Multi-factor Quantitative Stock Selection Strategy Based On LightGBM

With the rapid development of artificial intelligence,the application of AI in commerce,engineering,military,finance and other fields has become more and more extensive.And one of the typical application scenarios is the field of quantitative investment.Quantitative investment is different from traditional qualitative investment.It is the process of using computer programmed instructions and adopting certain mathematical models to implement investment strategies.It avoids artificial cognitive biases and subjective assumptions.The investment market has the characteristics of numerous stocks and complex index factors.It is difficult to analyze by manpower.It happens that machine learning,deep learning and other knowledge are gradually mature,and the computing power of computers has been greatly improved.With the help of AI algorithms and the powerful computing power of computers,help investors process massive amounts of data and mine data characteristics to achieve better investment results,which is also a new trend in future development.This paper focuses on the current hot quantitative investment environment in the market,and constructs a set of multi-factor quantitative stock selection strategies based on machine learning algorithms.This paper selects the top 50 stocks in the CSI 300 by market capitalization as the research object,and obtains their price data for a total of 5 years from December 2014 to January 2020 for research.In terms of selection factors,this paper consulted a large number of literature and securities firm research reports,and referred to the "China A-share Market Quantitative Factors White Paper" published by Wudaokou School of Finance,Tsinghua University.There are 9 types of high-quality factors including finance,technology,basics,emotions,growth,risk,share,momentum,and style factors,with the number reaching 260,constructing the factor pool of this paper.Next,this paper builds a predictive model based on the newly proposed machine learning algorithm LightGBM in recent years,and combines a data processing algorithm to try to optimize LightGBM.At the same time,this paper also constructs traditional machine learning algorithm models such as support vector machines and random forests to be used as control experiments.Experimental results show that compared with support vector machines and random forests,the LightGBM algorithm has improved prediction accuracy and AUC value,and the training speed has been greatly improved.We constructed investment strategies based on the predicted results of the model,and verified them through back test verification,and the final selected stocks achieved good results.The strategic returns were 32.62%,57.41%,39.34%,38.85%,and 42.43%.Compared with the CSI 300 benchmark,they obtained 8.12%,41.73%,13.52%,44.78%,and 21.87% of excess returns,and the maximum withdrawal is low.It proves the feasibility of the strategy in this paper,and also provides a new way of thinking for the existing quantitative investment model.