| The stock market is a highly complex dynamic system, and which’s trend variation isdecided by the political, economic, psychological and other factors. The traditionalquantitative forecasting methods which based on the mathematical statistics can notaccurately describe the stock market, and artificial neural network have the ability to solvenonlinear problems, network learning capabilities and the fitting ability. The artificial neuralnetwork can achieve the image of the nonlinear relationship between the variables in arbitraryprecision, and approaching the price of securities as time transform function..So far, for the different stock market, many foreign scholars have established theprediction model and good prediction method to obtain a good prediction.However, due to thehistory of the development of only ten years of securities market in China, is far from perfect,the popular and proven experience and mature foreign markets may not be suitable for China’sstock market. Back Propagation is a common forecasting methods for the stock price.Toomany parameters confusion the BP neural network computing, leading more computation andless accuracy. Based on the study at home and abroad, this article propose the BP neuralnetwork input variable selection method for the stock price forecasting. First principalcomponent analysis to reduce the dimension of the input vector; and then using the methodwhich combine Analytic Hierarchy Process and Delphi adjust the information structure of theinput vector; last compare the input vector group in a simulation experiment, and these grouphave gotten by two kinds of methods.The results show that the new primary compositionvector group has better performance for the BP neural network stock prediction. |
