Research On Multi-factor Stock Selection Based On Regression Method And GBDT

In recent years,Quantitative Investment has attracted the attention of investors because of its stable investment performance.Domestic quantitative product market scale constantly expand.As a professional discipline,Quantitative Investment has gradually stepped into the horizons of securities firms and scholars,and has become the main research direction in the field of financial engineering.The multi-factor stock selection model is the most mature and extensive stock selection model used in the field of Quantitative Investment,It is also an enduring research topic of domestic securities firms and scholars.It was based on financial investment theories such as Capital Asset Pricing Model(CAPM)and Arbitrage Pricing Theory(APT),the core idea of the multi-factor stock selection is to capture relevant factors that can explain stock returns then use historical data to excavate the effective factors to construct investment strategies and strictly in accordance with the quantitative model to build a portfolio,using active portfolio management to obtain excess returns.With the development of machine learning various computer algorithms have been applied to multi-factor stock selection,multi-factor stock selection strategy has also been continuously updated and enriched.In this thesis,We take Constituent Stocks 300 as the research object,establish the a multi-factor stock selection model under the regression method and GBDT stock selection model.We using Fama-Macbeth test and correlation test to screen factors,using Forward Stepwise Regression to removal of redundant factors,then select effective factors to build multiple regression model.We use parameter optimization and factor training to establish the GBDT multi-factor model.During the backtrack period,using two models to predict the stock returns,then select the top 15 stocks to build a portfolio based on the return rate.Through the investment portfolios of the 12-period dynamic holding,comparing the combined performance of the two models.By comparing the investment portfolio of the two stock selection models,the following conclusions could be drawn:(1)Compared with the CSI 300,both models are profitable,It also verifies the feasibility and effectiveness of the two models in China’s stock market;(2)In terms of portfolio overall performance,The profit performance of the multiple regression model is better than the GBDT stock selection model,but from the perspective of win rate and risk,the GBDT stock selection model is more stable and have a high win rate,However,Multiple regression models have greater risk and volatility.