Forecasting foreign exchange rates using artificial neural networks: a trader's approach

摘要

This study investigates the use of two different types of the Artificial Neural Networks (ANNs), Feed-Forward (FF) Neural Network and Nonlinear Autoregressive with Exogenous Input (NARX) neural network, in forecasting the exchange rate of the US dollar against the three major currencies: the Euro, the Pound and the Yen. Although the ANNs technique is not very common in economic discipline, the results are expected to be more accurate in terms of market timing ability and sign prediction than those of the standard econometric techniques such as ARMA. ANNs are, in fact, capable of dealing with high-frequency data as well as the nonlinearities in exchange rate movements. Our results support the notion that ANNs is an effective method in forecasting the exchange rates. The NARX networks output shows a significant market timing ability. Both FF and NARX proved to forecast at a higher accuracy (sign prediction) than random walk and ARMA models.