surprising. Whereas people are pre-theoretical entities
found in the world, economic agency is a theoretical construction elaborated as
part of the development of a family of models. In philosophical terms, we might
therefore describe the view that economists should forget about economic agency
and directly study people instead as an expression of normative
phenomenalism. This would be the thesis
that the proper objects of scientific attention are manifest phenomena, which
should be described directly rather than by way of intermediate theoretical
kinds. This is not a view we find on display elsewhere in the philosophy of
science. Though strong empiricists, such as Bas van Fraassen (1980, 2002), deny
that we are entitled to ascribe model-independent reality to the unobservable objects
of reference used in scientific theories, I have never heard anyone insist that
physicists ought to stop modeling fields and manifolds and go back to
generalizing directly about rocks and tables. Elsewhere in this volume, I
extensively discuss the reasons why economics has attracted a level of
anti-theoretical hostility not
encountered by other sciences (aside from evolutionary biology). I suggest that
this discussion is useful background for explaining the eagerness with which
revolutions in economics are promoted on grounds we don't encounter elsewhere
in science. In the present essay, I will assume that normative phenomenalism,
especially as applied arbitrarily to and only to economics, is not rationally
motivated. This assumption does not foreclose the possibility that either
current critique or the future course of economic science could reveal the idea
of economic agency, either in general or in some common particular form, to be
unhelpful. Use of any given agency concept in science is subject to requests for justification; but the mere
fact that economic agency is abstractly constructed establishes no prima facie case against such justification.
It might be objected that normative phenomenalism is a fair
standard for application to economics in particular because economic theory,
unlike physical theory, generalizes only or mainly over observable types. In
this connection, MŠki (1986, 1992) points out that Friedman's (1953) famous
methodology isn't in fact the standard sort of instrumentalism it's typically
taken to be because, unlike philosophical instrumentalists about the
unobservable entities of physics, Friedman assumes the objects of economics to
be manifest: consumers, firms, prices, etc.. He then doubts that economic
theory truly describes these objects, useful though it is for predicting their
trajectories. He does not doubt, as the instrumentalist does of bosons, that
the basic objects of economics exist. Though I agree with MŠki about what
Friedman thought on this question, I do not think that Friedman's opinion here
is correct. Because the words used by
economists, unlike 'boson', are derived from everyday vocabulary, it is easy to
forget that in their theoretical context they denote abstractions. Despite
slightly quaint philosophical jargon, Stigum (1990) offers nice examples of the
point: ÒWe have knowledge by acquaintance of the salary we received last year,
but we have knowledge by description only of what our income was, i.e., of the
maximum amount of money we could have spent last year and been as wealthy at
the end of the year as we had been at the beginning of the year É We have
knowledge by acquaintance of the price of our house, but only knowledge by
description of its current market valueÓ (p. 550). So it is with agency in
economic theory: we gaze upon and shake hands with people, but not with
economic agents. But, in the absence of an argument for normative
phenomenalism, this fact by itself no more implies that economists should stop
theorizing about agents, or equate them with people, then similar logic would
rightly advise them to stop theorizing about incomes or to equate incomes with
salaries. Again, this is not to deny the validity of requests for justification
on grounds internal to the goals of economics (as opposed to external
philosophical grounds).
The structure of the chapter is as follows. I will first
sketch the standard concept of the economic agent as featured in contemporary
microeconomics. I will then show why the practice of economists does not equate
this agent to a person, and why economists' longstanding interests in
'individualism' and 'microfoundations' should not be interpreted as suggesting
otherwise. This will show how, in detail, economists should respond to
criticisms reflecting normative phenomenalism. In section 5 I will indicate why
and how (some) behavioral economists propose to modify agency in light of
studies of people, in cases where normative phenomenalism is not assumed. The core of this argument involves
contesting the view held by increasingly many behavioral economists that their
program collapses into the ambition of the new 'neuroeconomics' to identify and
explain the processes by which brains comparatively value actual and
prospective rewards. I will maintain that what I will call 'modular neuroeconomics'
(as found in work by Glimcher 2003 and Caplin and Dean forthcoming) is
importantly different in its implicit attitude to standard economic agency from
a more reductionist version of neuroeconomics that has lately been stapled to
BE in would-be service of a paradigm shift (Camerer, Loewenstein and Prelec
2005). Having explained why modular neuroeconomics preserves rather than
challenges the standard concept of economic agency, I will defend the continued
use of that concept against calls for its replacement by objects and processes
identified through psychological and neuroscientific observation.
2
Economic agency
There is a clear historical path by which the standard
concept of the economic agent was developed (Mandler 1999). This agent first appeared
in the work of the early neoclassical theorists (Jevons and Walras) as a
maximizer of ceiling-less hedonic utility laboring under a finite budget,
subject to diminishing marginal returns from consumption within classes of
commodities he deemed to be close substitutes. I deliberately use the pronoun
'he', because at the point of his historical arrival the economic agent was
both normatively male in his status as a social atom and (more importantly for
present purposes) human, in that he relied on 'creature sensations' to both
form his close-substitute classes and to rank them with respect to the utility
they delivered. His agency revolved around his efforts, given his limited
means, to create the most appealing inner environment he could, as determined
by his own introspective judgment.
