How to Beat the Random Walk

Out-of-sample forecasting accuracy is a frequently used criterion for evaluating models of exchange rate determination. This paper shows that both UIP and PPP produce better exchange rate forecasts at the ten-year horizon than a random walk without drift. There are two novelties relative to previous studies. First, the effects of extending the horizons beyond four years have not been investigated. This is relevant because the influence of fundamental variables has been shown to increase with the forecasting horizon, and it may take considerably more than four years to reach the long run equilibrium in the case of exchange rates. Second, the exchange rate forecasts implied by uncovered interest parity have been neglected in this literature. UIP is typically rejected in empirical tests using data on short-term interest rates. However, long-term interest rates appear to be a quantitatively important determinant of nominal exchange rate changes.