Comparison of sectorial and financial data for ESG scoring of mutual funds with machine learning

Abstract Recently, funds and corporations have adopted environmental/social/governance (ESG)-related criteria as increasingly important criteria when evaluating investments. The ESG scores of mutual funds, which pool money from many individuals, are, therefore, significant as a sign of social commitment and deliver long-term results for investors deciding in which fund to invest. However, this scoring process is not transparent, and computing ESG scores requires an extensive review and monitoring of accountancy data. In this study, Machine Learning (ML) was employed to predict six ESG scores of 3192 mutual funds by comparing the performance of the models built solely using 13 variables of ESG-controversial sectorial involvement and 37 variables of financial data. In both cases, three classification and three regression algorithms were applied to the 72 simulations. The results demonstrate an acceptable fit of the Random Forest and Gradient Boosting algorithms in regression exercises (R 2 values of 60–80%) and good prediction capabilities for classification (accuracies of approximately 70–80%). The models obtained similar performance when predicting ESG scores either from financial data or ESG-controversial sectorial involvement, illustrating how financial variables are good predictors of ESG scores as data on direct involvement in non-ESG funds’ activities. Thus, a relationship between the financial performance of a fund and its ESG score exists and is predictable by ML, saving time and resources.