Although the early neoclassical agent was human, he was
already not a whole person. In sympathy
with Mill's refusal to follow Bentham in regarding all sources of satisfaction
– pushpin and poetry, a foot message and an end to poverty – as
lying on a single commensurable scale, Jevons (1871) took the economic agent to
be the aspect of the person
concerned with the consumption of 'lower' wants. We can fully understand what
was 'lower' about these wants only by going slightly outside the frame of
Jevons's text and importing some knowledge of the Victorian world-view. To some
extent the lowliness of economic wants lay in their materiality. But Victorian
idealism was closely bound up with the morality of social obligation: material
goods were 'low' in part because, unlike 'spiritual' goods, their consumption
as sources of utility was private; 'higher' wants were higher in part because
attending to them expressed commitment to public civilization. Given the
importance of atomism as a property of the economic agent for which he is
widely rebuked by his critics (including contemporary ones), this point merits
emphasis. The Victorians were pointedly and self-consciously divided amongst
themselves as regards metaphysical atomism versus holism, with the
scientifically minded such as Jevons inclining to the former and most
philosophers defending the latter. But neither Jevons nor Walras were moral
atomists; both rejected the idea that a
person should give his highest priority to what they regarded as his economic
interests.
Some readers might have jumped to the conclusion that I am
calling the early neoclassical agent 'human' because he had 'feelings'. It has
long been fashionable to contrast the 'cold calculator' featured in economic
theory with the warm, sentimental and impulsive beings celebrated by all
romantics and by most Western humanists. Though this is important for
understanding sources of non-rational antipathy to economics, it seems to me
ethnocentric to view emotional parochialism and impulsiveness as the core
properties of the human; Western romantic humanism is a peculiar, not a
globally typical, idealization of human nature. Thus I would question the
long-run philosophical importance of contrasting the early agent's passions
with the later agent's lack of them. Instead, I suggest, what made Jevons's
economic agent human by contrast with his contemporary successor was the
former's grounding in consumption within the boundaries of his body. The early neoclassical agent was an aspect of the
human animal. Thus there was an
implicit one-to-one mapping between these agents and human organisms, which all
applications took for granted.
As recognized by many writers, and reviewed succinctly by
Bruni (2005) and Bruni and Sugden (2007), Jevons's introspective agent was on
the way out before the twentieth century began; Pareto, in particular, worked
to reduce his defining properties to a mere disposition to consume in
accordance with representation by indifference curves. Following on this lead
using more powerful mathematical resources, the introspective agent was killed
stone dead in the ordinalist revolution of the 1930s and 40s led by Hicks,
Allen and Samuelson (Mandler 1999). As related by Ross (2005), however, what
never disappeared from most economists' (or other people's) informal conception
of the economic agent was the idea that he was still (as it were)
'ontologically grounded' in the human organism. By this I mean only that the
one-to-one mapping between agents and organisms presumed by Jevons and Walras
(henceforth, 'A -> O') remained the basic reference point for
understanding the place of agents in the empirical interpretation of economic
theory, even as the agent's human properties were steadily stripped away. There
were motivations for this conservatism, as we will see; it wasn't merely a case
of conceptual inertia. But, I will argue, we can make more consistent sense of
the character of most economics since Samuelson by dropping the attribution to
its foundations of the assumption of A -> O.
The ordinalist revolution did not so much modify the concept
of the economic agent as, to begin with, attempt to eliminate him. In the
canonical ordinalist texts, Samuelson (1938, 1947) set out to derive the
existence of sets of preferences mappable onto the real numbers by monotonic,
complete, acyclical, and convex functions from observable schedules of
aggregate demand. He would have preferred not to call these 'utility'
functions, but the lure of semantic continuity turned out to be a more powerful
force than his preference, and he quickly surrendered the point to convention.[1]
As the label 'revealed preference theory' was intended to suggest, his utility
functions were intended as descriptions of actual and hypothetical behavior,
not inner evaluations of experienced relative states of satisfaction. It is
common to attribute the motivation for this to the behaviorism and positivism
that dominated the psychology and social science of the 1930s, 40s and 50s, and
certainly this influence played its part. However, imagining it to have been
the main, let alone the sole, motivation ignores the fact that Samuelson
completed a process that had been underway for decades in economics, and which
thus reflected a special dynamic internal to the discipline. This was the felt
pressure to make economics a social science
independent of any foundations in individual psychology. Cold war neuroses
demanding adherence to 'methodological individualism' did much to obscure the
point in retrospect. But as good Keynesians, Hicks and Samuelson were, in a
very important sense to which we will return later, uninterested in individual agents, a concept of which they merely
inherited from an earlier neoclassical theory they profoundly transformed. If
we let Samuelson's (1947) mathematics speak for itself, as he largely though
inconsistently does himself in Foundations, then among the short and general things we might say about the role
of the agent in revealed preference theory the most accurate is that there
isn't one. There is observable aggregate demand, and if this has certain
testable properties then the existence of continuous preference fields is
implied. What stabilizes these fields might or might not be properties of
individual psychologies; the revealed preference theorist disavows professional
interest in this question, a point on which Samuelson is explicit.
