The Industry-Stratified Stock Selection Strategy Based On Convolutional Neural Networks

Advances in artificial intelligence technology have reconstructed people's perception of data.Machine learning,one of the branch of artificial intelligence,has gradually entered the financial field.Machine learning algorithms are complicated,and different algorithms perform differently in different application scenarios.As a kind of machine learning,convolutional neural networks(CNN)are known for their unique and effective feature extraction abilities.However,there are few applied studies in the financial field,especially in the field of quantitative trading.How to apply the CNN to the field of quantitative trading,and to find its applicable scenarios have become the main issues of this article.This article firstly puts forward research questions in combination with the development process of machine learning and quantitative investment,summarizes the research significance,and briefly describes the research methods.Secondly,it sorts out the development process from the three aspects of machine learning,quantitative investment,and the combination of machine learning and quantitative investment,summarizes and compares the research status at home and abroad,and summarizes the key points and shortcomings in the existing research.Lastly,the article focuses on two main issues.The first problem studied in this paper is the performance of the CNN in‘regression' and ‘classification' tasks while clarifying the structural design.In this study,the article uses the Shanghai Composite Index data,by comparing the regression results of the CNN with the widely used BP neural network,and comparing the classification results of the CNN with support vector machines(SVM)and random forests,focusing on the feasibility of the application of product neural network in quantitative investment.The second problem studied in this paper is the industry-stratified stock selection strategy based on convolutional neural networks.In this study,the article uses the stock data of China's stock market.After data preprocessing,it firstly uses clustering analysis based on industry monthly returns to stratify by industry to construct different machine learning method application scenarios.Secondly,the IC analysis method is applied to select the factor data as the model input,and the CNN,SVM and random forests are applied to the stock selection strategy.By comparing the performance of differentmachine learning models,the CNN stock selection strategy is made.After further improvement,a stock selection model compounded by different machine learning methods is finally constructed and the applicability of applying it to quantitative investment is analyzed.This article finds that the application of the CNN to stock market forecasting has a stronger ability to predict stock price fluctuations than to predict specific stock prices.At the same time,this paper stratifies the stock market by clustering analysis,and finds that in different quantitative application scenarios,the prediction performance of different machine learning methods is different,and based on this phenomenon,a stock selection model compounded by different machine learning methods is constructed.