Adaptive smoothing neural networks in foreign exchange rate forecasting

Abstract

This study proposes a novel forecasting approach - an adaptive smoothing neural network (ASNN) - to predict foreign exchange rates. In this new model, adaptive smoothing techniques are used to adjust the neural network learning parameters automatically by tracking signals under dynamic varying environments. The ASNN model can make the network training process and convergence speed faster, and make network's generalization stronger than the traditional multi-layer feed-forward network (MLFN) model does. To verify the effectiveness of the proposed model, three major international currencies (British pounds, euros and Japanese yen) are chosen as the forecasting targets. Empirical analyses reveal that the proposed novel forecasting model outperforms the other comparable models. Furthermore, experimental results also show that the proposed model is an effective alternative approach for foreign exchange rate forecasting. © Springer-Verlag Berlin Heidelberg 2005.