ESG And Cross-sectional Stock Returns Of China’s A Share Stock Market

While the global economy is developing at a rapid pace,human economic activities and lifestyles have had a huge impact on the Earth’s environment.Large-scale industrialization,urbanization and agricultural production have led to environmental and climate problems such as massive carbon emissions,overuse of resources,destruction of ecosystems and pollution discharges,which have largely constrained the economic development of countries and the healthy lives of human beings.In order to deal with environmental and climate problems,countries around the world have been deepening the concept of sustainable development and putting forward carbon emission reduction programs one after another.As a responsible big country,China proposed a "dual-carbon"strategic goal in 2020,which is to "strive to peak carbon dioxide emissions by 2030,and strive to achieve carbon neutrality by 2060".The report of the twentieth Party Congress further pointed out that "respecting,adapting to and protecting nature is an inherent requirement for building a modern socialist country in an all-round way." This points out the direction for promoting the transformation of China’s economic structure and realizing sustainable development.To realize the goal of the "dual carbon" strategy and complete the transformation of the economic structure,not only does the government have to play a macro-regulatory role,but also enterprises,as the main body of economic activities,have to actively practice the concept of ESG(Environmental,Social and Corporate Governance),and give full play to the power of the market to solve the problem of sustainable development.Because of this,under the background of the "dual-carbon" strategic goal,vigorously developing ESG concepts has been the consensus of the management and China’s listed companies,which has led to a rapid growth in the scale of ESG investment.According to statistics,as of December 31,2022,the scale of China’s ESG public fund has reached 518.2 billion yuan.ESG indicators provide an objective measure of a company’s performance in terms of environmental,social,and corporate metrics,as well as a standard for companies in terms of sustainable development,which directly affects the market’s assessment of a company’s ESG performance.If ESG is to play a positive role in China’s economic transformation,it is objectively necessary to first assess the validity and authenticity of ESG indicators.If the traditional asset pricing methodology is adopted,examining the information content of ESG indicators by directly analyzing the relationship between ESG indexes and stock returns may lead to biased results due to two factors.One,ESG indicators are often linked to multiple variables.Omitting certain variables would not accurately examine the additional information provided by ESG indicators,but including a large number of variables in the model at the same time would raise the issue of dimensional catastrophe,which again would not accurately examine the information content of ESG indicators.Second,there is not only a linear but also a non-linear relationship between the ESG index and stock returns,and only linear information can be examined with the help of traditional means,which leads to the underestimation of the information content of ESG indicators.In view of this,with the help of CSI ESG index data,① this paper utilizes machine learning methods to overcome the deficiencies in existing asset pricing research and examine the information content in ESG indexes.Specifically,first,we examine the information content of the four original CSI ESG index data,and then through supervised machine learning methods to do noise reduction and purification of the four original CSI ESG indexes to derive a single ESG index,and on this basis,we examine the information content of ESG indexes after removing the noise.Second.Based on big data and various machine learning methods,we test whether the information covered by ESG indicators can be explained by existing indicators by overcoming the dimensional catastrophe problem in asset pricing research.Finally,the incremental information that can be provided by ESG indicators is identified as much as possible by eliminating the part of ESG indicators that can be explained by other pre-existing indicators and taking into account the possible nonlinear relationship based on the traditional linear relationship research.The main findings of this paper are as follows:First,by using supervised machine learning methods to denoise and refine the four original CSI ESG indices and derive a single ESG indicator,this study found:(1)There is a significant negative relationship,i.e.,ESG risk premium,between ESG indices in the A-share market of China and stock returns.(2)After controlling for multiple pricing factors,cross-sectional predictive indicators,replacing ESG indices,and conducting various robustness tests,there still exists a significant negative relationship between ESG indices and stock returns.(3)Heterogeneity analysis shows that the negative relationship between ESG indices and stock returns is more pronounced in stocks with lower arbitrage costs,lower liquidity constraints,lower information uncertainty,and lower degree of mispricing.(4)Economic mechanism analysis indicates that rational risk expectations of investors,especially the risk compensation required for bearing risks,are an important reason for the negative relationship between ESG indices and stock returns.