Application Of BP Neural Network In Securities Analysis

With the development of market economy and the perfection of financial market,the relationship between stock market and real economy is becoming more and more close.The stock market not only serves as a way of listing and financing enterprises,but also provides a platform for investors to obtain income.In the process of stock investment,the fundamentals analysis and technical analysis is one of the two very important means of analysis,and had experienced a long history of development.However,with the development of artificial intelligence,especially the maturity of data mining technology,quantitative investment gradually sought after by investors,more and more investors and researchers began to study the application of data mining in investment analysis.This paper introduces the common analysis methods in the securities market and the development of data mining,then introduces the artificial neural network algorithm and deduces the formula.In this paper,the ups and downs of the Shanghai and Shenzhen 300 Index(HS300)as the object of study,select the time span for January 1,2012 to January 1,2017,six years of 1213 trading days to close the original data to BP neural network model Empirical analysis method,by adjusting the number of hidden layer nodes,learning rate and neural network hidden layer layers,and then compared with the support vector machine method to get the optimal model.At the same time,this paper also analyzes the application of support vector machine(SVM)in forecasting index revenue.By comparing the radial basis function,optimizing the penalty coefficient and gamma,we get the optimal result and compare with the results of BP neural network algorithm.The results show that data mining technology,especially BP neural network can be applied to the securities market,and achieved good prediction effect.By comparing BP neural network and support vector machine,the former has achieved better prediction accuracy,but the stability is poor.Because the BP neural network is the black box model,the training rate and the hidden layer number do not have obvious correlation with the actual result,but the network layer has a positive correlation with the prediction accuracy rate.