ESG-Driven Corporate Clustering and Stock Market Efficiency

We test the return predictability of environmental, social, and governance (ESG) scores by adopting the k-means clustering algorithm in ranking portfolio construction. The performance of the ESG score–based long–short portfolios indicates that, within the S&P 500 universe, firms with lower ESG scores outperform those with higher ESG scores. Moreover, the use of the machine learning–based clustering approach enhances the performance of these zero-cost portfolios compared with the traditional ranking method that relies on simply ordered and equally sized ranking buckets. Factor analysis fur ther suppor ts the robustness of the ESG-based portfolio outperformance, even after controlling for risk factor exposures. In addition, the factor analysis suggests the presence of potential underlying drivers contributing to this outperformance.