A Multi-factor Stock Selection Model Based On Seven Algorithms Such As Stacking Random Forest GBDT SVM Adaboost And Others

This article describes a quantitative investment strategy that uses seven kinds of algorithms such as Stacking to select the stocks that should be investedThe Stacking algorithm used in this article is an Ensemble algorithm In this algorithm,we ensemble six algorithms including Random Forest,GBDT,SVM,Adaboost,KNN,and Decision Tree.Adding Stacking,There are a total of seven algorithms.This article uses this seven algorithms to classify data.When training aigorithms,we need to input data with labels.Then we can classify sample by this model.The stock pool used in this paper is a constituent stock of the CSI 300.We select several factors of these stocks as a feature,and then use the monthly rate of return of the stock as a label.Since the rate of return is a continuous value,we mark the monthly rate of return greater than 15%as +1,that is to say,it can be bought,otherwise it is marked as-1,and we don't consider buying it.After preparing the training set data,we trained the Stacking model.Then we substitute the factor value of next month to forecast the result and select the stock that can be bought.After one month,the model is used again for prediction.For the stocks that can be bought,if it has already been held before,it will continue to hold.Otherwise,it will be bought.If it is bought but not in the stock group that can be bought we sell it.In this paper,we use the factor value of May 2017 and the income of June as the training data,then we conduct the strategy from July 2017 to January 2018.The frequency of transfer is 20 trading days.It's roughly equivalent to a monthly transfer.The results of this strategy show that the return is very good.