A Comparison between Among Econometric, Time Series and Neural Network Models for Exchange Rate Prediction

This research examines and analyses the use of Artificial Neural Network (ANN) models in foreign exchange (FX) forecasting and trading models. Also, the performance of five linear models of exchange rate of Iranian Rial/US dollar is compared with that of ANNs. The linear models are Box-Jenkins methods (auto regressive integrated moving average), two forms of random walk process (with a variable drift and without it) and three different specifications based on the purchasing power parity (PPP) hypothesis. The aim is to examine whether potentially highly nonlinear neural network models outperform traditional methods? We employ ANNs to study the nonlinear predictability of exchange rate for the Iranian Rial /US dollar. The comparative exercise has been conducted both in-simple and out-of-simple, at the 1-, 6- and 12-month forecast horizons. In general, the results confirm the difficulty in forecasting exchange rats, and reaffirm those obtained in previous literature which shows that the performance of econometric models of the exchange rates is inferior to that of a random walk (RW). In the direct comparison between structure econometric, time series and artificial neural network, the experiment with monthly data indicates that there ANNs clearly improves forecasting the exchange rate.