The Reliability of ESG Disclosures as Indicators of Environmental Performance - A Machine Learning Approach to the Analysis of ESG Reports and Carbon Intensity in the Steel Industry
Abstract
This thesis examines the relationship between ESG reporting and actual environmental performance in the steel industry, focusing on Scope 1 carbon intensity. Using natural language processing, we developed an Environmental Index to quantify the content of ESG reports and performed sentiment analysis to evaluate their tone. The Environmental Index did not significantly correlate with carbon intensity. This lack of a significant correlation may stem from three possible factors: limitations in the methodology used to construct the index, inaccuracies or misrepresentations in corporate ESG disclosures, or errors in third-party carbon intensity data. However, sentiment analysis revealed that a positive tone in reports was associated with lower carbon intensity, particularly in annual and integrated reports.
The study also finds that companies publishing standalone ESG reports often have higher carbon intensity. Similarly, firms emphasizing renewable energy were also found to have higher carbon intensity. This suggests that some companies may strategically use ESG reporting to influence stakeholder perceptions. These findings suggest a gap between ESG disclosures and actual emissions, raising concerns about the credibility of such reports.