Forecasting the JSE All Share Index Using the Neural Network Techniques

Abstract

The research investigated the application of the Neural Network (NN) technique in forecasting the All Share Index (ALSI) on the JSE Securities Exchange market. The ALSI was regressed with some of the market drivers: the US Dollar/ rand exchange rate, interest rate, Brent Crude oil, Bankers Acceptance spot rate . Data from 2003 to 2005 was used to train, validate and test the NN. The results obtained were compared to those derived from the statistical forecast tools: ARIMA and Regression Analysis. The Mean Square Error between the actual and the forecast ALSI movement was below 0.5% at 95% probability, thus indicating a 95% or more efficiency of the NN. The accuracy of the NN in predicting the direction of the trend of the ALSI was as high as 89%. The NN is useful as a strategic decision making tool for investors and fund managers who track the ALSI.