An Automated Framework to Predict Environment, Social and Governance (ESG) ratings using Machine Learning

Abstract

In recent years, there has been a growing emphasis on the Environmental, Social, and Governance (ESG) approach to evaluate the long-term sustainability and social viability of corporations. This research proposes developing an automated framework that uses a company’s financial components to predict its ESG score via machine learning and multivariate time series methods. Most renowned companies have now started to report their yearly ESG ratings and with time the number of companies reporting ESG components is likely to increase. Consequently, there exists a wide area of research with a lot of potential due to the availability of ESG data, which can be analyzed for a better understanding of a company’s portfolio. Our findings show that machine learning methods achieved a minimum RMSE of 13.856 and MAPE of 32.731%, while time series forecasting methods achieved a minimum RMSE of 4.237 and MAPE of 8.494%. The Consumer Staples and Health Care industries were most predictable using machine learning, while Consumer Staples and Real Estate were most predictable using time series methods. MENAP and South America were the most predictable regions with both methods. These results can help up build an automated predictive framework that can aid financial analysts, portfolio managers, and policymakers in enhancing ESG score predictions and improving decision-making processes.