All this makes it easy to imagine that, and how, the agent
might have disappeared altogether from economic theory had the discipline
technically matured in a slightly different context. Indeed, someone might well
argue that the agent did substantially
disappear despite the fact that the word 'agent' soon made a comeback in the literature following on Samuelson.
There are three possible interpretations to be distinguished here. By
'interpretations' I refer not to claims about what historical economists
actually intended, but to attributions that might be offered by philosophical
reconstructions that apply retrospective principles of charity in full knowledge
of contemporary economics. The possible interpretations are:
(1)
The role of the agent was eliminated from microeconomic theory
after World War Two.
(2)
Postwar microeconomic theory retained a concept of the agent,
but with substantial modifications that imply abandonment of the commitment to
A ->
O (whether or not many economists, who are not in the philosophy business,
noticed this).
(3)
The absence of agents in Samuelson's version of revealed
preference theory was an idiosyncratic wobble in the evolution of microeconomic
theory; the reappearance of the word 'agent' in subsequent canonical texts
indicates stronger continuity with early neoclassicism than Samuelson
suggested, in particular, continued ontological orientation around A -> O.
Contemporary paradigm-shifters based in BE, along with Sen
and his followers, adhere to interpretation (3) and then, in rejecting social
atomism, take themselves to be calling for the overthrow of a historically
unified neoclassical tradition. (Thus they often refer to the contemporary
mainstream as 'Walrasian'.) I will defend interpretation (2).
Let us now hoist the target of the conflicting interpretations onto the table. Again, there can be no dispute that Samuelson's avoidance of the word 'agent' failed to stick as a practice: the subtitle of Rubinstein's (2006) elegant formulation of the core elements of microeconomic theory, which deserves to be regarded as authoritative on matters of current convention, is ÒThe Economic AgentÓ. I will summarize the part of Rubinstein's formulation that might plausibly be taken to be definitive of economic agency. This is the part that can be stated independently of any assumptions about representations or computations taken to be aspects of agents' psychologies; were such assumptions to be incorporated into the definition of agency then the question distinguishing the defenders of interpretations (2) and (3) above would necessarily be begged in favor of the latter. Note that the judgment about what to regard as 'definitive' that I will express below is mine, not Rubinstein's. Note also that Rubinstein's formulation reflects the consolidation of postwar consumer theory provided by Debreu (1959), rather than the less exact version found in Samuelson (1947); this is a point that will be important in the later discussion.
The agent is a reference point for ascription of a utility
function. Utility functions are constructed from preference functions or
represent preference relations. A preference function or relation generalizes a
series of answers to a series of evaluative questions about elements x, y,
É, n of a set X, with one answer per question of the form 'x
is preferred to y' (x
y), 'y is
preferred to x' (y
x), or 'x and
y are interchangeable in preference
ranking' (I). Rubinstein shows
that two forms of generalization are equivalent:
(1)
Preferences on a set X are
a function f that assigns to any
pair (x, y) of
distinct elements in X exactly
one of x
y, y
x, or I,
restricted by two properties: (i) no order effect: f (x, y) = f
(y, x); and (ii) transitivity: if f (x, y) = x
y and f (y, z) = y
z then f (x, z) = x
z and if f (x, y) = I
and f (y, z) = I then
f (x, z) = I.
(2)
A preference on a set X is
a binary relation
on X
satisfying (i) completeness: for any x, y
X, x
y or y
x; and (ii) transitivity: for any x, y, z
X if x
y and y
z then x
z.
A utility function is a representation of a preference
relation according to:
represents
if for all x, y
X, x
y if and only if U(x)
U(y).
If the foregoing is taken to restrict the conception of an
agent then it follows that an agent's preferences are not lexicographic (Debreu
1960). This also follows from conceiving of preference relations as continuous.
From Debreu (1954, 1960), any set of continuous preferences is represented by a
continuous utility function.
The agent distributes her investments in alternative
feasible states of the world in accordance with the weak axiom of revealed
preference. I use a formulation of my own here instead of Rubinstein's: for two
complete states of the world x, y: x
y, if the agent pays opportunity cost c + y in
exchange for x, then the agent
will never pay opportunity cost c +
x in exchange for y. This implies that the agent's behavior will be
consistent with the hypothesis that she maximizes a utility function according
to which U(x)
U(y).
When agents are located in markets where they encounter
consumption problems, more is generally assumed of them. In particular, it is
supposed that when they are faced with alternative investments in
quantitatively measurable combinations of elements (bundles) from their utility
functions, their preferences satisfy monotonicity (for any element x
X, x +
x), continuity, and convexity (consumption behavior is
consistent with representation by neoclassical indifference curves). Stronger
assumptions, particularly that utility functions are differentiable, are
typically added if we are concerned to show that a particular model of a
consumer's optimization of consumption given a budget is explained by reference
to her preferences. Note that economists are almost never moved by this concern
except when they are engaged in explicit justification of abstract theory
– that is to say, when they're not actually doing economics.
