Journal of Economic Behavior and Organization 5 (1984) 35-55.

North-Holland

ON THE BEHAVIORAL AND RATIONAL FOUNDATIONS OF

ECONOMIC DYNAMICS

Herbert A. SIMON
Carnegie-Mellon University, Pittsburgh, PA 15213, USA

Received August 1983, final version received October 1983

Existing uncertainties about the correct explanations for economic growth and business cycles
cannot be settled by aggregative analysis within the neoclassical framework. Current disputes in
theory rest largely on ad hoc, casually empirical, assumptions about departures from perfect
rationality under uncertainty. Such disputes can only be settled by painstaking microeconomic
empirical study of human decision making and problem solving. Microeconomic research of the
kinds that are required can receive powerful guidance from the theories of human thinking that
have been developed and tested over the past twenty five years by cognitive psychologists.

1. Introduction

The study of the movement of economic systems through time has focused
on two phenomena of particular interest and importance: the business cycle
and long-run changes in technology and productivity. One of the impressive
features of Joseph Schumpeter’s economic theories is his unification of these
two phenomena, accounting for business cycles as the byproducts of
discontinuous advances in technology. The task I have set myself in this
paper, however, is not to argue for this, or any other, particular theory of the
business cycle or of economic development. Instead, I wish to look at the
logical foundations of such theories, and particularly the assumptions they
make about the behavior of the human actors in the economic system.
If I may be permitted a personal remark, I observe that I am one of a very
small number of pre-Keynesians present at this meeting.’ By a pre-Keynesian
I mean someone whose habits of thought about economics were already
formed, or at least partially formed, before the publication of the General
Theory in 1936. By that year I had already studied Ely’s textbook, taken
Henry Simonds’ course in price theory, and attended Henry Schiltz’s courses
in econometrics. Of course, you cannot draw any strong conclusions about
my economic views from those facts. Milton Friedman and Paul Samuelson

‘This paper is based on a talk given at at Conference on The Dynamics of Decentralized

(Market) Economics, sponsored by Industriens Utredningsinstitut in Stockholm, Sweden, August
29-31, 1983.

0167~2681/84/$3.00 0 1984, Elsevier Science Publishers B.V. (North-Holland)

36 H.A. Simon, Behavioral and rational foundations o/ rumomic dynamics

are also pre-Keynesians, trained at the very same University of Chicago that
I attended.
The significance of having a pre-Keynesian past is that it subjects one to
occasional attacks of deja vu. Rational expectations theory, though
innovative enough, does not come as a complete novelty to anyone who has
reflected on Frank Knight’s Risk, Uncertainty, and Profit. Reviewing the
current explanations for the business cycle that are offered to us by
monetarists, rational expectationists and supply siders, we may even convince
ourselves that there is a genuine fifty-year cycle - not a business cycle but a
cycle of business cycle theories.
Given the current disarray of macroeconomics, and the bewildering
multiplicity of theories purporting to derive macroeconomic phenomena from
microeconomic assumptions, it may be less useful to offer new ideas than to
ask how we can winnow out the many and conflicting ideas we already have.
In this paper, I will discuss the logical structure of some quite conventional
ideas about how an economy like ours might operate. My main concern will
be with the question of how we can subject these ideas to empirical test and
use them effectively in the formulation of social policy.
Finally, I am among the economic majority who regard the heresies of
Keynes as small (albeit important) heresies. Just as Luther accepted the Bible
while rejecting the Pope, so Keynes’ modes of reasoning in the General
Theory are only locally heretical. His general form of argumentation is the
one that is standard in economics: what might be called ‘what would 1 do if I
were a rational man’ argumentation. So my remarks will cover Keynes as
well as the classics.

2. Foundations of classical and neoclassical theory

The economic theories that we call classical and neoclassical are based
ostensibly on two tautologies and a postulate of rationality. The two
tautologies are Say’s Law (perhaps better referred to as the national income
identity when intended as a tautology) and the monetary identity (alias the
classical Quantity Theory of Money). The first of these says that, at any level
of production, the income generated just covers the costs incurred. The
second says that money changes hands at just the rate needed to cover
total transactions.’
These laws can be thought of as describing a system in neutral
equilibrium. A system governed by the national income identity can remain
at full employment, or at a zero level of production. A system governed by
the monetary identity can have stable prices or inflation, a high velocity of
money or a low one. There is nothing in either law to produce a change in

‘These brief statements are intended to identify the tautologies, not to characterize them with
any precision.
 H.A. Simon, Behavioral and rationalfoundations of economic dynamics 31

any of its variables - except, of course, in response to a change in one or

more of the others. Nor is there anything inherent in the equations that
forces the variables to remain constant.
We must be careful about thinking that, in labeling a law a tautology; we
thereby make it innocuous. In fact, the tautological character of a law can
only be determined in a context of the theory and observables in which it is
embedded. Newton’s Second Law, F =ma, is often referred to as a tautology
- and so it is in a situation where only mass and acceleration can be
observed. Then it is simply a definition of force. However, if independent
operations exist for measuring the force applied to a body, then F =ma is an
empirical law, for it asserts that the measured force must equal the product
of the independently measured mass and acceleration. (See ‘The axioms of
Newtonian mechanics’, ch. 6.1 in my Models of Discovery; and ‘The
axiomatization of physical theories’, ch. 6.5 of the same volume.)3
In much the same way, economic arguments can be, and frequently are,
derived from Say’s Law or the monetary equation by postulating something
(often, ceteris par&us) about certain of the variables in the law and then
deducing a relation among the others. Thus, we have the familiar argument
that an increase in the quantity of money raises the price level (the velocity
of money and the level of real product held constant). Say’s Law lends itself
to parallel arguments about an increased level of output ‘creating’ new
demand by generating income.
Whenever we encounter such an argument, ‘based on’ Say’s Law or the
monetary equation, we must keep clearly in mind that the argument is really
‘based on’ the ceteris paribus or other auxiliary assumptions that are being
made. We should not be seduced into accepting the argument because of its
tautological appearance without careful consideration of the empirical bases
for these auxiliary assumptions. Such assumptions commonly have the effect
of transforming tautologies into conditions of equilibrium.
The third component of classical theories, the postulate of rationality, can
take many forms. It’s job is to get markets cleared by providing an
opportunity for profit or increased utility whenever supply and demand do
not balance, and to make sure that resources are used up to the proper
margins. For a wide range of assumptions, rationality implies that, in
equilibrium, people will have no motivation to modify their behaviors, and
resources will be fully employed. The equilibrium need not, of course, be
static. Instead, variables, exogenous or endogenous, like innovation, capital
accumulation and population increases may produce a constantly changing
steady state, but always with full employment of resources.
What is notably absent from these classical foundations is any specification
of the mechanisms that restore equilibrium when it is disturbed. In fact, the

