Evaluating companies’ performances according to environmental, social, and governance (ESG) standards has become a central task in the financial industry. We show a novel solution to fine-tune transformer-based models for the ESG domain. By combining ESG ratings with text documents from annual reports, we were able to train an ESG sentiment model that outperforms traditional text classifiers at predicting the ESG behavior of companies by up to 11 percentage points. Moreover, we show practical applications of our ESG sentiment models by predicting individual sentences and by tracking ESG-related news coverage over time.
Paper
Full text
NLP for Responsible Finance: Fine-Tuning Transformer-Based Models for ESG
Semantic Scholar · Computer Science · 2022
Abstract
Evaluating companies’ performances according to environmental, social, and governance (ESG) standards has become a central task in the financial industry. We show a novel solution to fine-tune transformer-based models for the ESG domain. By combining ESG ratings with text documents from annual reports, we were able to train an ESG sentiment model that outperforms traditional text classifiers at predicting the ESG behavior of companies by up to 11 percentage points. Moreover, we show practical applications of our ESG sentiment models by predicting individual sentences and by tracking ESG-related news coverage over time.