Comparing exchange rate forecasting models: Accuracy versus profitability

In this paper we compare the rankings of alternative exchange rate forecasting models using two different evaluation criteria: forecast accuracy and profitability in forward market speculation. Either or both of these criteria may be useful to the practitioner depending on the forecasting application. We use both time-series and static and dynamic structural models to construct forecasts for the Canadian dollar/U.S. dollar and German mark/U.S. dollar exchange rates over the period 1976 :12–1984: 9. Our results confirm earlier findings that simple time-series models such as the random walk rank highest in forecast accuracy. The random walk also ranks high in terms of profitability for the German mark, but for the Canadian dollar the profitability rankings are quite different than the accuracy results. For both currencies we find that some models are very profitable in forward speculation, which is evidence against the speculative efficiency hypothesis but may be consistent with the existence of risk premia in foreign exchange markets.