A Comparative Study Of Linear And Nonlinear Machine Learning Algorithms In Quantitative Investment

Multi-factor model is one of the most widely used methods in the field of quantitative investment,but hundreds of factors make the traditional model meet unprecedented challenges in factor selection and function form.Nowadays,many studies have shown that using the advantages of machine learning algorithm in feature selection and return prediction,multi-factor quantitative investment can accommodate more factors and achieve better investment performance.However,the feature selection mechanisms of linear and nonlinear machine learning algorithms are not the same,and their applications in revenue prediction have their own advantages and disadvantages.Therefore,this paper compares the differences between two linear machine learning algorithms,Lasso and Elastic Net,and two nonlinear machine learning algorithms,Random Forest and Gradient Boosting Decision Tree,in factor selection and investment performance.By integrating the capabilities of different models,the performance of multi-factor quantitative investment is further improved.Based on the samples of the constituent stocks of CSI 300 index from January2007 to November 2021,this paper selects eight categories of 244 factors,including quality,value and momentum,for empirical research.The empirical results show that,first of all,from the comparison of investment performance,Lasso and Elastic Network,two linear machine learning algorithms,can effectively overcome the influence of factor correlation and noise,and achieve investment performance that exceeds the traditional linear regression.Random Forest and Gradient Boosting Decision Tree,two nonlinear machine learning algorithms,can better depict the complex nonlinear relationship between factor and stock sectional return,and achieve investment performance that exceeds the linear model.Secondly,from the perspective of the differences of important factor,linear machine learning algorithms more preference is directly related to stock price movements of the momentum factor,and the nonlinear of the machine learning algorithm about the company the value of the fundamental characteristics of the higher class factor and the importance of quality class factor,its influence on yield is more complex and long,and the method of nonlinear better able to recognize this relationship;Finally,in order to integrate the ability of different models in factor selection and return prediction,the above four machine learning models were integrated using simple averaging integration and Stacking integration,and the long-short portfolio obtained monthly average returns of 2.74% and 2.67%,sharpe ratio of 1.17 and 1.24,respectively.It is much higher than the 0.83% monthly average return rate and 0.32 Sharpe ratio of CSI 300 index,and also higher than the1.90% monthly average return rate and 0.91 sharpe ratio of traditional linear regression.In general,this paper applies machine learning algorithm to multi-factor quantitative investment,compares the difference between linear and nonlinear machine learning algorithm in important factors and investment performance,and also obtains good empirical performance in the exploration of model integration.The research results of this paper have certain reference value for quantitative investment practice.