A novel hybrid model based on EMD-BPNN for forecasting US and UK stock indices

Abstract

Stock index forecasting poses an interesting challenge to the interdisciplinary intelligent finance communities. The application of new signal processing and data mining technologies such as empirical mode decomposition (EMD) and artificial neural networks (ANN) has opened new possibilities for financial time series analysis and prediction. This paper proposes a novel hybrid model for forecasting stock indices using EMD and ANN. First, EMD is used to decompose a stock index time series is into many intrinsic mode functions (IMF) at several levels, which are then selected to form a feature vector as the input to an ANN. Then, a back propagation neural network (BPNN) is trained on each level to forecast IMFs at the corresponding level. The forecasting results are then combined to form the forecasting of the index trend in the immediate future. The proposed model is compared with ARIMA, GARCH and BPNN models. The historical data test on US and UK stock indices shows the superiority of the proposed model in terms of forecasting directional symmetry.