‘Simon (1977).
38 H.A. Simon, Behavioral and rationalfoundations of economic dynamics

only endogenous variable that is at all productive of dynamics in classical

theory is the volume of capital, which is determined by the equilibrium
requirements, and which in turn moves the system to new equilibria. In this
respect it is much easier to derive a theory of economic development from
the classical postulates than a theory of the business cycle. (Whether the
theory thus derived is correct is a different question.)
I need to qualify the assertion that classical theory provides no
specification of the mechanisms that restore equilibrium. The theory does
assert that if the demand for a commodity is, for instance, below the supply
at the market price, then suppliers will reduce the price and/or restrict the
supply. What the theory does not provide, without additional postulates, is
information about the relative magnitudes of the two adjustments or the
rates at which they will occur.
The incomplete specification of the adaptive mechanisms leads to a
problem and an opportunity. The problem is that it is precisely these
mechanisms that determine the possibility and structure of cyclical
movements about the equilibrium. The opportunity is to extend the postulate
of rationality to specify a rational mode and rate of adjustment. Economic
search theory and the theory of rational expectations are examples of
responses to this challenge.

3. The ‘stylized facts’ of dynamics

In his admirable little book, Growth Theory, Robert Solow reproduces the
six ‘stylized facts’ that, according to Kaldor, characterize the process of
economic growth in advanced economies:

(1) Real output per man (or per man-hour) grows at a more or less constant
rate over fairly long periods of time.
(2) The stock of real capital grows at a more or less constant rate exceeding
the rate of growth of labour input.
(3) Moreover, the rates of growth of real output and the stock of capital
goods tend to be about the same, so that the ratio of capital to output
shows no systematic trend.
(4) The rate of profit on capital has a horizontal trend.
(5) The rate of growth of output per man can vary quite a lot from one
country to another.
(6) Economies with a high share of profits in income tend to have a high
ratio of investment to output.4

There is a rather high level of agreement among economists, although not

%olow (1970, pp. 2-3).

 H.A. Simon, Behavioral and rational foundations of economic dynamics 39

complete agreement, that these ‘facts’ do reasonably describe what has gone
on in the economic growth of the advanced countries. A growth theory may
be regarded as a plausible first approximation if it can predict (or retrodict)
them; otherwise it is implausible. There is no guarantee that more than one
theory will not pass the test, in which case new facts must be adduced that
can discriminate among the competing theories, or some other criterion must
be introduced that can select the ‘correct’ one.
Robert E. Lucas (1977) has provided a similar list of ‘stylized facts’ to
describe what happens during business cycles. If his list will not command
universal assent, it does not differ in important particulars from the lists that
have been drawn up by others (e.g., by Mitchell in the essay whose title I
have alluded to). Whatever the quarrels may be among business cycle
theorists, they are not mainly quarrels about these facts.
‘Technically, movements about trend in gross national product in any
country can be well described by a stochastically disturbed difference
equation of very low order. These movements do not exhibit uniformity
of either period or amplitude, which is to say, they do not resemble the
deterministic wave motions which sometimes arise in the natural
sciences. Those regularities which are observed are in the co-movements
among different aggregative time series.
The principal among these are the following. (i) Output movements
across broadly defined sectors move together. (In Mitchell’s terminology,
they exhibit high conformity; in modern time series language, they have
high coherence.) (ii) Production of producer and consumer durables
exhibits much greater amplitude than does the production of
nondurables. (iii) Production and prices of agricultural goods and
natural resources have lower than average conformity. (iv) Business
profits show high conformity and much greater amplitude than other
series. (v) Prices generally are procyclical. (vi) Short-term interest rates
are procyclical; long-term rates slightly so. (vii) Monetary aggregates
and velocity measures are procyclical.’
A business cycle theory may be regarded as a plausible first approximation
if it can predict (or retrodict) these facts; otherwise it is implausible. Again,
we may have to introduce additional considerations if we find ourself faced
with several theories capable of explaining these facts.
A growth theory or a business cycle theory cannot be regarded as veridical,
then, merely because it reproduces the appropriate collection of stylized facts.
The empirical claims contained in these lists are so general that many
theories, many different sets of structural equations, might tit them. But
different structural equations will lead to different policy recommendations,

