Research On The Application Of Multi-factor Quantitative Stock Selection Model Based On BP Neural Network

In recent years,thanks to the rapid development of big data and artificial intelligence,the progress of quantitative investment in China has been increasingly accelerated.According to statistics,in March 2020,China has 7343 listed companies in global Stock Exchange,their total market capitalization totals 105.71 million.Compared to the beginning of 2019,China adds 382 new listed companies,and increase of more than 30%year-on-year.At the same time,there are more and more investors,and quantitative investment has further attracted widespread attention from investors.The problem of how investors go about selecting stocks that will make them gain from more than 7,000 companies has led to the further progress of stock selection models.Multi-factor stock selection model is one of the most widely used stock selection models by investors and investment institutions,which make multi-factor models continue to improve.Nowadays,in the era of big data,the density of stock data is getting bigger and bigger,and the processing of big data needs a reasonable and efficient technology.Deep learning is highly dependent on data.The larger the amount of data is,the better the performance is,and the BP neural network in deep learning is able to handle big data and solve complex problems with unique advantages.This paper takes CSI 300 constituent stocks as the research subject,research using stock data in Wind Financial from December 2013 to December 2021.To address the problem of intricate stock data,this paper uses principal component analysis to decrease the dimensionality of stock factor data.In terms of effective factor selection,this paper proposes to use the random forest model for factor selection,firstly adjusting the hyperparameters of the random forest,and later selecting the final effective factor according to the feature importance.In this way,the factors of each facet are retained,and effective information in the factors is fully retained,and the selected operational factors are more beneficial to the stocks for multi-factor stock selection.To address BP neural network quantitative stock selection model has the problems of difficult to select learning rate and overfitting and high risk,this paper proposes and constructs Adam W-BP-MV neural network quantitative stock selection model,which integrates Adam W algorithm,L2 regularization,Dropout method and mean-variance model(MV)to optimize the BP neural network quantitative stock selection model.The model uses the Adam W algorithm to implement the learning rate adaptive gradient so as to continuously update the weights and thresholds,and optimizes the hypermastigote of the Adam W algorithm so that the model has a more higher prediction rate.Then the L2 regularization term and Dropout layer are added to solve the overfitting problem of BP neural network and improve the generalization ability of the model,and the optimal dropout rate is obtained by comparing the dropout rates of different dropouts.Finally,the mean-variance model(MV)is utilized to reduce the risk of the portfolio.The optimized model is trained using by selected effective factor data,and also compared with BP neural network quantitative stock selection model.The experimental results show that the Adam W-BP-MV neural network quantitative stock selection model has better prediction accuracy and the portfolios selected by the model can achieve higher returns with lower risks.It can be observed that the improvement of the BP neural network quantitative stock selection model is effective.