In light of the foregoing, our prior question about the
ontological presumptions around agency in postwar economic theory comes down to
this: what import should be attached to saying that a reference point for
ascription of a utility function, as just defined, is an 'agent'? 'Reference
point' here just means an element of some index constructed for a particular
analytic exercise; so all the weight lies on concept of the utility function.
It should be evident that what I identified earlier as the 'human' properties
of Jevons's agent make no appearance in the definition. Nor, at least until the
rise of BE, did they play any explicit role in interpretations of the formalism
in applications. Now, there is no room for serious doubt that in the Western
intellectual tradition the prototypical agent is the goal-pursuing aspect of a
single person over the course of her biography from the dawn to the demise of
her mature competence in practical reasoning (Ross 2002). The idea has a
relatively clear and constant conceptual core from the work of Aristotle
through Kant. From this perspective it should seem puzzling that Samuelson's
avoidance of reference to agents didn't continue to be respected: reference
points for ascriptions of utility functions don't seem particularly to resemble
philosophers' agents. Why then is it standard practice we find Rubinstein
reflecting in using 'the economic agent' as his subtitle in 2006?
In aiming to be empirical scientists, rather than members of
the community of mathematicians who study constrained optimization,[2]
economists necessarily suppose that their theory gives a general description of
some class(es) of empirical phenomena. At the most crude level of description,
there seem to be two alternatives here: the theory can be about people, or it
can be about emergent systems of production, consumption and exchange, in a
context of agnosticism about who or what the ultimate units of these activities are (if there need to be ultimate such units at all; see Ladyman and Ross 2007
for reasons to doubt this). Once the issue is put this way, it might be
supposed that the answer to the question at the end of the previous paragraph
is obvious: utility functions must be proxies for individual flesh-and-blood
consumers lest we implicitly endorse mysterious 'group minds' that don't
decompose into individual minds; methodological individualism follows from
metaphysical atomism. If utility functions map one-to-one onto people for
philosophical reasons, then in light of the same philosophical tradition
according to which A -> O, a theory of the utility function is a theory of the
agent.
However, economists are usually reluctant to accept
important professional doctrines simply on philosophical grounds, as they
should be. One consequence of the public prominence of the Chicago School has
been to greatly exaggerate the perceived commitment to methodological
individualism in workaday economics. Agnosticism about microfoundations need
not imply – as it certainly didn't for Keynes or Samuelson –
endorsement of a transcendent Hegelian spirit which, in addition to thinking
about itself and moving history along, also produces, consumes and trades. The
respectable scientists who work today in complex systems theory (who are
respectable as scientists regardless of whether one shares their confidence in
their approach) believe in emergent processes and entities, behavior of which
cannot be derived from behavior of their constituents in vitro, but generally do not believe that
feedback-regulated dynamical systems are manifestations of Spirit. Of course,
complex systems theory did not yet exist in the 1950s. But this didn't deter
Samuelson from haughty indifference about the atomic material contents of the
economist's structural black boxes. (For example, at one point in the Foundations
[p. 87] he effectively implies that the
firm in production theory is not a 'company' in the everyday sense, since the
latter but not the former may make profits; but, he says, studying
institutional contexts that allow companies to gather rents is not the
economist's business. This would imply that it is also not in the economist's
brief to say why people form companies in the first place.) The real liberator
of economists from the ball-and-chain of microfoundations was Keynes, who
enjoyed emphasizing that the concerns of the philosophers in whose company he
had been intellectually trained were of no practical import in the dangerous
concrete world where policy was called upon to keep revolution at bay. Keynes
made economics both theoretically autonomous and professionally thrilling, and
these two attractive aspects of the profession as it set about reorganizing the
postwar order were closely related to one another. The conquering
macroeconomists of the Bretton Woods era were neither metaphysical atomists nor
metaphysical holists; they were practical structuralists who left metaphysics
to others.
I have already alluded, in my reference to 'Cold War
neuroses', to one reason this golden moment didn't last. Opposing Stalinism
obviously didn't rationally require that
anyone swear fealty to methodological individualism; but war is no friend to
subtlety (nor, as emphasized by Mirowski [2002], were the military funding
sources that fueled the expansion of postwar science, including economics[3]).
It cannot be rigorously demonstrated, but nevertheless seems very likely, that
extra-theoretical political factors in the postwar democracies constituted the
most decisive influence on economists' return to the rhetoric of social atomism. Because such rhetoric was also
widely associated – by the loosest, Humean, kind of relation – with
defense of markets against 'collectivists', and because economists are indeed
appreciators of markets, Chicago School celebrities readily promoted the idea
that economic theory has both descriptive and normative individualism built
into its core.
