A comparison of predictability of exchange rates between developing and developed economies

Abstract

In this paper, we argue that government interventions in the foreign exchange market are more prevalent for developing countries, and such interventions will cause exchange rates to disengage from their economic realities. As a result, exchange rates for developing economies will tend to be more predictable. Our results show that for in-sample predictions, models that employ monetary variables perform better for developed economies in terms of root mean square errors (RMSE) than models that use time series variables. Hence, it shows that the movements in exchange rates are more closely linked to the macroeconomic factors of these countries. However, for developing economies, time series models outperform monetary models. Overall, neural network models do not perform as well as the other models examined here in terms of RMSE or MAPE (mean absolute percentage error). Nevertheless, for out-of-sample predictions and as measured by correct sign percentage (CSP), neural network models perform better than the other models Our results show that exchange rates are not predictable for all the developed economies studied here. On the other hand, exchange rates movements are predictable for developing economies as indicated by the results of Pesaran-Timmerman test ( 1992) which are statistically sig nificant for five out of seven of the developing Asian countries