(5)With the impact of the policy event announced by General Secretary Xi Jinping on September 22,2020,regarding the "dual carbon" goal,we found that the negative relationship between ESG indices and stock returns has been significantly enhanced after the release of the "dual carbon" goal,indicating that the "dual carbon" goal deepened investors’ understanding of ESG risks.Second,using several commonly used machine learning and deep learning methods,this study further analyzes the relationship between ESG indicators and cross-sectional stock returns.Referring to existing research(such as Qu et al.,2018;Xu et al.,2022;Wang and Shi,2023;Liu et al.,2019;Pan et al.,2016),traditional cross-sectional predictive indicators for the Chinese market are constructed,mainly covering company financial data and stock trading data,and supplemented by investor sentiment indicators constructed using the Oriental Financial Network Stock Bar,forming the "traditional indicators" dataset.At the same time,based on the traditional indicators dataset,the "traditional indicators+ESG indicators" dataset is formed by adding four ESG indicators.This study uses OLS,Ridge,LASSO,Enet,PLS,and NN neural network,these six methods,and trains models using the "traditional indicators" dataset and the "traditional indicators+ESG indicators"dataset,and then compares the performance of the two models.The findings are as follows:(1)Compared to machine learning methods,OLS method is prone to overfitting and dimensionality trap due to high-dimensional data,resulting in estimation failure.Therefore,when using the OLS method,there is no difference in predictive power between models trained on the "traditional indicators" dataset and the "traditional indicators+ESG indicators" dataset.(2)Compared to the OLS method,machine learning methods show significantly better predictive power,and the deep learning method NN neural network performs better than many machine learning methods.(3)The model trained on the"traditional indicators+ESG indicators" dataset performs better than the model trained on the "traditional indicators" dataset,and the long-short portfolio constructed based on the former has higher returns than the latter.(4)By constructing portfolios using different methods,adjusting returns using multiple factor models,and conducting tests using different time periods,the above results remain robust.Third,following the approach of Cao et al.(2022),this study uses machine learning methods to construct models based on the "traditional indicators+ESG indicators" dataset and the "traditional indicators" dataset separately.By calculating the difference(forecast divergence)between the predicted values of the "traditional indicators+ESG indicators"and the "traditional indicators," the information covered by ESG information is measured to identify the linear and nonlinear relationships between ESG information and stock returns.This study found:(1)There is a significant positive relationship between the integrated forecast divergence indicator and cross-sectional stock returns.The long-short portfolio constructed based on the integrated forecast divergence indicator can achieve a monthly market-value-weighted return of 3.83%,proving that ESG indicators contain rich information.(2)After controlling for multiple pricing factors,cross-sectional predictive indicators,replacing ESG indicators,and conducting various robustness tests,there still exists a robust positive relationship between the integrated forecast divergence indicator and cross-sectional stock returns.(3)Heterogeneity analysis shows that the positive relationship between the integrated forecast divergence indicator and cross-sectional stock returns is more pronounced in stocks with lower arbitrage costs,lower liquidity constraints,lower information uncertainty,and lower degree of mispricing.This not only aligns with the findings in Chapter 3 but also demonstrates that ESG indicators do not provide erroneous pricing signals.(4)ESG indicator importance analysis shows that excluding any ESG indicator and re-constructing the integrated forecast divergence indicator results in weaker predictive power compared to the integrated forecast divergence indicator constructed without excluding ESG indicators.Among them,excluding the Social Responsibility Index S leads to the most significant decline in the predictive power of the integrated forecast divergence indicator,followed by the Composite Index ESG,the Environmental Index E,and the Governance Index G.The above results to some extent indicate that social responsibility factors play the most important role.In summary,this study adopts a research perspective with Chinese characteristics,references international academic frontiers,uses various emerging research methods in the fields of financial asset pricing and statistics,combines traditional methods,and conducts multidimensional and in-depth discussions on the relationship between ESG and cross-sectional stock returns.Compared to existing domestic and international research,the potential innovations of this study are reflected in the following aspects:First,the innovation in research perspective.Based on the perspective of the relationship between ESG and cross-sectional stock returns,this study deeply explores the information content of ESG indicators through theoretical analysis and empirical tests.It complements existing relevant research at home and abroad and promotes the progress of ESG asset pricing research.Second,the innovation in research content.Many authoritative ESG-related studies assume that ESG indicators contain high-dimensional information beyond traditional information and construct theoretical models based on this important premise to demonstrate the possible impact of ESG on stock prices,information content,and various factors.However,due to limitations in measurement methods and other technical means,this important assumption has not received suficient empirical support.This study overcomes the limitations in existing asset pricing research by using machine learning methods and provides deep-level tests on the information contained in ESG indicators.It also combines asset pricing risk theory and behavioral theory to discuss the sources of information contained in ESG,thereby providing preliminary empirical evidence on whether ESG is conducive to the construction of China’s characteristic capital market and valuation system.Third,the innovation in research methods.This study uses machine learning methods to study the relationship between ESG and cross-sectional stock returns,overcoming technical issues such as the curse of dimensionality,low signal-to-noise ratio,overfitting,etc.,in asset pricing research.It identifies the linear and nonlinear relationships between ESG indices and stock returns,and obtains more accurate and reliable research conclusions.