Though I contend that this was indeed more a matter of
rhetoric than logic, it would be seriously mistaken to suppose that the only
reason economic theory didn't continue down Samuelson's agent-free path is the
purely external, sociological one that its popular image was captured by cold
warriors. In the first place, as I argue elsewhere in this volume, the
completeness of the capture is often exaggerated. In the second place,
economists were not unaware that most of their applied work continued to focus
on aggregate magnitudes and relations. Economists had reasons, grounded in microeconomics rather than metaphysics,
for thinking that agency couldn't be
excised from their theoretical foundations. I will concentrate on two.
First, the invention of game theory (GT) by von Neumann and
Morgenstern in 1944 allowed economists to model the interactions of
idiosyncratically varying utility functions rendered interdependent by
contingent distributions of scarcity. Nothing in the mathematics stipulates
that these must be interpreted as the utility functions of people; indeed, in the most useful contemporary economic (as opposed to psychological) applications of GT,
they represent objectives of firms rather than of humans (Ghemawat 1998; Klemperer
2004; Milgrom 2004). However, GT required the enrichment of utility theory that
von Neumann and Morgenstern (and then Savage) provided in order to incorporate
players' uncertainty about the valuations of and information available to other
players. This enrichment was elucidated at every step by heuristics drawn from
folk psychology, and thus the non-mathematical version of the vocabulary of
game theory is full of psychological notions: beliefs, conjectures, aversion,
attraction. Furthermore, and more substantively, GT made it possible for
economists to use the core elements of their conceptual toolkit (constrained
optimization and opportunity cost) to systematically study individual choices
in strategic contexts and so, like good opportunistic scientists, they duly
embarked on such study. If we are to base our views of disciplinary boundaries
on what scientists actually do instead of on philosophical doctrines about how
the world is objectively carved,
then we must agree that the early game theorists thereby widened the scope of
economics, regardless of whether a revealed-preference purist would approve.[4]
Finally, GT seemed to demand progressive deepening of links between economics
and psychology as it technically evolved over the past 35 years. It can
be given a strictly behaviorist
interpretation, according to which one uses it to guide inferences about
players' stable behavioral orientations through observing which vectors of
possible behavioral sequences in strategic interanimation are Nash equilibria.
But the power of such inferences is often limited because most games have
multiple Nash equilibria. Efforts to derive stronger predictions led a majority
of economic game theorists in the 1980s to interpret games as descriptions of
players' beliefs instead of their
actions. On this interpretation,
a solution to a game is one in which all players' conjectures about one
another's preferences and (conditional) expectations are mutually consistent.
Such solutions are, in general, stronger than Nash equilibria, and hence more
restrictive. As pointed out in criticism by Binmore (1990), the resulting
'refinement program' draws game theorists not just into psychology but deep
into philosophy, since it
requires them to study their own 'intuitions' about which chains of argument
must be pursued if an agent is to count as 'rational'. In this context the idea
of agency looks fundamental to
microeconomics.
Second, the formal completion of general equilibrium theory by Arrow and Debreu (1954) required the concept of an 'economy' to be strictly regimented, and this in turn demanded imposition of strong general constraints on 'participants' in such economies (Debreu 1959). In particular, it was necessary to assume that the participants could rank all possible states of the world with respect to value, and that they never change their minds about these rankings. Again, nothing required that 'participants' be interpreted as coextensive with people. As argued at length in Ross (2005), if agents in general equilibrium are identified with utility functions, then the fact that changes in utility functions imply changes in agent identity is an excellent reason not to identify such agents with people. However, an important part of the intended point of general equilibrium theory, all the way back to Walras, has been to serve as a framework for thinking about the consequences of changes in exogenous variables, especially policy variables, for welfare. Regardless of whether descriptive individualism is persuasive as social metaphysics – the reader will have gathered that I think it is not – there remain the best of reasons for endorsing normative individualism: improvements and declines in the feelings of particular people about their well being is what most people, as a matter of fact, mainly care about, so for an economist to regard anything else as the appropriate topic of welfare analysis is to implicitly impose the economist's parochial value scheme on society. Policy makers should ignore the advice of such economists.[5] Thus if the loci of preference fields in general equilibrium theory are not at least idealizations of people, then it is not evident why efficiency, the touchstone of general equilibrium analysis, should be important enough to warrant touchstone status.
Theoretical developments in the 1970s added economic
substance to this philosophical concern. The 'excess demand' literature of that
period, centering around the Sonnenschein-Mantel-Debreu theorem (Sonnenschein
1972, 1973; Mantel 1974, 1976; Debreu 1974), showed that although all general
equilibria are efficient, there is no unique one-to-one mapping between a given
general equilibrium and a vector of individual demand functions. (Put more
directly, for a given set of demand functions there is more than one vector of
prices at which all demand is satisfied.) In tandem with the Lipsey-Lancaster
(1956) theory of the second-best, Sonnenschein-Mantel-Debreu challenged the
cogency of attempts by welfare economists to justify policy by reference to merely
inferred (as opposed to separately and empirically observed) subjective
preferences of consumers. Note that this problem arises whether one assumes an
atomistic or an intersubjective (and aggregate-scale, sociological rather than
psychological) theory of the basis of value. Nevertheless, the excess demand
results shook the general postwar confidence that if one attended properly to
the aggregate scale then specific properties of individuals could be safely
ignored.
