The Prediction Of Stock Market Based On BP Neural Network And Autoregressive Model

As the investment field of high income with high risks, stock market's operation is a complicated nonlinear system, which is easily affected by many aspects. In this field, in order to get the maximum of investment income and the minimum of investment risks, the investors continuously explore its inherent laws, and find the effective analysis methods and tools. So the research of the stock predictive methods has theoretical significance and application value.Stock market has strong randomness and nonlinearity, but artificial neural network is a nonlinear dynamic system, which has good self-adaptive ability, self-learning ability, and generalization ability, can carry out the map of nonlinear relationship between variables at arbitrary precision. That characteristic of neural network can satisfy the requirement of the stock prediction.Autoregressive model prediction is a short-term linear prediction method of high precision, which is suitable for various types of time series. In the processing of the modeling, it used a series of statistical methods to tests the model's applicability, in order to adjust the model's steps, until achieve the satisfied result, which only considers the time series'characteristics to predict. The stock market itself is affected by many unpredictable and complicated factors, such as politics and economy, which could be showed as random disturbance term in AR model.The paper first picks up three different types of stock indexes'combination, and analyzes and discusses deeply the indexes'functions and characteristics; second, builds up the BP neural network and autoregressive models, deeply analyzes the two models'theories, structures, and algorithms, and uses the MATLAB software to program; then, through picking up the real stock's data, using the build up models of BP neural network and autoregressive models separately to proceed rolling prediction, in order to predict the opening price, closing price, highest price, and lowest price; at last, for different methods and predicted stock indexes data by different indexes, separately proceeds contract horizontal and vertical analyses, in order to achieve the combination of the best predictable effect.