Beyond ESG Ratings: Informational Value of Specific ESG Factors for Corporate Carbon Management

As ESG disclosure and climate governance requirements become increasingly institutionalized, understanding how firms respond to external sustainability pressures through specific governance mechanisms and how these responses relate to carbon performance has become an important research issue. Moving beyond aggregate ESG ratings, this study examines the informational value of specific ESG-related factors for corporate carbon management, including energy management systems, innovation capability, human capital, market valuation, and board governance. Using data on Taiwanese listed and over-the-counter companies from the ESG database of the Financial Supervisory Commission, the Taiwan Economic Journal, and the Leadership ISO Survey, this study employs a time-lagged design linking 2023 firm characteristics to 2024 carbon emission intensity. Multiple regression analysis is the primary method, with firm size, leverage, capital intensity, profitability, and firm age included as firm-level controls in extended models; exploratory data analysis (EDA) serves as a supplementary diagnostic for data distribution, nonlinearity, and variable operationalization. Cross-year, alternative dependent-variable, and supplementary analyses are used to assess the stability of the estimates. The results show that ISO 50001 certification is positively associated with subsequent carbon emission intensity, whereas the structural characteristics of R&D investment and independent director governance show more stable negative associations. Average salary, female director representation, and Tobin’s Q yield inconsistent results. Proportion-based board measures outperform director-count and threshold-based measures, but no clear critical-mass threshold is supported. Results also vary under alternative carbon-performance measures, suggesting that carbon intensity and absolute emissions capture different dimensions of environmental performance. These findings indicate conditional statistical associations rather than causal effects.