Both the theory of individual choice under uncertainty and
welfare theory are extensions of core
microeconomic theory. Therefore, the fact that both embroil economists in
issues about agency is not a slam-dunk argument for interpreting that core
using the standard semantic label chosen by Rubinstein. However, here it is
important to remember that if the pressure to regard economics as being about
agents isn't decisive, the basis for resistance to such an interpretation isn't
very powerful either. As observed above, in denying that macroeconomics had
necessarily to be derived from microeconomics, Keynesians expressed commitment
to pragmatism, not philosophical holism: they left microeconomics behind
(Keynes) or blithely cast aside its early neoclassical commitments (Hicks and
Samuelson) because they thought that rigid fealty to Jevons and Walras stood in
the way of exercising available capacities to control policy-relevant economic
relationships and magnitudes. Therefore, if we come around to the view that
psychologistic GT is relevant to policy, as all behavioral economists believe,
then the same attitude that led Samuelson to drop agents from his foundations
should inspire us to put it back. Furthermore, if psychologistic GT is relevant
to policy because of variations in individuals' utility functions and attitudes
to risk, then it seems our idea of welfare is implied to be richer than merely
the vague utilitarian commitment to maximize community indifference curves that
characterizes most economics applied at the scales of national and international
policy.
I think that these considerations do defeat interpretation
(1) of the place of agency in postwar economic theory. Economics is motivated
by a broader set of empirical observations than merely noticing that ecologies of self-maintaining entities
collectively demand more consumption goods than the world can provide; it is
equally fundamental to the discipline as we now find it that these entities
have available to them and use importantly different strategy sets and
strategies for coping with specific aspects of their scarcity problems. Once we
have got as far as talking about 'entities with varying utility functions and
strategy sets' then it would simply be conceptually obtuse to deny that our
focus is on agents. Indeed, we should arrive at this conclusion with some
relief. It spares us the need to try to make general sense of preference or
consumption while not being able to say that there is any kind of thing that
is, in general, a possible locus for having preferences and consuming. Let me
be careful in framing the significance of this point. I don't wish to make
philosophy seem too important here, and I don't believe that we can aspire to
close the whole conceptual system by reducing basic economic concepts to some
extra-economic bedrock. Instead, preference, consumption and agency,
operationalized together as a triad, plausibly constitute a collective
conceptual primitive for economics, and as long as this doesn't leave economics
stranded apart from other sciences this should be regarded as foundations
enough. My point here is just that leaving agency in the picture doesn't
seriously compromise foundational elegance given that preference and consumption are already admitted.
Therefore, declining to identify
utility functions with agents would give more weight to philosophy – refusing to 'say what
comes naturally', just out of philosophical scruples – than doing so.
However, giving up the radical ambition to eliminate agency
from economic theory need not carry us, with Sen and the behavioral economists,
all the way to interpretation (3). I will argue over the course of the
remaining sections of the chapter that although economics is about agents, it
is not best regarded as staked to A -> O.
Before I launch into this, let me deflect a potential charge
that I have announced battle with a straw opponent. It might be objected that
the paradigm shifters have no need to accept a generalization as strong as A -> O,
and, indeed, do not insist on it. They
will agree that many applications of economics treat firms, households, unions
and even countries as agents. Furthermore, they will note – indeed, will
emphasize – that models inspired by neuroeconomics focus on sub-personal
agents (Montague, King-Casas and Cohen 2006, p. 438). This idea of representing
people as communities of agents – synchronic, diachronic or both at once
– goes back to the very dawn of BE (Strotz 1956), and so has some claim
to being regarded as among its basic points of departure from neoclassicism.
These points are duly acknowledged. I do not claim that any
economists of note maintain A -> O as an analytic or metaphysical
necessity. They are thus open to extending the concept of agency to apply it to
entities other than whole individual people, and they do regularly so extend
it. However, my key point is precisely that behavioral economists must regard these as extensions. They join classical economists and early
neoclassicals in regarding whole individual people as the paradigm or reference
cases of agents. This is an essential assumption underlying any campaign to
bring aspects of human psychology into the foundations of economic theory
– as opposed to simply conjoining aspects of economics and psychology
when specifically studying individual human choice. Now, if some who have employed
paradigm shifting rhetoric want at this point to say that the latter idea is
all they ever had in mind to promote, then disagreement dissolves. As noted
above, I do not aim to tighten membership in the club of economists so as to
exile the students of individual choice to another province where they must
call themselves psychologists; such rigidity about disciplinary boundaries is
silly. However, I claim that we dissolve the alleged basis for suggesting that
economics is in theoretical crisis or would benefit from a paradigm shift if we
give up the idea that the paradigmatic economic agent is a whole adult person.
I will argue that the postwar practice of, and the direction of theoretical and
practical progress in, economics is such that economists should be seen as
venturing away from base camp whenever they turn their attention to
non-aggregate phenomena. The contemporary concept of the agent is primarily a
theoretical construction that facilitates modeling of aggregate phenomena; and
it does a better job of this then would an agent fleshed out according to the
profile of the human being furnished by psychologists.
