Association between ESG disclosure in annual reports and ESG performance

Abstract

​The ESG framework is an increasingly popular topic in accounting and finance research due to its novelty and societal importance. Worldwide environmental and social sustainability concerns have created a need for investors and other stakeholders to accurately evaluate the sustainability level of their investments, suppliers, customers et cetera. The evaluation is aided by increasing regulation and standardization of companies’ ESG reporting and performance, and the emergence of ESG performance rating agencies. However, ESG disclosure is voluntary and uncontrolled by authorities in many parts of the world like the U.S., the target area of this study. This leaves an information gap between companies and stakeholders that utilize ESG information in their decision-making.

​This thesis attempts to narrow the ESG information gap between companies and their stakeholders by analyzing the association between textual ESG disclosure in annual reporting and ESG performance. The theoretical background for the study is the positive signaling theory, which implies that companies disclose ESG information in their annual reports to send positive signals about their ESG efforts and improving ESG performance. The method for textual analysis of this study is to calculate the ESG disclosure intensities in annual reports by using a large language model FinBERT that classifies the reports’ text segments into ESG and non-ESG contents. To study the association between ESG disclosure and performance, descriptive and predictive OLS regression models have been created. The models contain control variables to reduce extraneous effects of, for instance, firm size, capital structure, and profitability on ESG performance and therefore enable an isolated assessment of the association between ESG disclosure and performance.

​The results of the regression models provide assurance for a robust, statistically significant positive association between textual ESG disclosure intensity in annual reporting and ESG performance. Despite the robustness of the relationship, the model cannot predict future ESG performance at a high level of accuracy. Therefore, the usability of this study’s predictive regression model is limited to providing indications instead of accurate predictions for companies’ future ESG performance based on their textual ESG disclosure intensity. The regression results show, however, that the incremental effect of textual ESG disclosure as a determinant of ESG performance is increasing, which implies that text analysis could prove to be more viable in this context in the future.

​To conclude, this study narrows the ESG information gap between companies and stakeholders by proving the positive signaling theory correct in the context of textual ESG disclosure intensity in annual reporting. The study also provides useful data regarding other ESG performance determinants for future ESG performance regression analyses.