Applying Ensemble Learning Algorithm In Alpha Strategy

Since the Sino-US trade war broke out on March 23,2018,China's A-share market has declined in an all-round way.As of June 22,the total market value of Shanghai and Shenzhen stock markets has shrunk by 7064.2 billion yuan,causing heavy losses to investors.How to effectively prevent the systemic risk brought by the market is of great significance to every stock market investor.Alpha strategy as one of the quantitative investment strategic models,by long stocks while short stock index futures,when the market crash,stock index futures profits can make up for the loss of stock positions.Alpha strategy is designed to pursue absolute returns that are less correlated with market fluctuations.In addition to effectively guarding against systemic risks brought about by the market,it can also prevent the occurrence of unilateral long-term capital trapped.The key to a successful Alpha strategy lies in the building of the stock selection model,which largely depends on the model's ability to grasp such factors as valuation,finance and solvency.In this paper,931 stocks of SMEs in A-share market are selected as the stock pool,and 28 common factor values in the last six years are selected as the characteristic variables.The following month's stock returns are used as the label values to construct a two-class problem.When the next month's stock returns are greater than 0,the next month's returns are marked as "1";When the next month's stock returns are less than 0,the next month's returns are marked as "0".Then choosing stocks from big to small based on the probability of the next month's stock returns that are marked as“1”.Comparing the stock portfolio with the market index(Shanghai and Shenzhen 300 index)to see if it can win the market.If it wins the market,it shows that the stock portfolio can obtain positive Alpha returns,and then from the yield index and risk index to evaluate the effect of stock selection.It is found that the stock portfolio is able to outperform the stock market,but has a larger withdrawal rate.In order to reduce the withdrawal rate,this paper comparing and analyzing the characteristics of the existing hedging tools as well as the proportion of margins paid,and find that stock index futures have a significant advantage in hedging effect.Finally,we analyze how to determine the position of stock index futures and the problems of holding stocks in a long term.On the method of stock selection,this paper applying machine learning to quantitative stock selection.Comparing the classification results of Boosting ensemble learning algorithm represented by XGBoost,Bagging ensemble learning algorithm represented by Random forest,and stacking ensemble learning algorithm based on multiple models.Finally we find that stacking ensemble learning algorithm has a certain error correction ability,which can improve the classification effect of a single model,but the operation time is longer.