3.
Animal agents
As explained in the previous section, the agent in postwar
economic theory is an abstraction. There are no manifest folk entities onto
which agents need numerically map. In neuroeconomics, neurons and groups of
neurons may be agents. In development economics, agents are statistically
relevant households. In much macroeconomics since the 1970s, entire populations
of countries are modeled as if they reflected a single 'representative' agent.
By contrast, as also described above, the agent of BE is not abstract: she (no
longer gendered, as in Jevons's time) is a manifest, living, breathing animal.
More specifically, she is a social animal
with a complex, multi-part control system that is too decentralized to produce
the relentless consistency of the agent as previously defined.
Behavioral economists and their supporters among
psychologists, philosophers and others have lately been remarkably successful
in convincing other economists that in modeling agents they been neglecting
important empirical considerations, and should feel chastened by discoveries
coming from cognitive science generally and cognitive neuroscience particularly
(Camerer, Loewenstein & Prelec 2005). To cite one example, as Rubinstein
(2007, p. x) says Ò[t]en years ago it was difficult to publish a paper in the QJE
which included a 'present-bias' assumption.
These days it is almost impossible to publish a paper in the same journal which
ignores present-bias, let alone one which criticizes the approach.Ó
The discoveries that are supposed to chasten mainstream economists can be broadly sorted into four sets: (1) findings that people don't reason about uncertainty in accordance with sound statistical and other inductive principles; (2) findings that people behave inconsistently from one choice problem to another as a result of various kinds of framing influences; (3) findings that people systematically reverse preferences over time because they discount the future hyperbolically instead of exponentially; and (4) findings that people don't act so as to optimize their personal expected utility, but are heavily influenced by their beliefs about the prospective utility of other people, and by relations between other peoples' utility and their own. All of these are taken to threaten the supposed 'dogma' of mainstream (typically called 'neoclassical' or 'Walrasian') economics that people are rational and self-interested. The findings in sets (1) – (3) directly undermine (attributed) assumptions about peoples' practical consistency. Set (4) is often emphasized as undermining assumptions about narrow self-interest. This is an assumption which, it is quite easy to show, few economists make outside of institutionally constrained settings that specifically justify it (Cox 2004; Weibull 2004). However, to the extent that people's preferences drift with those they pick up from reference groups, this will further undermine intertemporal consistency. Of course, none of these putative discoveries undermine the standard model of economic agency unless it is supposed that the paradigmatic economic agent is a natural (including socially constructed) person.
Rebel flags would not be flying from the battlements of top journals if many economists did not find the call for self-chastening persuasive. In aiming to resist it, I owe an account of this disposition to be humbled. The main part of the explanation, I believe, lies in the simplified history of their discipline that most economists imbibe from textbooks. Philosophers, whose discipline largely consists in its history, are apt to under-appreciate the extent to which economists, like most scientists preoccupied with achieving strikes into new terrain rather than consolidation behind the lines, typically get by with shallow narratives about the development of their paradigms. Any history of economics that gathers all 'neoclassicals', from Jevons through Samuelson to Chicago, into a single relatively homogenous doctrine is bound to be a caricature. So then working economists, highly alert to what works and doesn't work in the practice of modeling, can be readily brought to admit that the caricatured picture needs a fundamental make-over if they are to have a conceptual and methodological framework that is truly adequate to their knowledge and judgment. In addition, in my experience, no small number of economists suffer from an analogue to post-colonial guilt over their discipline's perceived arrogance as self-nominated 'queen of the social sciences'. The less nuanced BE manifestos tend to have a populist air; allowing that psychology might partly re-write basic economic theory is an obvious way to send a clear signal that economists have put imperialism behind them.
In the simplified history of thought that often frames
casual (and some not-so-casual) methodological reflections in economics, it is
acknowledged that economists have a long history of ignoring psychologists.
This, it is then frequently supposed, has stemmed from a conviction on
economists' part that, in regarding people as narrowly selfish and materially
motivated, they operated with a more realistic understanding of at least the rational
parts of behavior than psychologists. But
now, it is thought, BE empirically vindicates the psychologists, while still
allowing an indispensable role for economists because of their training in
formal modeling. In embracing the call for paradigm change inspired by BE,
then, economists can refute the charge that their minds are closed to
theoretical change motivated empirically and by non-economists, particularly
the oft and unfairly neglected psychologists.
This impressionistic history of interdisciplinary relations isn't entirely false, of course; economists do have an established tradition of distancing themselves from psychology. As alluded to in the previous section, in the late 1930s and 1940s two threads in economic theory that had been developing separately were tied together. One thread was Keynes's focus on aggregate structural features of large economies without regard to the kinds of individual agents or actions that compose them[6] – that is, the then-new macroeconomics. The other was the attempt, clearly set in play by second-generation neoclassicists (Pareto and Fisher) near the turn of the century, to squeeze the psychological assumptions about economic agents down to a minimal core – ultimately, to nothing but consistency of preference rankings plus the idea that no agent would be content to consume only one type of good, no matter how cheap it became ('non-monomania'). Note that the second assumption is a substantial psychological hypothesis, and much more plausibly true of human beings than the first. Then, with Samuelson, as we saw, the need for even this final plausible human property was eliminated; we don't need to hypothesize non-monomania if we can use properties of observed demand to yield downward-sloping marginal utility functions empirically. This has frequently been interpreted, following the lead of Robbins (1935, 1938), as at last making a clean break between economics and psychology.
