Forecasting South African exchange rates: Comparing models and techniques

This article evaluates recent exchange rates models and forecasting techniques on South African exchange rate data on a monthly basis over two decades, both using ex post and ex ante forecasts. Techniques considered include stochastic time series analysis, multiple regression and time-varying parameter regression. It is shown that forecasts from the Frenkel-Bilson, Bilson, Frenkel and Frankel model specifications are more accurate than those of the Dornbusch-Frankel, Hodrick and Hoffman and Schlagenhauf models. It is further demonstrated that variable parameter techniques outperform the rest in accuracy.