Research On Alpha Quantitative Trading Strategy Based On Improved XGBoost

In the field of quantitative investment,machine learning methods have provided new ideas for the construction of investment strategies.In this context,this paper proposes an improved XGBoost model.Based on the improved model,an Alpha strategy is constructed in the Chinese A-share market and applied to the quantitative investment field.The improved model can be represented as RF-BOA-XGB model,which consists of three parts: using the random forest model to screen important factors,using Bayesian optimization algorithms to select the optimal hyperparameters of the XGBoost model,and using the XGBoost model to predict the monthly stock return of Chinese A shares.First,this paper uses a total of 164 factors of financial factors,market factors,and risk factors as the initial factor library.After using the random forest model to select important factors,the Bayesian optimization algorithm is used to analyze the XGBoost model.Five hyperparameters are tuned,and the optimal hyperparameters are substituted into the XGBoost model,and then the XGBoost model is used to predict the monthly return of A-share stocks on a rolling basis.Secondly,according to the prediction results of the RF-BOA-XGB model,the quintile stock portfolio is divided according to the monthly stock return rate,and a long-short portfolio is constructed.Finally,this paper tests the performance of the long-short portfolio for a total of 136 months from July 2008 to October 2019,and compares it with the HS300 index and other five benchmark models to evaluate the the performance of long-short portfolio and the performance of the RF-BOA-XGB model.The empirical results show that: First,through the screening of effective factors,this paper finds that market factors,risk factors and the development factor among financial factors can effectively predict the monthly stock return of Chinese A shares in the long run.Second,compared with the HS300 index,the RF-BOA-XGB long-short portfolio has obtained higher total yields,compound annualized yields,monthly average yields,and Sharpe ratios,and has a lower maximum drawdown rate.The portfolio can beat the HS300 index in the long run.Third,the RF-BOA-XGB long-short portfolio can obtain significant Alpha in the Chinese A-share market.Fourth,by comparing the RF-BOA-XGB model with other five benchmark models,it is found that the combination method can indeed improve theprediction performance of the RF-BOA-XGB model.