Predicting Trend of High Frequency CSI 300 Index: Based on Empirical Mode Decomposition and BP Neural Network

Abstract

As an indispensable part of research on the financial market, the prediction of stock market indexes helps investors to make effective investment strategies. The instability of stock sequences, the asymmetry of market information, and the herd mentality make stock indexes more difficult to be predicted. To improve the prediction accuracy of existing models, this article aims to use the hybrid algorithm to study the 10-minute data of CSI 300. Three statistical methods: Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), and Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), are implemented to decompose the stock index sequence and obtain several intrinsic mode function (IMF) components. Back Propagation (BP) neural network is established to predict the IMFs respectively, and the final stock index prediction is obtained through the direct sum of the IMFs regression results. Finally, we conclude that the BP neural network hybrid model based on CEEMDAN has higher prediction accuracy.