A data mining approach to financial time series modelling and forecasting

This paper describes one of the relatively new data mining techniques that can be used to forecast the foreign exchange time series process. The research aims to contribute to the development and application of such techniques by exposing them to difficult real-world (non-toy) data sets. The results reveal that the prediction of a Radial Basis Function Neural Network model for forecasting the daily $US/$NZ closing exchange rates is significantly better than the prediction of a traditional linear autoregressive model in both directional change and prediction of the exchange rate itself. We have also investigated the impact of the number of model inputs (model order), the number of hidden layer neurons and the size of training data set on prediction accuracy. In addition, we have explored how the three different methods for placement of Gaussian radial basis functions affect its predictive quality and singled out the best one. Copyright © 2001 John Wiley & Sons, Ltd.