Research Of Quantitative Stock Selection Strategy Based On Ensemble Learning

Both multi factor stock selection and "high transfer" stock forecast belong to the category of quantitative stock selection in essence,and occupy an important position in the field of investment.From the public information,multi factor stock selection model is mainly designed by cross-section regression and scoring methods in the domestic market,"high transfer" stock forecast mainly uses scoring method with strong subjectivity.Cross-section regression method considers the relationship between factors and stock returns as a simple linear relationship,and the scoring method mainly depends on the investors' market experience.In order to overcome the shortcomings of the above two methods,this paper regards the stock selection as a classification problem.Additionally,ensemble learning is one of the most important classification algorithms in machine learning.Consequently,this paper will build a quantitative stock selection model based on ensemble learning.Firstly,the construction of multi factor stock selection model need select the effective factors.In this paper,several different style factors are used to construct the factor library,including quality factors,growth factors,and technical indicators and other factors.Because of the serious speculation in the Chinese market,the speculative factors are constructed on the basis of the existing research,and detected by a single factor test model.Then the AdaBoost algorithm(LR AdaBoost algorithm),based on the regularized logistic regression model,is used to construct the multi factor stock selection model,and two kinds of stock selection models of M1 and M2 are proposed.Finally,these two stock selection strategies are used to construct multi factor investment model.Through the entire A share market test from March 2013 to December 2016,the result shows that the annualized return rate of M1 model and M2 model reached 33.36%,36.7%,respectively.In order to build a suitable "high transfer" stock selection model,a scoring model is constructed by using K nearest neighbors,CART algorithm and logistic regression.And compared with the random forest,AdaBoost algorithm and the LR_AdaBoost model constructed in this paper,the result shows that the scoring model and the stock selection model based on LR_AdaBoost algorithm have higher precision ratio,and the precision ratio of each method is much higher than the share of "high transfer" stock in the market.At last,the above four kinds of ensemble learning methods are used to construct "high transfer" stock selection model,then the model is applied to the investment portfolio of "high transfer" stock from 2010 to 2016,the result shows that the scoring model performs best On the whole,and the annualized return rate reached 86.86%.In this paper,the single factor validity test model and the multi factor model based on ensemble learning are constructed,and "high transfer" stock selection model are also constructed by different ensemble learning classification algorithms.In this way,ensemble learning and quantitative stock selection are fully combined.And the stock selection model is constructed with the income exceeding the HS300 index,which will provide effective reference value for investors.