Neural Network Modeling And Analysis For Stock Price Prediction

Through the analysis of predictability in the stock market,this paper proposes a method to predict the trend of the stock prices of the CITIC Securities and the China Life based on the BP neural network.In addition,refer to the analysis of the stock market price fluctuation.However,the disadvantage of the traditional BP neural network could not be ignored,such as its slow convergence speed,easy to fail into local minimum,which may cause the problems of the imperfect practice and the low prediction precision.Moreover,this paper has improved this method,which is added the momentum and optimize the algorithm which combined the variable step size.In contrast,under the same sample and parameters,this paper verify the method which is optimized could be more accurate in predication compared to the traditional BP neural network through experiment.The traditional single hidden layer BP neural network can approximate any function in theory,however,as it has not apply the proper parameter,which could may lead the huge error in predication,even fain into predication.As a result,this paper has done a lot contrast experiments under the hidden layer nodes and excitation function types.As a consequence,this paper has determined the number of nodes and the form of fuction as the incentive function of predication model.Furthermore,the value indicators represent operating performance of the listed companies,and the technical indicators summaries the basic stock market.as a result,the value indicators could predict the development trend of the target enterprise in the future.As a contract,the analysis technical indicators could predict the trend of the stock price afterwards as well.Take these two indicators could refer the stock information directly of indirectly.As a result,this paper determines to combine these two indicators as the input variable in this model experiment,which has introduced and analysis each indicator in detail.Finally,this paper integrate all the information come from the indicators,this paper choose three value indicators and seven technique indicators index as the input variable in the model experiment.Meanwhile,take seven technique indexes as the input variable prediction results,under the contrast,this two method shows that the combination index as input variables to predict the effect is much better than the other one.This paper has determined value of the index for 15 trading days through experiment as a model input variables to predict the closing price of the stock for 1 day or 30 days,and the prediction results are analyzed.