Federal Reserve Bank of Minneapolis

Quarterly Review
Summer 1980

What's Wrong With Macroeconomics (P. v

Deficit Policies, Deficit Fallacies (p. 2)
The Search for a Stable Money Demand Equation (p. 5)
Rational Expectations and the Reconstruction
of Macroeconomics (p. is)
District Conditions (P. 20)
Federal Reserve Bank of Minneapolis
Q u a r t e r l y R e v i e w vol. 4, no. 3
This publication primarily presents economic research aimed at improving policymaking
by the Federal Reserve System and other governmental authorities.

Produced in the Research Department. Edited by Arthur J. Rolnick, Kathleen S. Rolfe, and Alan Struthers, Jr.
Graphic design and charts drawn by Phil Swenson, Graphic Services Department.

Address requests for additional copies to the Research Department,

Federal Reserve Bank, Minneapolis, Minnesota 5 5 4 8 0 .

Articles may be reprinted if the source is credited and the Research

Department is provided with copies of reprints.

The views expressed herein are those of the authors and not necessarily those
of the Federal Reserve Bank of Minneapolis or the Federal Reserve System.
 Federal Reserve Bank of Minneapolis Quarterly R e v i e w / S u m m e r 1980

Rational Expectations and the Reconstruction

ofMacroeconomics*

Thomas J. Sargent, Adviser

Research Department
Federal Reserve Bank of Minneapolis
and Professor of Economics
University of Minnesota

Fans of the National Football League may well have National Football League announced a rule change,
observed the following behavior by the Houston Oilers effective next Sunday, which gave a team six downs in
during the 19— season. At home against Kansas City, which to make a first down. Would we still expect
when confronted with a fourth down in its own end of Houston to punt on fourth down? Clearly not; at least
the field, Houston punted 100 percent of the time. The no one familiar with the game of football would.
next week, at St. Louis, in the same situation, Houston What this example indicates is that historical pat-
punted 93 percent of the time. The following week at terns of human behavior often depend on the rules of
Oakland, again in that situation, Houston again punted the game in which people are participating. Since much
100 percent of the time, as it did the subsequent week human behavior is purposeful, it makes sense to expect
at home against San Diego, and so on and on for the that it will change to take advantage of changes in the
rest of the season. In short, on the basis of the time rules. This principle is so familiar to fans of football
series data, Houston has a tendency to punt on fourth and other sports that it hardly bears mentioning. How-
downs in its own territory, no matter what team it plays ever, the principle very much deserves mentioning in
or where. the context of economic policy because here it has
Having observed this historical record, suppose it is been routinely ignored — and with some devastating
our task to predict how Houston will behave in the results.1 Adherents of the theory of rational expecta-
future on fourth and long in its own territory. For tions believe, in fact, that no less than the field of
example, suppose that next week Houston is to play an macroeconomics must be reconstructed in order to
expansion team at Portland that it has never played take account of this principle of human behavior. Their
before. It seems safe to predict that Houston will punt efforts to do that involve basic changes in the ways
on fourth downs in its own territory at Portland. This economists formulate, simulate, and predict with econ-
sensible prediction is not based on any understanding ometric models. They also call for substantial changes
of the game of football, but rather on simply extrapolat- in the ways economic policymakers frame their op-
ing a past behavior pattern into the future. tions.2
In many cases, we would expect this method of
prediction to work well. However, for precisely those Models must let behavior change
cases in which predictions are most interesting, the with the rules of the game
extrapolative method can be expected to break down. In order to provide quantitative advice about the
For instance, suppose that the Commissioner of the
C h a r l e s Whiteman and Ian Bain are responsible for impressing upon me
*This paper is based on remarks prepared for the September 1980 Inter- the many parallels between football and macroeconomics.
2
national Symposium of the Hosei University in Tokyo, Japan. This is the message of Lucas 1976.

