The Study Of Stock Price Prediction Based On BP Neural Network

On efficient market conditions, stock price reflect all information, short-term stock price random walk,and technical analysis can help investors achieve excess returns. However, a large number of empirical practice prove that the stock market is not efficient.The stock price run with a certain regularity. Stock price is actually a complex nonlinear function, so stock price have a certain predictability. Stock price fluctuations affect by many factors.These factors is also very complex. Therefore time series prediction based on statistical methods can not achieve satisfactory results. To solve this problem, with excellent analog performance of the artificial neural network was introduced for financial forecasting. In theory, it can be simulated within a certain range of accuracy of any nonlinear continuous function. Artificial neural network's novelty is that without understanding the cause of data, through the given training samples for machine training, we can establish the output and input variables of the function, and simulate the nonlinear process modeling. Using neural network to forecast stock price is very appropriate.In this paper, based on BP neural network theroy which is widely used to predict the practice we design optimization method for its shortcomings. Combined with price-sensitive operation of the various technical factors, we design of network structure and parameters continuing to experiment. Through empirical testing, we conclude:1. Prove the predictability of stock price of our stock market,and the operation of the Shanghai index is not haphazard. Our stock market does not meet the weak effective.2. Volume factor driving the use of gradient descent method and the parameters adaptive BP algorithm can optimize the performance of BP neural network. The model reflected in the share price forecast a good performance, so that the improved algorithm is feasible in practice.3. BP neural network can be in better accuracy of forecasting stock prices. Compared with the traditional linear prediction model, in the stock price prediction, neural network model has marked the superiority.This structure is:The first chapter is an introduction, describeing the background and significance of topics, research status, the research method, ideas, articles, structures and innovations. ChapterⅡof neural network theory, which is introduced the basic concepts of neural networks, feature. BP algorithm also described the mathematical basis and we also propose optimizations. ChapterⅢas a model for the design, feasibility analysis of the model first, and then design the network structure and initial parameters. An empirical analysis of ChapterⅣ, after pretreatment of the raw data to train the neural network model to determine the model parameters obtained empirical results. Finally the conclusion summarizes model predictions on China's A shares capacity, draw conclusions and points out the shortcomings and future research directions.