Measuring ESG Performance: A Text Mining Approach

Details

Supervisors

Vrins, Frédéric

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

In this Master thesis, we propose to measure the ESG performance of a company by relying on the text mining analysis of its sustainability disclosure. More precisely, we are deriving ESG ratings from the number of occurrences of ESG terms in Annual and CSR reports. Of course, one cannot expect such a technique to provide a fully accurate indicator of ESG performance. Yet, in light of the importance of ESG described above and given the large number of firms investors have to analyze, this is a first step towards a more efficient and qualitative assessment of ESG performance. Concretely, we are evaluating the performance of ratings built by a set of 8 different models (the sum of words frequency, the Multiple Linear Regression, the Naive Bayes, the Support Vector machine with non-linear kernels, the Random forest regressor and classifier, and the Linear Discriminant Analysis) against the official ESG ratings from the firm Refinitiv. Eventually, we find that this approach of measuring ESG is at best when relying on a Random Forest classifier. Yet, even in this optimal situation the predictions made by the model did not perfectly approximate Refinitiv's rating. Despite these disappointing results, our study allowed us to highlight certain interesting avenues for research on the development of Text Mining based ESG ratings.