Research On Reverse Investment Strategy Based On Boosting Model

This paper explores the combination of Boosting model algorithm of machine learning and reverse investment strategy.On the one hand,the paper hopes to retain the idea of reverse investment that aims to look for stocks with low market attention or unexplored but good fundamentals.On the other hand,intelligent trading using machine learning can not only analyze data quickly and efficiently,but also make Objective and accurate judgment to eliminate the influence of investors' irrational psychology,forming a more effective and feasible strategy.The stock pool selected in this paper is all A stock market since 2007,excluding stocks that have been ST or suspended within the three-month period;according to the main measurement criteria of the reverse investment strategy,various types of data information are screened and synthesized to the corresponding index system,and then the process of feature extraction,data preprocessing and important parameter adjustment of training period are carried out.After that,the Boosting model is learned by internal and external rolling training,stratified test analysis and comparsion and optimization of the model results and backtesting combinations,including internal subdivision Boosting model(GBDT,XGBoost,Adaboost)and other machine learning models(linear regression,support vector machine,random forest and naive Bayes)comparison.The stock selection portfolio is compared by whether the industry is neutral,the frequency of the reconciliation(monthly and quarterly)and whether the model stop loss optimization is added.Based on the above research process,this paper finds that the XGBoost classification accuracy rate slightly equals to Adaboost and GBDT,but the running speed and simulation backtesting effect is much better than the other two boosting models.Secondly,compared with other comparison algorithms,the XGBoost model performs better than linear regression,support vector machine,random forest and naive Bayes in classification accuracy and yield index,while the maximum retracement rate is bigger than linear regression.As for the operation speed,other models' transportation time is generally 2 to 8 times that of XGBoost.At the same time,according to whether the stock selection is neutral or not,it is found that with the number of stocks increasing,the yield index gradually declines and the maximum retracement rate increases.The best performance is that each industry is selected for 2 stocks,or The A-share market chooses 20 stocks.Compared with the quarterly adjustment portfolio,the monthly is better in indicators such as yield,Sharpe ratio and information ratio except the maximum retracement rate.After adding the stop loss condition to the model,the excess annualized rate of return increases significantly,reaching 28.3%.The maximum retracement rate also decreases significantly,and the optimization effect is obvious.On the whole,the reverse investment strategy stock selection scheme based on the Boosting model has obvious advantages in computing speed and excess return compared with other machine learning algorithms,which has certain feasibility and practical value in Chinese stock market.