On forecasting exchange rates using neural networks

The paper considers the modelling, description and forecasting of four daily exchange rate returns relative to the Dutch guilder using artificial neural network models (ANNs). Based on simulations it is argued (i) that neglected GARCH does not lead to spuriously successful ANNs and (ii) that if there is some form of nonlinearity other than GARCH, ANNs will exploit this for improved forecasting. For the sample data it is found that ANNs do not yield favourable in-sample fits or forecasting performance. These results are interpreted as indicating that the nonlinearity often found in exchange rates is most likely due to GARCH and therefore ANNs are recommended as a diagnostic for mean nonlinearity.