5Lucas (1977).
40 H.A. Simon, Behavioral and rational foundations of economicdynamics

and, whenever structure changes, to different predictions. Why not test

theories by comparing them with the aggregate empirical data instead of the
derived stylized facts? The general answer is that any theory that will
reproduce the stylized facts can very likely be fitted to the aggregate data
with an R2 value well over 0.9. [Nelson and Winter (1982, pp. 218-230)
provide an example and a discussion of the problem in the context of growth
models. Adelman and Adelman (195 1) constructed a classical illustration of
how a Keynesian macromodel can fit the stylized facts describing the
statistical structure of business cycles.] It seems also that the details of the
empirical data that go beyond the stylized facts in situations like these are
also well below the noise level, and contain little information that can be
used for model identification or prediction. As Leontief and Orcutt, among
others, have pointed out, the idea of validating economic theories with
aggregative data is a will-o’-the-wisp that has led econometrics down the
garden path of ever more sophisticated statistical methodologies. The
ponderousness of contemporary econometric techniques simply crushes the
fragile data to which they are applied.
In the physical sciences, when errors of measurement and other noise are
found to be of the same order of magnitude as the phenomena under study,
the response is not to try to squeeze more information out of the data by
statistical means; it is instead to find techniques for observing the phenomena
at a higher level of resolution. The corresponding strategy for economics is
obvious: to secure new kinds of data at the micro level, data that will
provide direct evidence about the behavior of economic agents and the ways
in which they go about making their decisions. [Support for this thesis
comes from surprising directions. See Lucas (1980, pp. 28%289.)] A major
purpose of this paper is to show how and why such empirical data are
essential for the future development of economics.

4. The formal mechanisms for dynamics

The mathematics of differential and difference equations delimits for us the

range of mechanisms that are available for representing the dynamics of
economic systems. The two most likely candidates are systems of linear
equations subject to random shocks, and non-linear systems (with or without
shocks) possessing limit cycles.
Linear systems call for exogenous shocks to drive them (e.g. sunspots, oil
shocks). The systems themselves need to be stable, so that they will tend to
return to equilibrium whenever displaced. The coefficients of all of the
adjustment mechanisms are free parameters. They can, of course, be
estimated from the empirical time series data themselves, but only with the
loss of large numbers of degrees of freedom.
Non-linear systems have the attractive feature that they may behave in
 H.A. Simon, Behavioral and rational foundations of economic dynamics 41

cyclical fashion, traversing so-called ‘limit cycles’, even in the absence of

exogenous shocks. To explain real data with them, not only their parameters
but also the forms of their functions must be estimated - or postulated.
Again, these estimates may be made from available time series data at the
cost of some degrees of freedom.
In the past forty years, an enormous number of dynamic models of these
kinds, small, medium and large, have been fitted to empirical data. As we are
all aware, these model-fitting exercises have not led to much consensus as to
what is the right theory. A chief problem is that the parameter values turn
out usually not to be very robust. As rational expectationists are fond of
pointing out, coefficients estimated from past data often fail badly as
predictors of the future. (Rational expectationisti are especially pleased with
this non-robustness of the coefficients of non-equilibrium theories because
their theory leads them to expect it. The robustness of their own coefticients
has not yet been tested extensively.)

5. The behavioral sources of long-term dynamics

We have now set the formal terms of our inquiry. We wish to know what
kinds of assumptions will allow us to build dynamic models that possess the
characteristics defined by the ‘stylized facts’ we have listed. But we will also
be closely concerned with the bases, logical or empirical, for these
assumptions. First, we will discuss models of long-term growth, and then
turn to models of the business cycle.
I have already indicated in very general terms some ways in which
dynamic movement is produced in long-term growth models and in models
of the business cycle. For long-term growth in classical theory, three
variables are important: one of these (technical innovation) produces a
continuing shift in the production function. The other two, population and
capital, produce a continuing growth in the factors of production. In most
models, the shift in the production function and population are exogenous,
while the capital formation is endogenous. A thoroughgoing pursuit of the
goals of classical theory would try to define the processes of innovation and
population growth as well as capital formation in economic terms - that is,
to turn all of them into endogenous variables.
In a formal way, it is perfectly feasible to produce a theory of technical
innovation based on the postulate of rationality. Since innovation is
presumably produced by the investment of human and capital resources, we
introduce a new production function for innovation, and equate the value of
the marginal product of innovation with its cost. From a formal standpoint,
we have simply replaced the task of estimating an exogenous variable, the
rate of innovation, with the task of estimating the parameters of a function,
the production function for innovation. The only obvious gain from the
42 H.A. Simon, Behavioral and rational foundations of economicdynamics

replacement is that we can now rest comfortably in the knowledge that

everything is proceeding rationally. Human rationality is now only bounded
by the characteristics of the external environment: the quality of the ore that
is mined by the innovation production process. Of course, if we examine the
metaphor too closely we see that ‘quality’ of the ore’ is a euphemism for
‘effectiveness of the thought processes of the human beings who are doing the
innovating’. We have not removed the constraint from the human brain, just
relocated it, so to speak, from one hemisphere to the other.
Something similar can be said about endogenizing the process of
population growth. All we need here is something like the marginal utility of
a child, and a reasonable respect for the technology of contraception. But
Gary Becker and others have already built such a theory for us. By gently
manipulating the part of the utility function that contains children, almost
any kind of population growth function can be matched. Of course it might
take considerable imagination to devise a function that would provide a
rational explanation for fluctuations in the American birth rate over the past
fifty years. Perhaps we would need, after all, some occasional alteration in
the utility function, some changes in taste, to accommodate the data.
The conclusion that I draw is that to build an interesting and useful
theory of long-term economic growth, even for developed countries, we have
to go behind the principle of rationality. First, with respect to the production
function we need an empirically based theory of innovation - a
Schumpeterian component.6 I say ‘Schumpeterian’ because the theory has to
explain not merely that most inventors and ‘risk capitalists’ are motivated by
the prospect of gain, but also has to specify just what the circumstances are
under which those prospects will seem rosy and the circumstances under
which they will seem bleak. Surely, a successful theory will have a large
historical component - the prospects of gain from investment in genetic
engineering are not independent of recollections of gain (one’s own or
others’) from investments in electronics. A phenomenon that depends
critically upon the strengths of beliefs in extremely uncertain events, as well
as upon acquisition of esoteric and rapidly changing knowledge about new
technological developments, must be influenced by psychological mechanisms
that go far beyond the principle of rationality.
Second, the rationality principle will also do little to explain rates of
population growth unless complemented by a historical and psycho-
sociological theory of the determinants of the ‘utility’ of children. If we object
that such historicizing, psychologizing and sociologizing are not the businsss
of economics, then we must conclude that the objector thinks that long-term
growth theory is not the business of economics.

