INTEGRATING EMD, CHAOS-BASED NEURAL NETWORK AND PSO FOR FINANCIAL TIME SERIES FORECASTING

Abstract

In capital market research, stock or index prices are notoriously difficult to predict, because of their chaotic nature. For chaotic time series, the prediction techniques of PSR (Phase Space Reconstruction) methods, which are based on attractor reconstruction, can be employed to extract the information and characteristics hidden of the dynamic system from the time series. However, the existence of noise which may mask or mimic the deterministic structure of the time series, can lead to spurious results. In this work, EMD (Empirical Mode Decomposition) is specially developed for analyzing such nonlinear and non-stationary data. Thus, the major of this study is to integrate PSR, EMD and NN techniques optimized by particle swarm optimization to attempts to increase the accuracy for the prediction of stock index. The effectiveness of the methodology was verified by experiments comparing random walk model for Nasdaq Composite Index (NASDAQ). The results show that the proposed PSR-EMD-NNPSO model provides best prediction of stock index.