Modifying the Rational Expectations Assumption in a Large World Model

Abstract

Many of the large econometric models in use around the world have introduced rational expectations(RE) as their main operating assumption over the last ten years, largely because of the issues raised by Lucas (1976). These include the Fair model, Minford’s Liverpool model, the quarterly models of the National Institute of Economic and Social Research, the London Business School and HM Treasury model in the UK, Multimod at the IMF, the Global Econometric Model (GEM) and a number of others. A considerable amount of effort has been spent in the academic literature on attempting to test the relevance of the RE assumption in the real world, we will not attempt to survey this literature here, a good introduction is the book by Pesaran (1987). This literature has not found overwhelming support for the RE assumption, but on the whole it has not fared too badly. We want however to make a clear distinction between these tests and the implementation of RE in an econometric model. Most of the standard tests of RE are attempting to test if agents use all available information in an efficient way. The model under study is usually not either complete or detailed and so forcing variables are often generated through unrestricted VARs or in some other, non-structural, way such as instrumental variable estimation. These tests may then be viewed as a test of a weak form of rational expectations. When the large econometric models make the RE assumption they are imposing a much stronger assumption, one which we feel is of a quite different nature. They are assuming that agents actually use that particular model to form their expectations. That is, to take an example, agents have full knowledge of the London Business School model; they believe it to be the true model of the economy and that they use it to form full model consistent expectations. Such an assumption has never been tested before it has been imposed on a model.