ESG는 조세회피에 영향을 미치는가? 머신러닝 기법을 이용한 ESG와 조세회피의 관계 재검토 = Does ESG Affect Tax Avoidance? A Re－examination of the Relationship between ESG and Tax Avoidance Using Machine Learning

다국어 초록 (Multilingual Abstract)

Prior studies on the relationship between corporate Environmental, Social, and Governance (ESG) activities and tax avoidance have reported conflicting results, leading to an ongoing academic debate.
This study posits that such inconsistency in findings stems not from the validity of any single theory, but from the following methodological limitations common in the existing literature：(1) conceptual differences in tax avoidance measures, (2) heterogeneity within composite ESG indices, and (3) uncontrolled endogeneity.
This study reconfirms the mixed findings reported in prior research. To explore the sources of mixed findings, the present study conducts the following stepwise analyses and subsequently presents corroborating empirical evidence. First, by replicating a multivariate regression used in the previous studies, we confirm that results significantly vary depending on the choice of tax－avoidance measure (ETR vs. BTD), the inclusion of control variables, and heterogeneity across rating agencies. Second, using a random forest analysis, we empirically demonstrate that ESG ratings are endogenous variables that are, to a considerable extent, predictable from financial characteristics such as firm size and financial soundness. Finally, by applying a double machine learning (DML) approach to address endogeneity, we estimate the potential causal effects of each ESG component on tax avoidance. The results of our DML analysis show that, once endogeneity is controlled for, most of the relationships observed in the conventional regression analysis lose their statistical significance. Our results imply that the associations reported in prior studies between ESG and tax avoidance may not reflect true causal relationships, but rather spurious relationships driven by common underlying factors such as firm size and financial soundness. The findings of this study highlight the risks of relying on averages to draw a simplistic, one dimensional interpretation of the relationship between ESG and tax avoidance. Because various factors, including firm size, corporate governance, and shareholder monitoring simultaneously influence both ESG and tax avoidance, it is difficult to definitely establish their relationship. This underscores the need to move beyond analyses based on composite ESG scores and to carefully account for the distinct motivations of each ESG component as well as the associated endogeneity issues. This study suggests that investors, policymakers, and tax authorities should adopt a more cautious approach when using ESG information. In addition, it highlights the importance of future research to further elucidate the complex relationship between ESG and tax avoidance.