Linear and machine learning analysis of ESG performance and carbon emission reduction Pathways

In the context of global climate governance and the ‘dual carbon’ target, corporate ESG performance has become a key driver of the low-carbon transition. This paper uses traditional econometric models to empirically investigate how corporate ESG performance influences carbon emission reduction performance. Machine learning models are employed to analyze the non-linear relationship, revealing that ESG performance positively affects carbon emission reduction, partly by reducing the shareholding proportion of short-term institutional investors. The study’s robustness is assessed through a variety of methods, including the instrumental variable method. Additionally, a heterogeneity analysis was conducted, which revealed that the ESG effect is more significant for dual-hatted enterprises due to their decision-making efficiency advantage and for enterprises in the eastern region due to their resource endowment advantage. Moreover, machine learning techniques overcome the constraints of conventional linear models by utilizing non-linear regression for hypothesis testing. The CatBoost model quantifies the heterogeneous effects of ESG segmentation dimensions, thereby revealing that ESG’s social dimension exerts a predominant influence on emission reduction. The study confirms the catalyzing effect of corporate ESG performance in empowering emission reduction through financial channels, and conducts a machine learning-based feature importance analysis to highlight the significant roles of social and governance factors in emission reduction, which provides a scientific basis for the precise allocation of ESG resources by enterprises.