“Nelson and Winter (1982).

 H.A. Simon, Behavioral and rationalfiundations of economic dynamics 43

6. The business cycle: Shocks

There is no dificulty in specifying various sources of unpredicted shocks as

a means for causing departures of an economy from equilibrium. The weather
is a perennial source of such shocks for economies that have an important
agricultural component. The Oil Shocks provided spectacular examples of
another sort, while wars, revolutions, and even elections supply a continuing
sequence of disequilibrating events.
If we were to carry the logic of neoclassical economics to its limit, we
would want to predict even these events (omitting the weather) as
endogenous products of rational cogitation. However, a more limited goal,
and the one usually accepted by economic theory, is to introduce them as
unpredictable, hence unexpected, events. Then, in rational expectations
theory, they need only be reacted to, not foretold.
Shocks, defined in this way, can make a valuable contribution to the
testing of economic theories. For if they are genuinely unpredicted, they can
be used after the fact to identify economic models and improve our estimates
of their parameters. If there has been any systematic discussion in the
economic literature of methods of exploiting this information, it has escaped
my attention.
In business cycle theory we are concerned with the ways in which
economic agents respond when they are faced with departures from
equilibrium, or with expectations of such departures. Rather than discuss the
matter in general terms, I think it more useful first to take up some
examples: the inventory cycle, responses to price changes, and ‘permanent’
(or is it ‘natural’?) unemployment.

7. Inventory cycles

A businessman may receive signals from the economic system indicating

that he is currently supplying more of his product than can be sold in the
market at the current price. The usual signals are that inventories of finished
product are increasing in stores or warehouses, or if the product is made to
order, that new orders are arriving more slowly than products are being
shipped. He may respond to these signals in a variety of ways - even
ignoring them at his peril. He may reduce prices; he may reduce the
production rate. If he reduces the production rate, he may reduce his labor
force, put them on shorter hours, or simply divert some of them to
maintenance work of some kind. In addition to price and product
adjustments, he may increase (or decrease) his advertising, reorganize his
sales efforts and sales organization, modify his product, or promise shorter
delivery times. The actions he might take are limited only by his ability to
imagine possible decision variables, and the range of values that each might
take.
44 H.A. Simon, Behavioral and rational foundations of economic dynamics

Moreover, there are many signals besides increasing inventories that might
trigger the actions. The excess supply may be predicted prior to the actual
accumulation of inventories or the depletion of orders and the prediction, in
turn, may be based on many kinds of information, ranging from astrology,
through the pages of the Wall Street Journal, to the latest turn of the
Wharton model.
A business cycle theory must, somehow, capture or sum up these reactions
- not for any single businessman, but for the aggregate of enterprises. This
is usually done by introducing aggregate equations that are copies, writ large,
of what is imagined to be the response process at the micro level. Of course
the copy is not supposed to reproduce all the detail and variety of the
individual functions, but only their gross, most important, features.
Now since management scientists and operations analysts have been
rummaging about in business firms for many years, and since production
scheduling has been one of their favorite concerns, we have a great deal of
information - although it is rather unsystematic information not easily
summarized in statistical distributions - about how inventories and
production rates are actually controlled. I would like to make some
generalizations, based on this knowledge, that I think could be verified by
systematic sampling. First, production responses are more common than
price responses, at least in the early stages of inventory buildup. This is
probably not unrelated to the fact that prices are usually constructed by
marking up costs. Second, in most business firms the response strategies are
not formalized; responses are made ‘judgmentally’ by executives without
detinite rules to guide them. Third, the decisions about rates of production
are often made by different persons, using different procedures, than the
decisions about prices. These are ‘stylized facts’ at the micro level, accepted
by most persons who have spent much time in business firms.
To avoid dropping into ‘anything-can-happen’ mode, let me describe an
actual forma1 decision rule that has been devised to govern the production-
level decision, and which has had at least some modest application to real
factory situations: the so-called ‘linear decision rule’ devised by Holt,
Modigliani, Muth and Simon (1960) (two Keynesians, a rational
expectationist and a behavioralist!). Simply stated, the rule determines the
desired level of production for the next scheduling period as a function of
current inventories and expected values of sales for future periods. The level
selected by the rule maximizes the present value of the expected profit stream
for the given sales estimates, production function and cost of holding
inventories. (A generalized form of the rule also makes decisions about the
level of the labor force.) The price of the product, and the possibility of
changing the price, are entirely ignored in the decision; the problem is to
produce at minimum cost the stream of product that will be required to meet
the predicted future sales.
 H.A. Simon, Behavioral and rational foundations of economic dynamics 45

