Reasearch On Quantitative Stock Selection Model Based On Convolutional Neural Network

In recent years,with the advent of artificial intelligence such as AlphaGo,deep learning has gradually become popular and leaded a new learning trend.At the same time,China's quantitative investment industry is also rapidly developing,but compared with foreign quantitative investment industry,there is still a huge gap.In essence,quantitative investment is to predict future earnings by modeling financial data,which is similar to deep learning.Therefore,it is feasible to combine the two together.Throug applying deep learning to quantitative investment,on the one hand,it provides a new research idea and modeling method for quantitative investment,on the other hand,it helps investors to more accurately predict stock prices,reduce investment risks and obtain investment income.This paper mainly studies the application of convolutional neural networks in quantitative stock selection.The research data are the constituents of CSI 300 Index from January 1,2008 to January 1,2018.Firstly,this paper will add a technical indicator dimension.Convert onedimensional time series into two-dimensional matrix data for training and prediction of convolutional neural network models.Secondly,in order to use data more effectively,this paper divides the data into five training sets and five test sets by window scrolling.Thirdly,the paper studies the influence of the convolution kernel size,the number of convolution kernels,the number of fully connected neurons,the dropout probability,and the optimizer on the prediction results of the convolutional neural network model and optimizes the parameters of the model.After that,this paper compares the prediction results of convolutional neural network model with the results of logistic regression,BP neural network and LSTM neural network.Finally,this paper constructs the corresponding CSI 300 stock selection strategy based on convolutional neural network for research and analysis and compares it with the same CSI 300 stock selection strategy based on logistic regression,BP neural network and LSTM.By comparing the model accuracy of convolutional neural network with logistic regression,BP neural network and LSTM neural network,we find that the prediction of convolutional neural network is better than the other three models.In addition,this paper builds a convolutional neural network based on convolutional neural network.The strategy gained 186.6% of total revenue and 125.51% of excess returns during the 5-year backtest period.The annualized return is as high as 23.44% and the Sharpe ratio is 1.47.In order to better evaluate the effectiveness of the model's stock selection,this paper constructs a corresponding longshort hedging strategy.After long and short hedging,the strategy achieves a total return of 160.55%,an average annualized return of 21.11%,and a maximum drawdown of-2.39%.The average sharp rate is as high as 5.18.Finally,compared with the performance of logistic regression,BP neural network and LSTM neural network in quantitative stock selection,the results show that the total return,excess return and Sharpe ratio of the convolutional neural network strategy are higher than the other three models,and the convolutional neural network is superior to the logistic regression model in each year.Therefore,convolutional neural networks is an effective quantitative stock selection model.