Exchange rate forecasting in the West African Monetary Zone: a comparison of forecast performance of time series models

It has become an undisputable fact in economics and finance that conventional exchange rate determination models cannot outperform the random walk model in out-of-sample forecasting. We evaluate the empirical veracity of this well-known fact in the West African Monetary Zone (WAMZ). We compare the out-of-sample forecast accuracy of the random walk hypothesis vis-a-vis the Autoregressive Moving Average (ARIMA) model, Generalised Autoregressive Conditional Heteroskedastic (GARCH) based models, and Vector Autoregressive (VAR) model. The root mean square error (RMSE) is used as the measure of forecast accuracy. We find evidence to refute the body of economic literature that supports the view that forecasts from the RWM are unbeatable. We show that if a non-linear RWM is estimated, and the RMSE is used as the measure of forecast performance, the VAR model, the ARIMA model, and the GARCH(-M) model generally outperform the RWM. However, when the assumption of linearity is sustained, the RWM convincingly outperforms all other models. We show that the type of model to use to achieve forecast accuracy depends on the time horizon, and the country for which the forecast is to be made.