Now I do not wish to claim that the linear decision rule solves the right
optimization problem. But in my many conversations about the rule with
practicing managers, I have never heard it criticized for its omission of price
considerations, or as being over-simple. From this, I am tempted to conclude
that the other rules used in industry are probably no more complex, and
moreover, that there is probably a considerable variety of them, using
different signaling information and different decision variables. Different
decision rules and different forecasting procedures imply different dynamic
behavior of the firm, and of the economy when the behavior of many firms is
aggregated. Moreover, the rules and forecasts that firms use are subject to
learning and fashion. That would be no great matter if the changes in them
were random and independent, but they almost certainly are not.
It would not require an unconscionable amount of research effort to
obtain, by investigations carried out in individual firms, a fairly good picture
of the strategies actually used by firms to modulate their production levels.
Such a picture would certainly provide a more satisfactory basis for an
aggregate production decision rule than a complete reliance on aggregate
time series for selecting relevant variables and estimating parameters. It
could also provide some information about the circumstances that might
cause firms to change their response strategies, and hence some basis for
anticipating structural changes in the aggregative model.
In investigating and describing the production decision rules used by
business firms, one aspect of particular interest would be the use or non-use
of forecasts of future sales, and the procedures for arriving at those forecasts.
A decision rule that simply reacts to current inventory levels is a feedback
system that is guaranteed to generate temporally damped cycles, but to
amplify them each step backward in the production chain - from ultimate
customers to producers of capital goods. Hence, if we found that such
passive, reactive feedback systems were common (as I strongly suspect may
be the case), we would have one source of explanation for Lucas’ first and
second stylized facts.
Introduction of sales forecasts into the decision rules could change the
dynamics of a system drastically. If sales estimates were based on the
extrapolation of recent history, they would not make a fundamental
difference to the dynamic behavior of production rates. However, if
expectations were formed in other ways (e.g., in whatever ways it is supposed
that ‘rational’ expectations are formed), the sales forecasts could be either
strongly stabilizing or strongly destabilizing. Possibilities for destabilization
would be especially great if many firms drew upon the same sources of
information for their forecasts.
Since these conclusions follow from well-known general properties of
feedback and feedforward control systems, I will not try to justify them in
detail here. I simply wish to point out that we cannot be satisfied that a
46 H.A. Simon, Behavioral and rational foundations ofeconomicdynamics

theory explains the observed facts of the business cycle without knowing a
good deal about the precise forecasting methods that it assumes. That brings
me to the topic of rational expectations. But instead of discussing it in the
context of inventory cycles, I would like to consider how businessmen
respond to changes in the price level, and the relation of these responses to
all types of business cycles.

8. Responses to price level changes

An attractive feature of theories that incorporate rational expectations is

that they purport to give a general explanation of how businessmen estimate
future sales and prices, and hence to foreclose the need for any empirical
inquiry into forecasting. We can ask, even of these theories, what information
they use to estimate the structural parameters that determine equilibrium,
but I wish, instead, to finesse that question and explore in another direction.
If expectations are formed rationally, in the most clear-eyed sense of that
phrase, then the corresponding economic systems tend to steer very rapidly
toward equilibrium and it becomes very difficult to understand how they can
experience consequential cycles of any kind. It is interesting to see how Lucas
deals with this difficulty. In brief, he handles it by assuming that forecasters
are rational, but not very rational. He starts with the proposition that ‘one
would like a theory which accounts for the observed movements in quantities
(employment, consumption, investment) as an optimizing response to
observed movements in prices’.’ But ‘optimizing response’ does not rule out
delusions or illusions. For the critical assumption in Lucas’ model, the
assumption that creates the cycle, is that there is ‘confusion on the part of
agents between relative and general price movements’.8 The unmoved mover
is not the assumption of rationality, but an auxiliary assumption
(unmotivated by empirical data) of a very particular failure of rationality -
a failure of the businessman to interpret correctly the price information that
is known to him.’
It appears that rational expectations theory, thus emended to account for
the business cycle, does not at all relieve us of the burden of empirical
inquiry at the microeconomic level. Instead, it instructs us loudly and clearly
that to obtain a validated theory of the cycle, we must know not only what
price signals the businessman receives and responds to, but also what
misapprehensions he suffers under while he is interpreting these signals.
Without such empirical inquiry, we are faced with the most casual of

‘Lucas (1977, p. 14).

“Lucas (1977, p. 22).
‘Of course, in neoclassical theory these failures are not treated as irrationalities, but as
rational responses to imperfect information. Whatever they are called, they will be understood
only if they are subjected to empirical study at the microlevel. They cannot be deduced as
consequences of the rationality assumption.
 H.A. Simon, Behavioral and rational foundations of economic dynamics 47

empiricisms. Removing this difficulty would seem to call for a rapid

convergence of the research program of rational expectations to the program
of behavioralism.
The problem to which I have just called attention - the critical role of
‘irrationalities’ in Lucas’ business cycle theory - is not peculiar to that
theory, or to rational expectationist theories, for that matter. Departures
from rationality are endemic in classical and neoclassical economic theories.
In earlier days, they were often described as ‘stickinesses’. In partial
equilibrium analysis, they take the form of ceteris paribus assumptions. In
contemporary dynamic simultaneous-equation models, they most often show
up as response rate parameters that determine which of the mechanisms in a
system operate rapidly and which operate slowly. Whatever their form,
conclusions about dynamic behavior and short-term equilibrium are
extremely sensitive to them.
Since this last point is as important as it is obvious, I would like to
perseverate upon it for a moment. In our research on near-decomposability,
Albert Ando and I gave a formal rationale for partial equilibrium analysis,
and for separating the short-term from the long-term dynamics of any
system. The idea is that if a system can be ‘nearly decomposed’ into discrete
subsystems by ignoring small coefficients, then the short-term dynamics of
each subsystem can be determined by examining it independently of the
others. This very strong conclusion, conversely, demonstrates the criticality of
the choice of ‘small’ coeficients for the inferences one draws about a system’s
dynamic behavior.
The formal analysis of near-decomposability teaches us that rationality
assumptions by themselves cannot provide macroeconomics with a
satisfactory microeconomics. In addition, there remains at the very least the
empirical task of determining which of the parameters is small enough to be
ignored in partial equilibrium or (what is the same thing) short-term analysis.
And a behavioralist would argue that the best way to do that is to look at
behavior in the small - at the level of the individual firm.
Rational expectations theory moves the locus of parameter estimation, but
does not eliminate the task. Moreover, as the particular example of Lucas’
business cycle theory shows, it usually requires also its own complement of
‘stickinesses’ to make it fit the facts. The particular stickiness that Lucas
proposes for his theory is a form of the money illusion. Let me try to put
that illusion (as well as others) in a little broader psychological perspective.
The rationality assumptions of classical theory require implicitly that the
economic actor attend to all of the important variables about which he has
to make decisions or that can inform him in his decisions. Whether that is
easy or hard to do - or even possible - depends on how many relevant
variables there are in the world. In any realistic description of the
environment of a human decision maker. the variables and information to
48 H.A. Simon, Behavioral and rational foundations ofeconomicdynamics