15
 Federal Reserve Bank of Minneapolis Quarterly R e v i e w / S u m m e r 1980

effects of alternative economic policies, economists model people as making decisions in dynamic settings
have constructed collections of equations known as in the face of well-defined constraints. Included among
econometric models.3 For the most part, these models these constraints are laws of motion over time that
consist of equations that attempt to describe the behav- describe such things as the taxes that people must pay
ior of economic agents — firms, consumers, and gov- and the prices of the goods that they buy and sell. The
ernments— in terms of variables which are assumed to hypothesis of rational expectations is that people under-
be closely related to their situations. Such equations stand these laws of motion. The aim of the research is
are often called decision rules since they describe the to build models that can predict how people's behavior
decisions people make about things like consumption will change when they are confronted with well-under-
rates, investment rates, and portfolios as functions of stood changes in ways of administering taxes, govern-
variables that summarize the information people use to ment purchases, aspects of monetary policy, and the
make those decisions. For all of their mathematical like.
sophistication, econometric models amount to statisti-
cal devices for organizing and detecting patterns in the The Investment Decision as an Example
past behavior of people's decision making, patterns A simple example will serve to illustrate both the
which can then be used as a basis for predicting their principle that decision rules depend on the laws of
future behavior. motion that agents face and the extent that standard
macroeconomic models have violated this principle.
As devices for extrapolating future behavior from
Let kt be the capital stock of an industry and rt be a tax
the past under a given set of rules of the game, or
rate on capital. Let zt be the first element of zt9 a vector
government policies, these models appear to have
of current and lagged variables including those that the
performed well.4 They have not, however, when the
government considers when it sets the tax rate on
rules have changed. In formulating advice for policy-
capital. We have rt = eTzt, where e is the unit vector
makers, economists have routinely used these models
with unity in the first place and zeros elsewhere.7 Let a
to predict the consequences of historically unprece-
firm's optimal accumulation plan require that capital
dented, hypothetical government interventions that can
acquisitions obey8
only be described as changes in the rules of the game.
In effect, the models have been manipulated in a way 00
a > 0
which amounts to assuming that people's patterns of
(1) kt = \ k t _ x - a Z $ E t T t + j 0 < X < 1
behavior do not depend on those properties of the
J=0 0 < 8 < 1
environment that government interventions would
change. The assumption has been, that is, that people
will act under the new rules just as they have under the
old, so that even under new rules past behavior is still a 3
Lucas and Sargent 1979 provides a brief explanation of econometric
models and their uses in macroeconomics.
reliable guide to future behavior. Econometric models 4
This evidence is cited by Litterman (1979) and his references.
used in this way have not been able to accurately 5
Sims (1980) and Lucas (1976) describe why econometric models can
predict the consequences of historically unparalleled perform well in extrapolating the future from the past, assuming no changes in
interventions.5 To take one painful recent example, rules of the game, while performing poorly in predicting the consequences of
changes in the rules.
standard Keynesian and monetarist econometric mod- 6
For an example of such research and extensive lists of further references,
els built in the late 1960s failed to predict the effects on see Hansen and Sargent 1980 and Lucas and Sargent forthcoming.
output, employment, and prices that were associated 7
Here T denotes matrix transposition.
with the unprecedented large deficits and rates of 8
The investment schedule (1) can be derived from the following dynamic
money creation of the 1970s. model of a firm. A firm chooses sequences of capital to maximize
Eo
Recent research has been directed at building econ- (*) $/{Akt ~ " fa ~ * M )2 }
ometric models that take into account that people's be-
havior patterns will vary systematically with changes w h e r e / i , / 2 , / 3 , D > 0; 0 > P > 1; and Eq is the mathematical expectation
in government policies or the rules of the game.6 Most operator conditioned on information known at time 0. The maximization is
subject to (k t -1, z t ) being known at the time /. Maximization problems of this
of this research has been conducted by adherents of the kind are analyzed in Sargent 1979. The parameters X, a, and S can be shown to
so-called hypothesis of rational expectations. They be functions o f / i , / 2 , / 3 , and d.

