A EMD-based Combining Forecasting Model For Stock Index

Stock market is an important part of financial market.In stock market,corporates can raise necessary funds and investors can rearrange their portfolio.It is a strong support for economics.Shanghai Stock Exchange is founded in 1990 s.It witnessed the great development of Chinese stock market during these years and its eager for effective and subjective studies for its development.That explains the importance of stock index forecast in both theoretical and practical fields.A great many kinds of forecasting methods have been proposed until today and they can be divided into mainly two kinds.One is the Statistical methods such as the Markov process,ARMA,ARIMA and GARCH.The other one is the intelligent algorithm such as ANN.However nowadays,a new method different from these two kinds is used in financial time series analysis,that is the empirical mode decomposition.It is based on Hilbert-Huang transform and is able to decompose a time series into several intrinsic mode functions,which in financial fields reflect changes in the time series of different periods.In this study,a combined forecast method based on empirical mode decomposition is proposed.It combines Hilbert transform,ARIMA,hidden Markov model,fuzzy neuron network and RBF neuron network to take advantages of each of these methods.In the first place,the stop rule of empirical mode decomposition is studied based on the properties of stock index time series.Stop rule has great influence on the stability of decomposition results.It also restricts intrinsic mode functions to ensure them meaningful in Hilbert transform.A quantitative index for these problems is proposed in this article and a comparative study is done based on this index.The result shows that different stop rules have different effects on the decomposition results.In addition,it is found that the tolerance of the stop rule in financial fields should be set to be larger than it in physical fields to ensure the stability of the results.In the next part,the intrinsic mode functions of SSE are calculated.They are divided into short-term volatility,medium-term volatility and long-term volatility.It is shown that the medium-term volatility and long-term volatility reflects the fluctuation of the Chinese stock market,which proves that empirical mode decomposition is available in analyzing Chinese stock market.In the last part of the article,two model bases are set based on the different properties of short-term volatility and mid-long-term volatility to forecast their values.Moreover,a RBF neuron network is proposed.It takes the forecasted values of each intrinsic mode function as inputs and get the final forecast result.An empirical study based on the Shanghai Stock Exchange Index shows that this model takes advantages of its component models and is more accurate than ARIMA,FNN.Its forecast is robust.This article provides researcher of financial time series a new direction.It also improved the empirical mode decomposition in the financial fields.Investors can make subjective decisions based on this study.