Research About The Combination Of Forecasts On Stock Index Based On Stepwise Regression Analysis And Neural Network

The market`s internal objective law often hides behind the changing stock market. This paper attempts to identify the inherent variation of the price of securities, and thus to predict it and provide decision support for investors. However, the securities system itself is so complex and the external factors are changeable. All of this makes it difficult to lead a right prediction of the prices, so the attempt to explore a more accurate stock price model will undoubtedly have a great value. Due to the instability and variability of the macro and micro economic situation,the predictive value of a single prediction model often have large errors in almost all of the forecasting model. The combined forecasting model can integrate useful pieces of information in the single model, decentralize the uncertainties of the single prediction model within itself and reduce the overall uncertainty. So,it has great theoretical and practical value to build a combined forecasting model to explore the law of the price fluctuations of securities.RBF neural network model,along with BP neural network model and GRNN network model, in this paper, are used to simulate and predict the CSI 300 stock index`s daily closing price and then collect useful information about t he three single models, build an optimal combination forecasting model to do the empirical research about stock’s price forecast. 3 parts make up this text. An exact analysis of the single prediction model and its application is showed in the first part. For the construction of combined forecasting model, it lays a basis. The second part describes the theory and methods of combination forecast, while selecting 3 kinds of neural networks, RBF BP and GRNN model, to construct optimal combination forecasting model. In the third part, an empirical analysis is conducted for stock index price forecast with the optimal combination prediction model and the three single forecasting model.The paper mainly studies the following three elements:(1) The stepwise regression analysis is applied to preprocess the input data of the models in order to filter out the most relevant index data with the stock index price as input data for each model. And then the RBF model is taken to exam the effectiveness of da ta preprocessing;(2) Based on the 3 neural network models, named RBF BP and GRNN, regarding minimization sum of the absolute relative error as objective function and making previous period’s forecast data together,we construct the combined forecasting model to empower the three individual model to predict the stock index.(3) Discuss the variable weight issues about the combination forecast. The empirical results show that: the stepwise regression analysis can indeed effectively screen the input data which is most relevant with the stock index price for the model; the optimal combination model can combine and integrate the useful information in each individual model and then use it effectively, thus improving the prediction results; in the short term forecast, compared to the optimal combination forecasting model with the same weight, the model with the variable weight can not significantly improve the prediction accuracy.