An integrated approach to esg index construction with machine learning

Abstract

This paper presents an integrated approach to the automated generation of Environmental, Social and Governance (ESG) ratings of companies from financial and textual data. Three different research avenues on ESG relationships are investigated, each presenting a machine learning model which approaches the ESG calculation from a different aspect. The first model generates annual ESG ratings, the second uses historical data to predict the ESG rating of the immediate next quarter, and the third predicts whether the ESG rating would rise, fall or remain stable year-on-year. The combination of these models provides a foundation for the construction of a fully automated ESG rating system.