Introduction 1.1 According to the rational expectations hypothesis people form expectations as if they know the process which will ultimately generate the actual outcomes. In practice this means that when an economist formulates a stochastic economic model, 'rationality' is imposed by modelling expectations as if the people forming them essentially know this economist's model (including the parameters of the stochastic error generating process), and thus utilize this information in forming their (mathematical) expectations. This process ensures a degree of model consistency and guarantees that expectational errors are unbiased.l
In 'justifying' this hypothesis its new classicist proponents typically resort to criticising alternative very simple and rigid expectations models and frequently attempt to associate the latter with Keynesian theory. In this manner Keynesians are often accused of underestimating the capabilities of people to understand and interpret the situations surrounding themselves and about which they form expectations.
Such associations and accusations are, of course, illegitimate; whilst Keynesians certainly hold different views to the new classists concerning the structures of underlying, economic processes, they do not reject the assumption of capable and informed expectation formation (see for example Lawson, 1981).
It does seem the case, however, that the latter assumption could be more fully exploited by Keynesian (and other) model builders, especially if there exist groups of people whose informed predictions are a relevant source of otherwise scarce information. This in fact, represents the main proposal of this paper: that in a situation where data limitations and other factors inhibit detailed economic analyses, a 'rational' response, for purposes of model construction, is to attempt to identify the subjective model implicitly adopted as a useful approximation to an underlying process, by a relevant informed person. For want of a better term this informed person will be referred to throughout as an expert.
To slightly misquote Muth (1961) 'the procedure can be rephrased a little more precisely as follows: that predictions of modellers (or more generally the forecaster's * Presented at 'The 8th Conference on Problems of Building and Estimation of Large Econometric Models' held in hSd2, Poland in December 1981. I am grateful to my colleagues on the Cambridge Growth Project, and in particular to William Peterson, for helpful comments on an earlier draft of this paper.
tSidgwick Avenue, Cambridge CB3 9DE, UK. t For interesting discussions of the rational expectations hypothesis see Friedman, 1979 andShiller, 1978, and for the defining article see Muth, 1961.