Font Size: a A A

Multi-factor Stock Selection Trading Strategy Design Based On Time-varying Weighted LightGBM

Posted on:2021-03-05Degree:MasterType:Thesis
Country:ChinaCandidate:S Q YuFull Text:PDF
GTID:2430330626454328Subject:Financial
Abstract/Summary:PDF Full Text Request
The computer performance has improved significantly with the development of science and technology in recent years,which has brought a rapid development of financial quantitative investment and machine learning,and a large number of transaction strategies emerging.Trading strategy is based on quantitative investment,which is a kind of active investment management combined with quantitative method and computer program.Mainly including quantitative stock selection,quantitative timing,stock index futures arbitrage,commodity futures arbitrage,statistical arbitrage and so on,quantitative investment is favored by investors for its higher efficiency and lower error rate.Among them,quantitative stock selection is the most popular,thus how to design a trading strategy that can obtain excess return rate is the focus of current investors.From the perspective of factors,this paper aims to build a stock selection model,design trading strategies,and ultimately obtain excess earnings through the construction of effective factors.In the empirical process,this paper constructs a multi factor stock selection model based on time-varying weighted LightGBM,selects the factor of the first trading day of each month from January 2011 to December 2019 as the data samples.Firstly,this paper constructs the high-frequency factor index:RVol_t?RSkew_t?RKurt_t?from the intra high-frequency data of each stock.Then we get the effective high frequency factorRSkew_t?after a sequence of analysis,and screen the CSI 500 based on the size of this factor,get a stock pool of each period,and then establish a low frequency factor library of the whole market and verify the effectiveness of the factor.The weight of each period is determined according to the correlation between factors and dependent variables,and each period changes dynamically.Finally,the LightGBM model is used for training and prediction.This paper carries out the total back test and realize a total strategic return of 118.41%,the annual return of 14.26%.Besides,the sharp ratio of the strategy is 0.4275,over 80%of the months outperform the CSI 300index,and the net value of the strategy is 2.1841,far exceeds CSI 300,which shows that the multi-factor stock selection strategy in this paper can obtain excess returns.
Keywords/Search Tags:Factor screening, Time-varying weighted, Trading strategy, LightGBM model
PDF Full Text Request
Related items