16
 Federal Reserve Bank of Minneapolis Quarterly R e v i e w / S u m m e r 1980

where Etzt+j is the tax rate at time t which is expected investment schedule, since via equation (2) all of these
to prevail at time (t+j). variables help agents forecast future tax rates. (Com-
Equation (1) captures the notion that the demand pare this with the common econometric practice of
for capital responds negatively to current and future using only current and lagged values of the tax rate as
tax rates. However, equation (1) does not become an proxies for expected future tax rates.)
operational investment schedule or decision rule until The fact that (2) and (3) share a common set of
we specify how agents' views about the future, Etzt+J-, parameters (the A matrix) reflects the principle that
are formed. Let us suppose that the actual law of mo- firms' optimal decision rule for accumulating capital,
tion for zt is described as a function of current and lagged state and
information variables, will depend on the constraints
(2) zt+l=Azt (or laws of motion) that firms face. That is, the firm's
pattern of investment behavior will respond system-
where A is a matrix conformable with zt? If agents atically to the rules of the game for setting the tax rate
understand this law of motion for zt, the first element r t . A widely understood change in the policy for
of which is zt, then their best forecast of r t +j is eTAJzt. administering the tax rate can be represented as a
We impose rational expectations by equating agents' change in the first row of the A matrix. Any such
expectations Etrt+j to this best forecast. Upon impos- change in the tax rate regime or policy will thus result
ing rational expectations, some algebraic manipulation in a change in the investment schedule (3). The de-
implies the operational investment schedule pendence of the coefficients of the investment sched-
ule on the environmental parameters in A is reasonable
(3) kt =\kt_\ - aeT(I-8A)~xzt. and readily explicable as a reflection of the principle
that agents' rules of behavior change when they en-
In terms of the list of variables on the right-hand counter changes in the environment in the form of new
side, equation (3) resembles versions of investment laws of motion for variables that constrain them.
schedules which were fit in the heyday of Keynesian To illustrate this point, consider two specific tax
macroeconomics in the 1960s. This is not unusual, for rate policies. First consider the policy of a constant tax
the innovation of rational expectations reasoning is rate zt+j = zt for all j > 0. Then zt = r„ A = 1, and the
much more in the ways equations are interpreted and investment schedule is
manipulated to make statements about economic pol-
icy than in the look of the equations that are fit. Indeed, (4) k t = \k(_j + h 0 z t where h 0 = -a/(l-<5).
the similarity of standard and rational expectations
equations suggests what can be shown to be true gen- Now consider an on-again, off-again tax rate policy of
erally: that the rational expectations reconstruction of the form zt = - r M . In this case zt = zt, A = - 1 , and
macroeconomics is not mainly directed at improving the investment schedule becomes
the statistical fits of Keynesian or monetarist macro-
economic models over given historical periods and that (5) kt =\kt_x + h0zt where now
its success or failure cannot be judged by comparing /*o =-<*/( 1+5).
the Rvs of reconstructed macroeconomic models with
those of models constructed and interpreted along ear- Here the investment schedule itself changes as the
lier lines. policy for setting the tax rate changes.
Under the rational expectations assumption, the Standard econometric practice has not acknowl-
investment schedule (3) and the laws of motion for the edged that this sort of thing happens. Returning to the
tax rate and the variables that help predict it (2) have more general investment example, the usual econo-
a common set of parameters, namely, those of the metric practice has been roughly as follows. First, a
matrix A. These parameters appear in the investment model is typically specified and estimated of the form
schedule because they influence agents' expectations
of how future tax rates will affect capital. Further,
notice that all of the variables in zt appear in the 9
The eigenvalues of A are assumed to be less than <H in absolute value.

17
 Federal Reserve Bank of Minneapolis Quarterly R e v i e w / S u m m e r 1980

(6) kt = X kt-i + hzt mand functions for assets are all examples of such
rules of choice. In dynamic settings, regarding the
where h is a vector of free parameters of dimension parameters of these rules of choice as structural or
conformable with the vector zt. Second, holding the invariant under policy interventions violates the princi-
parameters h fixed, equation (6) is used to predict the ple that optimal decision rules depend on the environ-
implications of alternative paths for the tax rate r t . This ment in which agents believe they are operating.
procedure is equivalent to estimating equation (4) from If parameters of decision rules cannot be regarded
historical data when r t = rt_x and then using this same as structural or invariant under policy interventions,
equation to predict the consequences for capital ac- deeper objects that can must be sought. The best that
cumulation of instituting an on-again, off-again tax rate can be hoped for is that parameters characterizing
policy of the form zt = - r M . Doing this assumes that a private agents' preferences and technologies will not
single investment schedule of the form (6) can be found change when changes in economic policy change the
with a single parameter vector h that will remain fixed environment. If dynamic econometric models were
regardless of the rules for administering the tax rate.10 formulated explicitly in terms of the parameters of
The fact that equations (2) and (3) share a common preferences, technologies, and constraints, in principle
set of parameters implies that the search for such a they could be used to predict the effects on observed
regime-independent decision schedule is misdirected behavior of changes in policy rules. In terms of our
and bound to fail. This theoretical presumption is investment example with equations (2) and (3), the
backed up by the distressing variety of instances in idea would be to estimate the free parameters of the
which estimated econometric models have failed tests model (X, a, <5, A). With these estimates, economists
for stability of coefficients when new data are added. could predict how the investment schedule would
This problem cannot be overcome by adopting more change if different A9s occurred.11
sophisticated and more general lag distributions for the
vector /z, as perhaps was hoped in the 1960s. Policymakers must choose among alternative
rules, not isolated actions
These ideas have implications not only for theoretical
General Implications
and econometric practices, but also for the ways in
The investment example illustrates the general pre- which policymakers and their advisers think about the
sumption that the systematic behavior of private agents choices confronting them. In particular, the rational
and the random behavior of market outcomes both will expectations approach directs attention away from
change whenever agents' constraints change, as when particular isolated actions and toward choices among
government policy or other parts of the environment feasible rules of the game, or repeated strategies for
change. To make reliable statements about policy choosing policy variables. While Keynesian and mon-
interventions, we need dynamic models and econo- etarist macroeconomic models have been used to try to
metric procedures which are consistent with this gen- analyze what the effects of isolated actions would be, it
eral presumption. Foremost, we need a new and stricter is now clear that the answers they have given have
definition of the class of parameters that can be necessarily been bad, if only because such questions
regarded as structural. The body of doctrine associ- are ill-posed.
ated with the simultaneous equations model in econo- In terms of our investment example, by selecting
metrics properly directs the attention of the researcher different values for the first row of A, we can analyze
beyond reduced form parameters to the parameters of the effects on current and subsequent investment of
structural equations which are meant to describe those switching from one well-understood policy for setting
aspects of people's behavior that remain constant
across a range of hypothetical environments. Although
such structural parameters are needed to analyze an 10
This is analogous to assuming that Houston's propensity to punt on fourth
interesting class of policy interventions, most often down does not depend on the number of downs per series determined by the
included among them have been parameters of equa- N F L rules.
11
As claimed in footnote 8, the parameters (A, a, 8) can be shown to be
tions describing the rules of choice for private agents. functions of the parameters C / 1 , / 2 , / 3 , ^ ) of the present value function being
Consumption functions, investment schedules, and de- maximized in (*).

