A random walk is compared with a Markov switching regimes process in forecasting exchange rates out of sample, using quarterly data on three currencies relative to the US dollar over the period 1973:3-1997:3. The results show that the relative performance of the models varies with the length of the post-sample period suggesting that the availability of more past information may be useful in forecasting future exchange rates.

This paper contains an assessment of three variants of the monetary approach to exchange rate determination when the dynamics of the information variables are described by a Markov switching regimes process which generates non-linear forecasts. A large information set is used and the empirical results are based on monthly data on six major US dollar exchange rates over the period 1978-90. The relevant cross-equation restrictions are tested statistically and the economic significance of the models is evaluated on the basis of appropriate volatility tests. The Markov model is compared with other popular stochastic processes.exchange rates, asset markets, Markov process, non-linear forecasts,

No abstract