| In recent years,China’s stock market has developed rapidly,the reform of the comprehensive registration system has continued to advance,and the process of institutionalization has accelerated.Under this background,the analysis and prediction of stock return has more important significance.Over the past 30 years,people have been digging deeper into pricing factors.Machine learning algorithms have unique advantages when using a large number of characteristic factors to predict the excess return of stocks.Under the multi-factor model based on machine learning algorithms,the absence of a single factor will affect the accuracy of the model’s prediction of excess return on stocks.The greater the accuracy decline,the greater the marginal contribution of the factor to the model’s predictive ability,that is,the stronger the contribution of the factor.Under different macroeconomic policy backgrounds,the contribution of this factor presents the characteristics of ebb and flow and has certain rules.This paper selects A-share listed companies on the Shanghai and Shenzhen Stock Exchanges as the research object.Based on existing research,factors are selected from the aspects of stock volume and price transactions and company characteristics,and single-factor testing is performed,and then through least squares regression and different machine learning algorithms,including four linear algorithms(partial least squares regression,principal component regression,least absolute shrinkage and selection algorithm,elastic net)and four nonlinear algorithms(decision tree,random forest,gradient boosting tree,neural network)to construct multi-factor Model,compare the accuracy of different machine learning algorithms to predict the excess return rate of stocks,and further explore the marginal contribution of different factors to the model’s prediction accuracy.From this,the machine learning algorithm and factors with better performance are selected,combined with the macroeconomic background,and the relationship between factor contributions is considered.The research conclusions of this paper are as follows: First,the nonlinear model is significantly better than the linear model in predicting the excess return rate of stocks.In the linear model,the principal component regression and elastic net perform best,and in the nonlinear model,the neural network and gradient boosting tree perform best.Secondly,the contribution of the volume price factor to the model prediction accuracy is superior to that of the fundamental factor,and the four types of factors that contribute the most are liquidity,momentum,growth,and profitability factors.Finally,in principal component regression,elastic net,and gradient boosting tree model,the factor contributions of the volume price factor and the fundamental factor have a negative correlation.In the neural network model,the momentum factor contributes more when the economy goes up,and the liquidity factor and profit factor contribute more when the economy goes down,and the growth factor contributes more when the monetary policy is tightened.Based on a variety of machine learning algorithms to predict the excess return of stocks,explore the contribution relationship of different pricing factors,will help investors understand the factor logic behind the stock market more comprehensively,and will also help regulators prescribe the right medicine in different market environments and target the performance of specific factors.Logical implementation of targeted regulatory measures. |