which he might attend (and to which he must attend to satisfy the strict
requirements of rationality) are innumerable. The hypothesis of bounded
rationality claims that human beings handle this difficulty by attending to
only a small part of the complexity about them. They make a highly
simplified model of the world, and they make their decisions in terms of that
model and the subset of variables that enter into it.
Now this approach may work very well (and may even give a good
approximation to Olympian rationality) if the number of very important
variables is small at any given time, and if this list of important variables
does not change from time to time without the change being noticed. The
decision maker of bounded rationality has both the problem of dealing with
a simplified model of the decision situation und of noticing when he needs to
change his model. Attentional mechanisms are critical to the quality of his
rationality.
If the price level is not changing significantly, then one does not suffer
from the money illusion if one fails to discriminate between real and money
prices. One good way to simplify the world model in a world of stable prices
is to ignore the distinction. That is almost certainly what most of us did in
our economic decisions twenty years ago. At some point in time, however,
we became aware that prices were no longer stable, and we began to take the
price level into consideration; we began to index our commitments. The
discrepancies between real and monetary quantities became so large that our
attention was drawn to them as important enough - among the myriad of
things we could consider - for our consideration.
Here we have a major discontinuity in the economic process, a major
structural shift taking place from one set of mechanisms to another. The
dynamics of the economic system depends on the timing of this shift, which
of course will not be made by all persons simultaneously. To handle this
discontinuity, and others like it, in our theories, we have to choose among
three alternatives. The first is to treat attentional shift as an exogenous
random shock. Then we simply give up the attempt to explain or predict it.
The second is to apply the familiar marginal arguments: We allocate just as
much effort to scanning the horizon for newly important variables as is
justified by the marginal value (in terms of improved decision making) of the
scanning. This second alternative is vacuous, providing us with no clue as to
how these marginal costs and benefits are to be estimated. The third
alternative is to study human attention and shifts of attention empirically in
order to develop a theory of individual and social determinants of focus of
attention. This important task of empirical inquiry has not yet been
addressed by the economics profession.

9. The debate on unemployment

As a final example of how auxiliary assumptions or irrationality, or rather,

 H.A. Simon, Behavioral and rational foundations of economic dynamics 49

bounded rationality, complement assumptions of perfect rationality in

economic theories, both classical and Keynesian, I should like to turn to the
much discussed topic of unemployment. The literature on this topic is so
enormous that I don’t really need to make any new arguments, but can
simply point to those that have been made and repeated many times. Let
me start with the notion, currently popular in some circles, of ‘natural level
of unemployment’. lo
The most familiar of Keynes’ local heresies was to introduce high viscosity
into the adjustment of the wage level in order to permit the economic system
to reside for considerable periods of time in a condition of below-equilibrium
employment. The radical (should I call it ‘ultra-classical’?) alternative to the
Keynesian heresy is to insist that labor supply and demand are always in
equilibrium, no matter what the official statistics of unemployment say. The
level of unemployment is natural - all of those wanting employment at the
current wage rate are employed.
The debate about natural levels of employment and unemployment can
turn into a very refined philosophical discussion of the meaning of the phrase
‘wanting employment’. Though I haven’t seen it done, it might even be
transformed into a classical discussion of free will versus determinism. But I
think the debate misses the real issue, which has been pointed out by
Thurow among others: Observed rates of unemployment fluctuate greatly
from year to year. A theory that dismisses these fluctuations as irrelevant
throws away an important aspect of the dynamics of the economic system.
Why do these fluctuations occur, and what impact, in turn, do they have
upon the other variables of the system?
A curious feature of the debate, nearly as true of the Keynesians as of their
opponents, is that almost no one (Thurow is again a recent and notable
exception) seems to question that there is a labor market in which workers
offer their services at one or another price and/or employers simultaneously
offer employment at one or another wage.
Thurow asks a very refreshing and revealing question: What in fact
happens if an unemployed person presents himself to a business firm and
offers to work (in a job for which he is qualified) at a wage below that being
paid to the current occupant of a position? Here again, I would prefer to
seek our answer by systematic empirical inquiry than by the methods of
casual empiricism. Since I am not familiar with any inquiry into the question,
I will hazard a casual answer. The offer would be refused, probably politely.
If the jobseeker persisted, on the grounds that he was simply behaving as his

“‘If awards were made for outrageous audaciousness in the coining of terms, certainly the
author of ‘natural level of unemployment’ would deserve one, as would Jack Muth, the author
of ‘rational expectations’. The virtue of such terms is that they win the argument instantly by
taking the breath away from would-be disputants, whose very skepticism now accuses them of
‘unnaturalness’ or ‘irrationality’, as the case may be.
50 H.A. Simon, Behavioral and rational foundations ofeconomic dynamics

