Neural Network design parameters for forecasting financial time series

Abstract

Neural Networks (NN) have been used extensively by researchers and practitioners to forecast financial time series. The forecasting accuracy of NN depends on several design parameters, and fine-tuning them to suit a particular financial time series is essential for attaining lower error levels and minimizing running time. This paper presents the results of a two-level full-factorial Design of Experiment developed to investigate the significant factors that influence the performance of NN in forecasting financial time series. The factors considered in this paper are NN type, number of neurons in the hidden layer, the learning rate of LM algorithm, and the type of output layer transfer function. The methodology is applied to the Morgan Stanley Capital International Index for United Arab Emirates.