Higher Order Neural Networks and their Applications to Financial Time Series Prediction

R. Ghazali (Malaysia and UK), A. Hussain, and M. Merabti (UK)

Keywords

High Order Neural Networks, Functional Link Network, PiSigma Network, Multilayer Perceptron, and time series.

Abstract

This paper investigates the ability of High Order Neural Networks as forecasting tools to predict the future trends of financial time series data. Two types of High Order Neural Networks are used which are the Functional Link and Pi-Sigma neural networks. The models were tested on two different financial data; the IBM common stock closing price and the United States 10-year government bond. The performance of the High Order Neural Networks is benchmarked against the performance of Multilayer Perceptron. The simulation results indicated that High Order Neural Networks performed better than the Multilayer Perceptron when compared using the annualized return and using small number of free parameters.

Important Links:



Go Back