18
 Federal Reserve Bank of Minneapolis Quarterly R e v i e w / S u m m e r 1980

the tax rate to another—that is, we can analyze the about the proper strategy for repeatedly adjusting tax
effects of different strategies for setting the tax rate. rates in response to the state of the economy, year in
However, we cannot analyze the effects on current and and year out. Strategic questions of this nature abound
subsequent investment of alternative actions consisting in fiscal, monetary, regulatory, and labor market mat-
of different possible settings for the tax rate zt at a ters. Private agents face the problem of determining the
particular point in time t = 7. For in order to make government regime under which they are operating,
predictions, we must specify agents' views about the and they often devote considerable resources to doing
law of motion A, and this is not done when we simply so. Whether governments realize it or not, they do
consider actions consisting of alternative settings for zt make decisions about these regimes. They would be
at one isolated point in time. This idea is so widely wise to face these decisions deliberately rather than
accepted as to be uncontroversial among decision ignore them and pretend to be able to make good
theorists (and football fans); but even today practicing decisions by taking one seemingly unrelated action
macroeconomists usually ignore it. after another.
To take a concrete example, in the United States
there was recently interest in analyzing what would
happen to the rate of domestic extraction of oil and gas
if the tax on profits of oil producers increased a lot on a
particular date. Would supply go up or down if the tax References
were raised to X percent on July 1 ? The only scientifi-
cally respectable answer to this question is "I don't
know." Such a rise in the oil-profits tax rate could be Hansen,Lars P., and Sargent,Thomas J. 1980. Formulating and estimating
interpreted as reflecting one of a variety of different tax dynamic linear rational expectations models. Journal of Economic Dynam-
strategies (A matrices), each with different implica- ics and Control 2 (February): 7 - 4 6 .

tions for current and prospective extraction of oil. Litterman, Robert B. 1979. Techniques of forecasting using vector auto-
regressions. Research Department Working Paper 115. Federal Reserve
For example, suppose that oil companies had rea- Bank of Minneapolis, Minn.
son to believe that the increase in the tax is temporary Lucas, Robert E., Jr. 1976. Econometric policy evaluation: a critique. In The
and will be repealed after the election. In that case, Phillips curve and labor markets, ed. K. Brunner and A . H. Meltzer.
they would respond by decreasing their rate of supply Carnegie-Rochester Conference Series on Public Policy 1: 1 9 - 4 6 . Amster-
dam: North-Holland.
now and increasing it later, thus reallocating their sales
Lucas, Robert E., Jr., and Sargent, Thomas J. 1979. After Keynesian macro-
to periods in which their shareholders get a larger share economics. Federal Reserve Bank of Minneapolis Quarterly Review 3
of profits and the government a smaller share. Yet (Spring): 1-16.
suppose that oil companies believed that the increase eds. Forthcoming. Rational expectations and econometric
in the tax rate on July 1 is only the beginning and that practice. Minneapolis, Minn.: University of Minnesota Press.

further increases will follow. In that case, the response Sargent, Thomas J. 1979. Macroeconomic theory. N e w York: Academic
Press.
to the tax rate increase would be the reverse: to
increase supply now and decrease it later in order to Sims, Christopher A. 1980. Macroeconomics and reality. Econometrica 48
(January): 1 - 4 8 .
benefit companies' shareholders. This example illus-
trates that people's views about the government's
strategy for setting the tax rate are decisive in deter-
mining their responses to any given actions and that the
effects of actions cannot be reliably evaluated in iso-
lation from the policy rule or strategy of which they are
an element.
What policymakers (and econometricians) should
recognize, then, is that societies face a meaningful set
of choices about alternative economic policy regimes.
For example, the proper question is not about the size
of tax cut to impose now in response to a recession, but

19