Research And Application Of Time Series Analysis Based On Neural Network In Stock Investment

With the development of market economy also the standardization and regularization ofChina stock market, the stock market occupys an increasingly important position in thefinancial investments, it becomes an integral part. So if we can accurately describe pricemovements and control stock market reasonable, not only have great influence on our dailylife,but also provide valuable reference to the sustainedable development of China’seconomy.Traditional stock forecasting methods cannot fit and analysis highly nonlinear,multi-factors of stock market well, there are problems such as the p rediction accuracy islow,the training speed is slow etc.Artificial neural network with its good linear approximation,adaptive and self-organizing characteristics has been widely used in the field of finance. Inthis paper, we establish multiple regression neural network model and principal componentneural network model, and use the models to test the prediction effect of BaoGang coposite.1）Multivariate regression neural network model:Multivariate linear regression is aprocess that use linear realationship to establish the regression model between the explainedvariables and several explanatory variables. In this paper, it presents a forecasting modelcombining linear regression and neural network, put the predicted results of multiple linearregression method as a neuron’s input layer of a Elman neural network, undergoing thelearning and training ability of neural network got the results.It shows that compared with thesingle neural network model,the accuracy and predictive speeds of combined model hasgreatly improved and be better suited to handle complex data information system.2） Principal component neural network model:Principal component analysis is adimensionality reduction method, it is a process that using linear transform multiple variablesinto a few principal components.Single input vector of Elman neural network is too complex,there exists a correlation among the datas.In order to reduce the redundancy of the data, weuse principal component analysis method to analyze the correlation between stock index andform a new set of training samples,which used as the training data of Elman neural network,this action not only reduces the input sizes of neural network model,but also simplifies thestructure of network and improves the learning rate.To test the model predictions results, establishing the same structure of Elman neural network and BP neural network model, while eliminating correlation of input factors andconstructe Elman neural network model based on principal component analysis. Usingsoftwares such as MATLAB and Eviews to predicted the closing price of BaoGang steelshares from April12,2013to March26,2014, and compares with single BP network modeland Elman network predictions.The simulation results show that Elman neural network modelbased on principal component analysis with higher accuracy, faster network speeds,can bebetter predict the movements of the stock price index.