Towards the estimation of ESG ratings: A machine learning approach using balance sheet ratios

Abstract

Despite the persistence of methodological inconsistency and uncertainty, ESG ratings are useful for assessing Environmental (E), Social (S), and Governance (G) risk, individually and as a system (ESG). The ESG rating class is the only investment selection parameter that measures asset class sustainability. This paper tests whether a selected set of balance sheet variables and a dynamic measure of systemic risk, observed at time t, have information content useful to identify a firm’s ESG rating class of at time t+1. Using EuroStoxx 600 firms for the period 2016–2021, we apply a Machine Learning (ML) model. Specifically, a Random Forest (RF) classification model estimates the ESG rating at time t+1 with unprecedented accuracy in the international literature. This agile and parsimonious model offers important information to the sustainable investor for making strategic investment decisions and paves the way for ESG rating estimation for unlisted companies and SMEs.