Method for generating ESG ratings for unlisted companies based on NLP and predictive modelling

Abstract

Recent studies show a growing interest in integrating sustainability in investment processes. To our knowledge, there has not been any reported work on identifying and predicting environmental, social, and governance (ESG) performance among Norwegian unlisted companies. The overall purpose of this study is to investigate if there are opportunities to develop a tool that utilizes machine learning, text mining, and unstructured text from open sources to screen investment candidates. This feasibility study focuses on simplifying manual screening and assessments of ESG in existing investment processes. When investigating this opportunity, we have been using a Random Forest model for predictions and further analysis. The study contributes to theoretical and empirical insights into how the use of supervised machine learning can streamline assessments and decision making in existing investment processes. The study's findings show that it is possible to predict ESG performance based on information retrieved from open sources. Findings from the analysis show that the model manages to sort and categorize the text documents extracted and thus have low prediction errors, good overall accuracy and a high correlation between predicted and actual values. Further findings from the study show that the model has a good ability to generalize across different uses and markets. The study suggests that investors should replace manual evaluation of ESG on unlisted companies with a more efficient solution, to ensure future economic growth in the investment portfolio. It is also suggested that the model should be further developed before it can be commercialized and distributed among investors and players in the financial industry.