Research On Multi-factor Quantitative Stock Selection Scheme Based On Voting Ensemble Learning Algorithm

Ensemble learning is now a very popular machine learning method.At present,the integrated method has achieved good rankings in many well-known machine learning competitions.The ensemble learning algorithm completes the learning task by constructing and combining multiple learners.It is sometimes called a multi-classifier system or committee-based learning.It often has significant advantages over the single learner.The integrated learning algorithm has achieved remarkable results in predicting financial market trends,processing unstructured text information,identifying financial risks,and improving investment plans.The voting integrated learning algorithm is a type of ensemble learning algorithm,and it has obvious advantages over the single algorithm in predicting stock indexes.However,the research literature on important issues such as how to use the voting integrated learning algorithm to design stock selection schemes,how to efficiently use the high-dimensional features of a large amount of data for quantitative stock selection,and how to build the better voting integrated learning algorithm based on a combination of better base learners is scarce.This paper mainly studies the multi-factor quantitative stock selection investment plan based on the voting integrated learning,constructs the better voting integrated learning algorithm based on a combination of better base learners,and hopes to obtain a return rate that exceeds the market.First of all,this article takes the constituent stocks of the Shanghai and Shenzhen 300 Index as the scope of stock selection and constructs a pool of 34 factors that affect stock returns.The stock which monthly increases greater than the HS300 index is marked as "1",otherwise it is marked as "0".Secondly,this paper uses the random forest algorithm to filter out 22 more important feature factors,and then compares and analyzes the multi-factor quantitative stock selection schemes based on six single algorithms.This article then filters out four relatively excellent single algorithms,once again use the four algorithms to constructs 11 types of Hard-Voting and Soft-Voting integration algorithms which are based on base learner combinations respectively and constructs 11 types of Bagging algorithms based on Naive Bayes.Finally,these algorithms are compared and analyzed to select the optimal multi-factor quantitative stock selection program.The research conclusions of this paper show that the best multi-factor quantitative stock selection scheme constructed in this paper is that the scheme based on the Hard-Voting algorithm which combines Naive Bayes and Support Vector Machines.Its stock selection backtest performance is better than that of six single algorithms,Bagging fusion algorithm based on Naive Bayes and the voting integration learning algorithms based on other base learner combinations,and at the same time far better than the Shanghai and Shenzhen 300 Index.In terms of return,its cumulative total return for 13 months of stock selection backtesting is as high as 123.5%.In terms of risk,the maximum drawdown is 3.43%.In terms of stability,it outperforms the benchmark of the CSI 300 index for 10 months in stock selection backtesting for 13 months.