economics textbook said he should, he would soon be referred for psychiatric

examination.
For whatever good or bad reasons, employers during periods of high
unemployment do not usually dismiss their current labor force and hire
another at a lower wage. Sometimes, in extremity, they threaten to close
down if their employees will not accept wage cuts. But I am reciting
(stylized) facts that are familiar to all of us, no matter how indirect our
information about labor markets.
Economic theory has been satisfied to substitute a hypothetical ‘labor
market’ - whether the sticky one of Keynes or the perfectly adjusting one of
the classical theory - for an actual study of the decisional and behavioral
processes that lead to employees being dismissed and to job offers being
made and being accepted or rejected. In the hypothetical labor market, the
actors are, in some definable sense, ‘rational’. But the assumption of their
rationality does not provide answers to the important economic questions
about the market. The actual level of employment predicted by such theories
derives not from the rationality assumption, but from the particular limits
that are supposed to operate on that rationality - whether they be
stickiness of wages (Keynes) or unexplained shifts in the balance of
preference for work and leisure (natural rate of unemployment).
One does not have to be a Marxist to notice an assymetry between the
roles of employee and employer in the real labor market. Before there can be
an offer or acceptance of employment, there must be a job to be filled. And
jobs exist in the minds of employers. Employers decide, from time to time,
that they want a work force of a certain size and composition. To be sure,
these decisions are not made in a vacuum; they are made in the anticipation
of profit, and in the light of a host of expectations about prices of output,
costs of production and the quality of labor that could be hired at various
wage rates. However they are made, there is a genuine decisional process
going on here, involving all kinds of guesses, hopes and fears about the
future, whose dynamics is central to the dynamics of the whole economic
system. Since armchair conjecturing about what this process must be has
only led to disagreement and confusion in the economics profession, it is
probably not too soon to begin investigating empirically what the process
actually is.
The picture I have just painted of the labor market is thoroughly
Schumpeterian. Schumpeter focused mainly on those employers who were
innovatively introducing new products, but the problem is not very different
for employers in general. Employers must envision and ‘create’ jobs, before
labor markets can operate. And if the number of jobs created is less than the
number of workers available at the going wage, the difference between these
numbers will show up in the official statistics as unemployment. It is less a
matter of explaining how there can be a departure from equilibrium than it
 H.A. Simon, Behavioral and rational foundations of economic dynamics 51

is of explaining how a system so constructed manages to keep close to

equilibrium (as it does) much of the time.
Again, I must stop short of stating the answer to the question I raise. For
my basic point is that, to find the answer, we must study the decision process
in much greater depth and with far greater thoroughness than it has ever
been studied.

10. Studying the bounds on rationality

A few main themes recur repeatedly in the previous sections. I should now
like to sum up the findings of those sections in a new set of ‘stylized facts’ -
this time, facts about the status and prospects of economic theory.

(I) There is broad agreement, at the level of aggregate statistics, on the

principal facts of economic growth in the developed countries, and the facts
of business cycles, unemployment and inflation.
(2) It is not likely that important new facts can be obtained by applying
sophisticated statistical techniques to aggregate time series. The residual
fluctuations in the data are mostly below the level of random noise.
(3) The picture revealed by the agreed-upon facts about growth and
cycles does not have high enough resolution to choose between major
alternative economic theories. But different theories, all consistent with the
aggregate data, lead to different policy prescriptions and to different
predictions whenever structure changes.
(4) The stylized facts of economic growth and business cycles cannot be
derived from the rationality postulates of classical and neoclassical theory
without introducing essential qualifications of those postulates. These
qualifications take the form of assumptions of bounded rationality (e.g., the
money illusion), or assumptions of exogenous shocks (e.g., unmotivated shifts
in work-leisure preferences). Both neoclassical and Keynesian writings are
mottled with such auxiliary assumptions, and their conclusions would not
follow if the assumptions were removed. In fact, with removal of the
auxiliary assumptions, which have mainly to do with the ways in which
expectations are formed about uncertain future events, the neoclassical and
Keynesian theories would become indistinguishable.
(5) The auxiliary assumptions one finds in neoclassical and Keynesian
analyses have no secure empirical base. They generally derive from the most
casual of casual empiricisms. They are not adequately supported by
aggregate data for the reasons already given. They are not adequately
supported by micro-level data (data derived from actual studies of the
decision-making processes of consumers and in business firms), because such
data have not been gathered systematically for most of the phenomena under
study.
52 H.A. Simon, Behavioral and rational foundations of economic dynamics

A rather obvious conclusion follows from these facts: that the continuing
progress of the economic theory of change and cycles requires massive
empirical work at the level of the economic agents who make decisions -
for example, the consumer and the firm. The dynamics of the economic
system depends critically on just how economic agents go about making their
decisions, and no way has been found for discovering how they do this that
avoids direct inquiry into and observation of the processes.
This thesis, sometimes called Behavioralism, has been opposed on a
number of grounds. First, it has been opposed as an abandonment of the
principle of rationality. I have tried to show here that the principle of
rationality has already been abandoned, and abandoned in fundamental
ways, by classicists, neoclassicists, Keynesians, and rational expectationists
alike. No new abandonment is proposed; just a major effort to discover
which modes of abandonment are empirically defensible.
A second objection (e.g., Blaug) is that Behavioralism has not yet produced
a complete theory that can compete with the classical one. It is not a theory,
but a set of hopes and promises of a future theory. Blaug argues that you
cannot beat something with nothing; there is no reason to abandon the
neoclassical theory until a new theory is in place. This is a Catch 22
argument: behavioralism can’t be developed without being pursued
empirically; but there is no reason for pursuing it since it does not offer, here
and now, an alternative to classical theory.
Behavioralism does not hold forward the hope of a theory that can be
arrived at in an armchair. The new theory can only be built through
extensive empirical research - on the scale, let us say, of the empirical
research of the last century or two in field biology or geology. Moreover, the
required research is of a kind that is novel to most economists. It does not
rest on statistical data derived from secondary sources, but calls for direct
observation of human behavior in the market and in the firm. The reason for
undertaking that research is that the classical program has failed: it patches
the rationality principle with ad hoc assumptions of bounded rationality, but
it has no way to choose among alternative patches without obtaining new
data. And macroeconomic studies are not a viable source for such data.
Empirical research on human decision making at the microeconomic level
need not be, and should not be, a blind accumulation of facts. Priorities
among the areas of decision to be investigated can be determined by the
needs of macroeconomics. But there is an even more powerful source of
heuristics for guiding the research. During the past twenty five years there
has been developed in cognitive psychology a theory of the information
processes that account for human problem solving and decision making for a
wide variety of tasks. l1 The theory has by now reached the point where it
has a good deal to say about complex, professional level human cognitive

’ ‘See, for example, Newell and Simon (1972)

 H.A. Simon, Behavioral and rationalfoundations of economic dynamics 53

activities, like making medical decisions, using financial statements to identify

and analyze corporate problems, or making investment decisions. In parallel
with the theory, a methodology has been developed for using verbal
thinking-aloud protocols and other verbal responses as empirical data for
testing hypotheses about information processes, and a powerful technology
for modeling the processes as computer programs and simulating their
behavior. The substantive and methodological content of this work in
cognitive psychology provide a theoretical framework for empirical work in
microeconomics that was not available a decade or two ago. Lack of
theoretical or methodological guidelines is no longer a credible reason for
delaying the construction of an empirically based microeconomics.

