Empirical Reasearch On ESG Multi-Factor Model Based On XGBoost And Other Boosting Algorithms

The predecessor of ESG(environmental,social,corporate governance)investment is socially responsible investment,which came from Europe and the United States,started late in our country.It means considering its impact on the environment and society except its financial situation to make the decision whether to invest or not.Since the goal of achieving carbon peak and carbon neutrality was proposed at the two sessions in 2021,ESG has also attracted much attention because it conforms to the national strategic layout.In the recent years of the COVID19 epidemic,many listed companies have experienced large fluctuations in financial revenue and other poor conditions.Traditional multi-factor models mostly only consider financial information,including traditional fundamental indicators and technical indicators,and have limited predictive ability during the epidemic period,while ESG rating can reflect a company’s sustainable development ability.Existing studies have confirmed that ESG investment can achieve long-term and stable excess returns.A survey shows that 90%of senior investors pay more attention to the performance of enterprises in ESG in terms of investment decisions and strategies since the epidemic.Therefore,this paper adds ESG-like factors into the traditional multi-factor model to build a portfolio,comparing the impact of quantitative strategy before and after adding ESG-like factors on investment returns,to provide a new investment strategy.With the gradual popularization of machine learning algorithm,scholars at home and abroad have widely applied machine learning algorithm to multi-factor stock selection model and obtained better returns.However,there are few researches that consider ESG factor into multi-factor model or apply machine learning algorithm.The main reason is that before 2020,the information disclosure degree of stock ESG is not high and the data of stock ESG is insufficient.However,machine learning algorithm often needs a large amount of data to have a more accurate effect.With the introduction of dual-carbon targets,the introduction of policies urging listed companies to publish ESG information,and the impact of the epidemic on the stock market,the ESG information of listed companies released by Wind,Huazheng and Shangdao Ronglv in recent two years has a high coverage rate for all A-shares,covering more than 4,000 A-shares with a large amount of data.However,XGBoost,LightGBM and other Boosting algorithms,which are relatively efficient ensemble learning models at present,have few studies on ESG multi-factor models applied to the domestic market,so this paper will conduct an empirical study based on these algorithms and the latest ESG data with the relevant data of all A-shares in China’s stock market.The research content and innovation of this paper are as follows:1.As a leading financial data service enterprise in China,Wind’s own ESG rating covers all A-shares for the first time in 2021,which has been highly concerned and recognized by various institutions such as academia,financial institutions and regulatory agencies.Based on the ESG data of Wind,Huazheng,and Shangdao Ronglv,this paper conducted effectiveness analysis,comparing the different influences of ESG scores of the three institutions on strategy returns.2.This paper explores the relevance of ESG scores provided by 3 mainstream ESG rating agencies in China,Wind,HuaZheng,and Shangdao Ronglv,adopting the weight method of CRITIC to make full use of the data of 3 agencies to construct Multi-ESG scores for the first time,and add them to the strategy for research.3.In order to fully explore the hidden information behind the ESG scores of different institutions and analyze the differences between the ESG data of different institutions,an ESG difference factor based on the differences and standard deviations of any two institutions was innovatively constructed.Its effectiveness was analyzed and added to the stock selection strategy to verify the effectiveness of the factor.4.There are few empirical studies combining machine learning algorithms with ESG multi-factor model.This paper,based on Boosting algorithms such as XGBoost,will conduct an empirical study to explore the strategic returns of ESG multi-factor model under different machine learning algorithms and get the conclusion that XGBoost has the strongest stock picking ability when adding ESG-like factors.