A Forecasting Model For Exchange Rate Changes

This study develops a discriminant analysis model to predict changes in exchange rates of foreign currencies over the medium term. Using annual economic data for 20 major countries for the years 1972 through 1978, the study identified relevant economic variables which correctly classified the currency of a country as appreciating or depreciating against the U.S. dollar one year in the future. Because annual data are available in April, the model would give approximately eight months early warning. Four variables--international reserves, money supply, price levels, and current balance of payments--were found to have significant explanatory power. The model determined approximate weights for each variable. Overall, the model had a prediction accuracy of 75%. A holdout sample of predictions for 1979 had a classification accuracy of 80%. The model predicts only the direction of exchange rate change one year in the future. It does not consider the magnitude of change nor movements in exchange rates during the year. An attempt at a three-way classification to isolate currencies which changed less than 10% gave a relatively low classification accuracy of 55%. (Author).