The transformation of macroeconomic policy and research prize Lecture, December 8, 2004 by

The course requires students to carry out quantitative analyses to address

specific questions. They use the methodology that Finn and I developed for

the study of business cycles to address policy issues. A typical exercise is to 

determine whether a proposal made by a public opinion leader or government

official will have the intended consequence. One question they were assigned

concerns financing of transfer payments. Their finding was totally counter to

then-conventional wisdom. I will return to this finding later because it is an

implication of Finn Kydland’s and my business cycle theory.

2. THE TRANSFORMATION OF MACROECONOMIC POLICY

In this section I first describe what macroeconomic models were before the

transformation and what they are after the transformation. Then I describe 

policy selection before and after the transformation. Before the transformation,

what is evaluated is a policy action given the current situation. Policy discussions

were in terms of questions such as whether the money supply should be increas-

ed or reduced. In his critique (1976), Robert Lucas established that questions

such as these are not well posed in the language of dynamic economic theory.

After the transformation, what is evaluated is a policy rule. A policy rule

specifies the current policy action as a function of the current economic situa-

tion. As Finn and I found, no best policy rule exists. Typically, the policy rule

is best given that it will be followed in the future. Any such rule is by definition

time-consistent, but except in empirically uninteresting cases, Finn and I

show that time-consistent rules are not optimal; indeed, they lead to bad 

outcomes. All that can be hoped for is to follow a good rule, and this requires

economic and political institutions that sustain this rule.

Macroeconomic Models before the Transformation

Macroeconomic models were systems of equations that determined current

outcomes given the values of the current policy actions, values of predetermined

variables, and values of any stochastic shocks. Thus, physical models and 

pre-transformation macro models have the same mathematical structure.

The basic mathematical structure of both is

x

t+

1

= f (x

t

,u

t

,

␧

t

).

The state or position of the dynamic system at the beginning of period t is x

t

,

the control or policy variables are u

t

, and the stochastic shocks are 

␧

t

.

With the system-of-equations approach, each equation in the system is 

determined up to a set of parameters. The simple prototype system-of-equations

macro model has a consumption function, an investment equation, a money

demand function, and a Phillips curve. Behind all these equations were a rich

empirical literature and, in the case of the consumption function, money 

demand function, and investment equation, some serious theoretical work.

The final step was to use the tools of statistical estimation theory to select the

parameters that define the function f.

372

K4_40319_Prescott_358-395  05-08-18  11.41  Sida 372

---

I worked in this tradition. In my dissertation, I formulated the optimal policy

selection problem as a Bayesian sequential decision problem. The problem is

a difficult one because the policy actions taken today affect the distribution of

the posterior distribution of the values of the coefficients of the equations.

The macroeconometric models organized the field. Success in macroeco-

nomics was to have your equation incorporated into the macroeconometric

models. Indeed, Lucas and I were searching for a better investment equation

when in 1969 we wrote our paper “Investment under Uncertainty,” a paper that

was published two years later in 1971.

A key assumption in the system-of-equations approach is that the equations

are policy invariant. As Lucas points out in his critique, which I delivered in 1973,

this assumption is inconsistent with dynamic economic theory. His insight made

it clear that there was no hope for the neoclassical synthesis – that is, the devel-

opment of neoclassical underpinnings of the system-of-equations macro models.

Fortunately, with advances in dynamic economic theory an alternative set

of tractable macro models was developed for drawing scientific inference.

The key development was recursive competitive equilibrium theory in Lucas

and Prescott (1971) and Lucas (1972). Equilibrium being represented as a

set of stochastic processes with stationary transition probabilities was crucial

to the revolution in macroeconomics.

Macroeconomic Models after the Transformation

Models after the transformation are dynamic, fully articulated model economies

in the general equilibrium sense of the word economy. Model people maximize

utility given the price system, policy, and their consumption possibility set;

firms maximize profits given their technology set, the price system, and policy;

and markets clear. Preferences, on the one hand, describe what people choose

from a given choice set. Technology, on the other hand, specifies what outputs

can be produced given the inputs. Preferences and technology are policy invariant.

They are the data of the theory and not the equations as in the system-of-

equations approach. With the general equilibrium approach, empirical

knowledge is organized around preferences and technology, in sharp contrast

to the system-of-equations approach, which organizes knowledge about 

equations that specify the behavior of aggregations of households and firms.

The Time Inconsistency of Optimal Policy

Before the transformation, optimal policy selection was a matter of solving

what the physical scientists called a control problem. This is not surprising, given

that the system-of-equations approach was borrowed from the physical sciences.

With such systems, the principle of optimality holds – that is, it is best to choose

at each point in time policy action that is best given the current situation and

the rules by which policy will be selected in the future. The optimal policy is

time-consistent, and dynamic programming techniques can be used to find

the optimal policy as in the physical sciences. This is true even if there is 

uncertainty in the model economy.

373

K4_40319_Prescott_358-395  05-08-18  11.41  Sida 373

---