This paper examines the monetary model of exchange rate determination. Using cointegration and error-correction techniques, forecasting models for DEM / USD and ITL / USD are constructed. In order to assess the forecasting properties of the constructed models, they are put to an out-of-sample test against the random walk. It is shown that when proper care is taken of the time series properties of the underlying data and short-run dynamics are incorporated in to the long-run monetary model, the estimated error-correction models can comprehensively beat the random walk over different forecasting horizons up to 1 year.