11. Requirements for a Schumpeterian dynamics

Let me end on a more substantive note, deriving from the earlier

discussion of the labor market. In the language of classical economics there
is little room for a fundamental asymmetry between the two parties to a
bargain. There is no room for a concept like ‘initiative’. In a Walrasian
tatonnement it is irrelevant who makes the first bid. Similarly, there is no
distinction in classical economics between action and inaction. Each time
period, the actors must choose some value of each decision value, and there
is no special value called ‘inaction’. An investment level of zero is not
qualitatively different in the theory from a level of 100.
In the everyday world, matters are quite different. We do distinguish
certain states as states of inaction. In a baseball game, O-O is a possible score,
but there is a difference between a tie game and a day on which no game
was played. Inaction is the state that prevails when no positive action is
being undertaken, and it is qualitatively different from action. Action has to
be motivated. The actor must attend to some stimulus, external or internal,
that leads him to consider action relevant to it. In the absence of motivation,
inaction prevails.
Applying these ideas to economic affairs, we arrive at a picture of them
that looks very different from Walrasian equilibrium. In the Walrasian
picture, there is at all times a fixed set of markets, each with its supply
function and demand function and a price fluctuating around the equilibrium
value. Markets are neither created nor destroyed. In an economy where
actions must be positively motivated (let me call it a Schumpeterian
economy), commodities are produced only after someone is motivated to
consider producing them. Investments are made only when someone is
motivated to pay attention to a potential investment opportunity and decides
to invest. Job slots are created only when employers attend to the need for
more workers and decide to try to employ them.
In a Schumpeterian world, human attention is a central endogenous
54 H.A. Simon, Behavioral and rationalfiundations ofeconomic dynamics

variable, whose determination is a key to the direction and level of activity of

the economic system, both in the long run and in the short. Attention, in
turn, is a function of the information and communication flows in the
society. There are habits of attention, a potent source of inertia and stability.
There are also trends and fads in attention - societies as well as individuals
may be said to have changing foci of attention.
There is a good deal of room for traditional economic variables in such a
world. Economic actors may attend to prices, and they may become aware
of, and be motivated to anticipate, the possibilities for price changes. One
might even say that a Schumpeterian model is simply a next step in the same
direction as that in which economic theories have been moving for forty
years. During that time we shifted from an almost exclusive preoccupation
with statics to a full concern with dynamics. We introduced expectations into
the dynamic model, first describing those expectations as adaptations to the
memory of the recent past. Then we (or the rational expectationists among
us) took the leap of making expectations depend on an understanding of the
economic system itself.
A very natural next step for economics is to maintain expectations in the
strategic position they have come to occupy, but to build an empirically
validated theory of how attention is in fact directed within a social system,
and how expectations are, in fact, formed. Taking that next step requires that
empirical work in economics take a new direction, the direction of micro-
level investigation proposed by Behavioralism.

References

Adelman, I. and F.L. Adelman, 1959, The dynamic properties of the Klein-Goldberger Model,
Econometrica 27, 596625.
Becker, G.S., 1981, A treatise on the family (Harvard University Press, Cambridge, MA).
Blaug, M., 1980, The methodology of economics (Cambridge University Press, Cambridge)
Hicks, J.R., 19.50, The trade cycle (The Clarendon Press, Oxford).
Holt, CC., F. Modigliani, J.F. Muth and H.A. Simon, 1960, Planning production, inventories
and work force (Prentice-Hall, Englewood Cliffs, NJ).
Keynes, J.M., 1936, The general theory of employment, interest and money (Macmillan,
London).
Leontief, W., 1971, Theoretical assumptions and nonobserved facts, American Economic Review
61, l-7.
Lucas, R.E., Jr., 1977, Understanding business cycles, in: K. Brunner and A. Meltzer, eds.,
Stabilization of the domestic and international economy (North-Holland, Amsterdam).
Lucas, R.E., Jr., 1980, Methods and problems in business cycle theory, in: Studies in business-
cycle theory (The MIT Press, Cambridge, MA).
Muth, J.F., 1961, Rational expectations and the theory of price movements, Econometrica 29,
315-335.
Nelson, R.R. and S.G. Winter, 1982, An evolutionary theory of economic change (Harvard
University Press, Cambridge, MA).
Newell, A. and H.A. Simon, 1972, Human problem solving (Prentice-Hall, Englewood Cliffs, NJ).
Samuelson, P.A., 1947, Foundations of economic analysis (Harvard University Press,
Cambridge, MA).
 H.A. Simon, Behavioral and rational foundations of economic dynamics 55

Schumpeter, J.A., 1934, The theory of economic development (Harvard University Press,
Cambridge, MA).
Simon, H.A., 1977, Models of discovery (D. Reidel, Dordrecht).
Simon, H.A. and A. Ando, 1961, Aggregation of variables in dynamic systems, Econometrica 29,
111-138.
Solow, R.M., 1970, Growth theory (Oxford University Press, Oxford).
Thurow, L.C., 1983, Dangerous currents: The state of economics (Random House, New York).
Turnovsky, S.J., 1977, Macroeconomic analysis and stabilization policies (Cambridge University
Press, Cambridge).