Research On The Method Of Forecasting Stock Price Based On The BP Neural Network

With the establishment of securities market in China in the early 1990s, Chinese stock market has obtained considerable development and become an indispensable part of the financial investment fields. It is undeniable that the deepening and development of stock market is accompanied by the prevailing market speculation and volatile fluctuations in share prices. Serious speculation can render stock market experience false prosperity, which misleads investors to adopt a wrong investment strategy and makes regulators take inappropriate regulatory measures, thus plunging the whole stock market into sheer danger. Therefore, if an effective forecasting method of stock price can be found, it is of important practical significance for both investors and regulators.This paper is supposed to make stock price prediction using the BP neural network technology. BP neural network is a powerful and widely-used machine learning algorithms, which is extremely adept in discovering nonlinear relationships between data and applicable to the stock price forecast problem. This paper takes into consideration all relevant factors affecting the stock price, from which transactional indicators and financial indicators are extracted, and then via Principal Component Analysis (PCA) to reduce dimensions, the principal component with reducing dimensions is regarded as the input variables of the network. In order to solve the inherent defects of BP neural network, the initial weight values and threshold values of the neural network are also optimized by the application of Mind Evolutionary Computation (MEC). By applying these steps, a "comprehensive model" is thus established. In this paper, four contrast models are also set up at the same time so as to highlight the necessity of considering comprehensive variables, dimension reduction as well as optimization.Based on stock data of military industry sector, using SPSS and MATLAB to build share price forecast model, this paper comes to a conclusion that according to the empirical analysis, the "comprehensive model" with the highest precision, in comparison with contrast model, can provide a decision-making basis for investors and regulators to a certain extent. The MSE、MAPE and the direction of accuracy of the "comprehensive model" are 1.1654,0.0562 and 84.62% respectively.