A MULTICOMPONENT NONLINEAR PREDICTION SYSTEM FOR THE S-AND-P-500 INDEX

The proposed stock market prediction system is comprised of two prepro cessing components, two specialized neural networks, and a decision ru le base. First, the preprocessing components determine the most releva nt features for stock market prediction, remove the noise, and separat e the remaining patterns into two disjoint sets. Next, the two neural networks predict the market's rate of return, with one network trained to recognize positive and the other negative returns. Finally, the de cision rule base takes both return predictions and determines a buy/se ll recommendation. Daily and monthly experiments are conducted and per formance measured by computing the annual rate of return and the retur n per trade. Comparison of the results achieved by the dual neural net work system to that of the single neural network shows that the dual n eural network system gives much larger returns with fewer trades. In a ddition, dual neural network experiments with the appropriately select ed filtering and decision thresholds managed to achieve an almost twic e larger annual rate of return when compared to that of the buy and ho ld strategy over a seventy month period. However, no claims are made t hat the proposed system is better than the buy and hold strategy when considering transaction costs.