A number of studies have contended that it is challenging to explain exchange rate movement with macroeconomic fundamentals. A naive model such as a random walk forecasts exchange rate movements more reliably than existing structural models. This paper confirms that it is possible to improve the forecast of structural exchange rate models, by explicitly accounting for parameter instability when estimating these models. Making use of the Kalman filter as an estimation method that accounts for time-varying coefficients in the presence of parameter instability, this paper indicates that forward exchange rates with different maturities predict the future spot exchange rates more reliably than the random walk model for the Rand exchange rates.