Despite their shared rejection of interpersonal comparisons of utility as unscientific, there is an important difference between the attitudes of Robbins and Samuelson toward scientific psychology. Whereas Robbins rejected the behaviorism then prevailing in psychology,[7] revealed preference theorists considered it to be a virtue of RPT – albeit, as I said earlier, a secondary one – that it was consistent with the up-to-date psychology of their time. They thus took it that ideas about how people internally represent their own preferences – most importantly for previous economists, their supposedly not subjectively liking each additional increment to their stock of a good as much as they subjectively liked the previous increment – are unscientific claims not just as economics but as psychology. This point can be used to smooth the narrative that supports the self-chastening attitude. One can say: at least some important postwar economists meant to remain responsible members of a partnership with psychology, but then the profession missed the bus at the cognitive revolution in the 1960s. Fortunately, the paradigm shifters can continue, thanks to findings in experimental economics, to the undermining of aggregate welfare measures by Sonnenschein-Mantel-Debreu, and to the way in which game theory evolved, the bus eventually came around again and economists could redeem the earlier error by this time climbing aboard.
In Section 2 I referred to the fact that the rise of the refinement program in game theory plunged economists deep into modeling of belief profiles and other objects conceptualized using the language of psychological states. This encouraged interpretations of agency consistent with A -> O. But it simultaneously introduced a tension into this commitment by inflating the computational demands on agents. The players of many refined games – e.g., those that find so-called 'sequential equilibria' (Kreps and Wilson 1982) – are computational prodigies, instantly updating all their beliefs, using all valid principles of Bayesian probability, upon receipt of any information. The capacities such refinements imply for agents are not plausible capacities of finite human beings whose inboard computational hardware was built by natural selection's incremental tinkering. And, sure enough, experimental economists duly showed that when people play the games analyzed by game theorists in laboratories, they often do not appear to behave like the agents in the models and they converge on vectors of strategies that are often not Nash equilibria (let alone subgame-perfect or sequential equilibria) according to the models (Camerer 2003). Thus, it seems to paradigm shifters, the 'assumptions' about agency of standard microeconomics need correction by the empirical facts of cognitive science.
The correction in question, according to the revolutionary manifestos, turns out to be drastic. People approximate traditional economic agency behaviorally in that they often accomplish their projects at bearable costs; but they don't exhibit any of the core computational properties attributed to economic agents by general equilibrium theory, rational-expectations macroeconomics, or game theory with refinement. 'Their' behavioral rationality typically turns out to really be natural selection's rationality, evolution having supposedly built rough situational rules of thumb ('heuristics') into people that serve them well as long as their environments are not too strange by comparison with their ancestral ones (Gigerenzer et al 1999). This critique then appears to be reinforced by cognitive neuroscience, which musters evidence for biases, heuristics and framing effects operating directly in the processing systems of the brain (Camerer, Loewenstein & Prelec 2005). Thus, it is concluded, economics collapses not just into abstract computational psychology, but all the way into computational neuroscience. That the word 'collapse' is not too strong is indicated by the sorts of things some neuroeconomists claim to discover. Recently, a team reported having determined from inspection of dopamine neurons that people do not value rewards by reference to their opportunity cost (Knutson et al 2007); they infer from this that economic theory requires revision. Open-ness to chastening from extra-disciplinary sources has gone remarkably far for any economist who admits that studies of the brain might imply revision in her view of opportunity cost as the basic state variable in microeconomics.
Once economics is taken to collapse into psychology, then discoveries in sets (1) – (4) above are naturally interpreted as tearing its standard theory apart. Furthermore, the news seems to have been getting worse since the early days of BE. Findings in sets (1) – (3) can, at least in principle, be accommodated by constructing new kinds of valuation functions. For example, people / agents can be taken to maximize within frames, even if not across them. Hyperbolic discount curves can be approximated by composing exponential ones of different slopes (Laibson 1997, 1998). However, cognitive science has lately been shaking free of a hyper-rationalistic and atomistic legacy of its own. The past decade has seen enormous upgrading of the significance attached to affect in explaining both mentation and behavior in people (Damasio 1994; Panksepp 1998). Furthermore, affect itself is increasingly understood as both responding to and conditioning dynamic social interaction, an approach to modeling that seems to be borne out by the discovery of mirror neurons (Frith and Wolpert 2004). As individual people appear less and less to be autonomous bearers and computers of valuations, whose preferences explain their exchanges but are unchanged by them, and come instead to be seen as resembling adaptive nodes in social colonies where valuations continuously modulate one another in interacting cascades,