Quantitative Trading Strategy Design Based On XGBoost Stock Selection Model

With the rapid development of data science and information technology,all kinds of machine learning and artificial intelligence algorithms are more and more applied in the financial field,and quantitative investment comes into being.Quantitative investment started earlier in the United States and other foreign markets,so its development is relatively mature.In contrast,quantitative trading in our country started late and accounted for a relatively small proportion.It still has a lot of room for development.Therefore,the research on the application of quantitative investment in our A-share market has very strong theoretical significance and realistic value.This paper constructs an integrated model based on XGBoost classification model and XGBoost regression model,and uses the daily frequency data of Chinese A-share allmarket stocks for model training and stock selection,hoping to design a quantitative trading strategy to obtain excess returns stably.First of all,this paper selects initial quantitative factors by referring to existing research results,and designs a characteristic factor--resistance to stock price changes according to its own experience.Then,according to the daily frequency data of all A-shares from January 1,2011 to February28,2023,we calculated the factor value from January 1,2022 to February 28,2023,and carried out data cleaning.Then,we use the factor data to adjust and integrate the XGBoost classification model and the XGBoost regression model.In this process,we found that using XGBoost regression model for secondary screening of classification results of XGBoost classification model can effectively improve the accuracy of stock selection model.Subsequently,we used machine learning’s own factor evaluation index,IC information coefficient and IR information ratio to analyze the factors,which not only verified that the characteristic index built in this paper--60-day stock price change resistance has a good performance,but also found that the removal of the common factors of the tail ranking of the three factor analysis methods can effectively improve the stock selection ability of the model.Then,through the analysis of the length of training set,the training target and the length of test set,it is found that when the length of training set is 120 trading days,the training target is 20-day yield and the length of test set is 30 trading days,the stock selection effect of the model is the best.Finally,this paper uses the data from January 1,2012 to December 31,2017 to backtest the model to obtain the most preferred stock strategy,and uses the data from January 1,2018 to February 28,2023 to simulate the trading to verify the effectiveness of the strategy.The results show that using the integrated model based on XGBoost classification model and XGBoost regression model to select stocks,allocate stocks according to the maximum Sharpe ratio portfolio,and adjust positions once every 20 trading days,can obtain an annualized return of 25.09%,a Sharpe ratio of 0.8 and an annualized volatility of 0.264.In the process of constructing quantitative trading strategies,this paper designs a characteristic factor with excellent performance--resistance to stock price fluctuations.As more and more quantitative factors appear "congestion",this provides a new choice and idea for everyone.In addition,by integrating the XGBoost classification model with the XGBoost regression model,this paper obtains better backtest returns than a single model,which also provides different ideas for quantitative investors on model integration.