Exchange rate forecasts with the Michigan quarterly econometric model of the US economy

Gandolfo et al. [Journal of Banking and Finance 14 (1990) 965-992] have shown that their continuous time model of the Italian economy produces better ex post out-of-sample forecasts of the exchange rate than either existing structural or random-walk models. When the Michigan Quarterly Econometric Model is used, it is found that ex post out-of-sample forecasts of the trade-weighted value of the US dollar produced by the model are also superior to forecasts of a random-walk model. However, ex ante forecasts in which all the exogenous as well as the endogenous variables are forecast are less accurate than those produced by the random walk. The price of imported goods in foreign currency, an exogenous variable in both the Michigan and Italian econometric models, is the key variable in the Michigan model which explains the divergence of the ex ante and ex post forecasting results.