It is well known that modeling exchange rates is difficult. Meese and Rogoff's (1983) results show that a random walk model performs as well as or better than a variety of structural models, where the forecasts from the structural models are based on the actual values of the future explanatory variables. Because of these and other results, the view has become fairly widespread that structural models of exchange rates are not very good. There is, however, somewhat of a dichotomy in the literature between those who deal with small models, where the focus is almost exclusively on exchange rates, and those who deal with large macroeconometric models, where exchange rates make up only a small subset of the endogenous variables. One might have thought, for example, that in a survey like Levich's (1985) both types of models would be considered, but the large models are given only one footnote (fn. 19, p. 1001). It may be that exchange rate determination within the context of large models has not been given a sufficient hearing. Exchange rate and interest rate equations are estimated and analyzed for 17 countries in this paper. This study is part of a larger project of constructing a multicountry econometric model. One of the aims of this paper is to see if the exchange rate equations that are part of my multicountry model also suffer from the Meese and Rogoff criticism. The results show that the view that structural exchange rate models are not very good may be too pessimistic. The theory upon which the multicountry econometric model is based is outlined in Section II. The exchange rate and interest rate equations are estimated in Section III and tested in Section IV.
