Research On Stock Forecasting Based On BP Network

Since the birth of the stock market, many investors and scholars has been attracted, to predict the trend of stock price is always the goal of what they find. However, there are a lot of factors affecting the stock prices, include not only the internal factors, but also the influence of external factors such as policy, moreover, the influence degree of various factors, such as the way is not the same, the stock prediction becomes a very difficult task. Data to predict the stock trading is a time series prediction. In the extremely complex systems of stock market, turbulence, nonlinear, high noise and other factors which decide the process of the stock prediction of complex and difficult, the traditional prediction method is very difficult to do, difficult to establish accurate mathematical model of effective. Neural network prediction method is a kind of time series. BP network is the most commonly used by neural network. Because the BP network could approach any continuous function between complex ability, and these abilities does not exist at the traditional methods, so the BP network is suitable for the prediction of time series, and do not need to make any assumptions on the time series analysis, only a BP network is used to fit the time sequence.Therefore, and the basic of the basic characteristics of the structure and algorithm of BP neural network. At last, a BP algorithm is proposed based on the improved grey difference, its application to stock market prediction, and analyzed effectively empirical. Based on GM (1,1) BP algorithm model was used to model the stock market prediction, through training, to seek the optimum network model. Using the trained model to forecast the stock, with the Shanghai composite index data as an example, with examples show that the improved BP network through learning and training, it could fit the stock data to make predictions, based on GM (1,1) prediction method of improved BP neural network grey difference can predict accurately the the next day’s closing price, compared with the pure BP, neural network and momentum adaptive learning rate adjustment algorithm prediction model error is smaller, change trend basically could agree, and satisfactory fitting results.