time series. This decision on the part of the most prestigious theorist of his day freed a generation of economists from the discipline imposed by equilibrium theory, and,as 1 have described, this freedom was rapidly and fruitfully exploited by macroeconometricians. Now in possession of detailed, quantitatively aCCUI’dte replicas of the actual economy, economists appeared to have an inexpensive means to evaluate various proposed economic policy measures. It seemed legitimate to treat policy recommendations which emerged from this procedure as though they had been experimel:tally tested, even if such policies had never been attempted in any actual economy. Yet the ability of a model to imitate actual behavior in the way tested by the Adelmans (1959) has almost nothing to do with its ability to make accurate conditional forecasts, to answer questions of the form: how would behavior have differed had certain policies been different in specified ways? This ability requires invariance of the structure of the model under policy variations of the type being studied. Invariance of parameters in an economic model is not, of course, a property which can be assured in advance, but it seems reasonable to hope that neither tastes nor technology vary systematically with variations in countercyclical policies. In contrast, agents’ decision rules will in general change with changes in the environment. An equilibrium model is, by definition, con- structed so as to predict how agents with stable tastes and technology will choose to respond to a new situation. Any disequilibrium model, constructed by simply codifying the decision rules which agents have found it useful to use over some previous sample period, without explaining why these rules were used, - will be of no use in predicting the consequences of nontrivial policy changes. 12
The quantitative importance of this problem is, of course, a matter to be settled by examination of specific relationships in specific models. I have argued elsewhere’ that it is of fatal importance in virtually all sectors of modern macro- economic models, primarily because of the faulty treatment of expectations in these models. Rather than review these arguments in detail, let me cite the most graphic illustration: our experience during the recent “stagflation.” As recently as 1970, the major U. S. econometric models implied that expansionary monetary and fiscal policies leading to a sustained inflation of about 4 percent per annum wo~11d lead also to sustained unemployment rates of less than 4 percent, or about a full percentage point lower than unemploy- ment has averaged during any long period of U. S. history. lo These forecasts were widely endorsed by many economists not themselves closely involved in econometric forecasting. Earlier, Friedman (1968) and Phelps (1968) had argued, purely on the basis of the observation that equilibrium behavior is invariant under the units change represented by sustained inflation, that no sustained decrease in unemployment would result from sustained inflation.