The Research Of The Stock Index Prediction With The RBF-GARCH Model

Since the establishment of the stock market,the research of stock market prediction has been widely concerned.However,because the stock market is influenced by many factors, there is a big difficulty for the reasonable prediction of the stock index.Now,there are a lot of methods for the stock index prediction,among which,the neural network and time series analysis are used mostly.Regarding of the characteristics of the two methods, we propose a new GARCH model with a RBF neural network for the first time.The difference between the new model and the traditional GARCH model is that we use a RBF network to replace the ARMA model in the mean equation.In this paper,firstly,we introduce and discuss the two main methods:neural network and time series analysis method.In the neural network method,through the empirical analysis,we found that the RBF neural network is superior to the BP neural network on the stock index prediction.Then,in the family of GARCH model from the time series,the empirical analysis shows that the GARCH and EGARCH model have a better fitting precision.Therefore,we combine the neural network and the family of GARCH model and propose a new model,the RBF-GARCH model,applied to the nonlinear time series prediction.The model takes the nonlinear and volatility clustering of the stock market into consideration and breaks through the traditional "black box" characteristics of neural network.In the mean equation of the model,the centers and widths of the Gaussian basis function are obtained by the k-means clustering algorithm,while the coefficient of the Gaussian basis function and the parameters in the variance equation are jointly obtained by maximum likelihood estimation.In order to improve the prediction effect,we also introduce the cuckoo optimization algorithm in this paper.Finally, after comparing with the RBF neural network, the GARCH model and the RBF-GARCH model,we find that the RBF-GARCH model optimized by the cuckoo algorithm has the best prediction effect.The empirical data of all the models in this paper uses the daily closing price series of Shanghai index sequence.