Ml-Based Methodological Framework for Feature Selection in Panel Dataset : An Application to ESG Controversies Prediction in Banks

Mistreating Environmental Social and Governance (ESG) concerns has serious drawbacks in every kind of organizations, and even more in banks. Deeply revolutionized in its taxonomy of risks, banking sector is herein evaluated focusing on the integration of ESG parameters that, when lacking, leads ESG-related controversies (ESGC) arise. Scandals from recent and old past abound, and gain a disruptive public attention. Thereby, the article approaches the almost uncharted territory of ESGC in banks, by means of machine learning. Aiming at selecting the set of features that are relevant in ESGC prediction, techniques belonging to feature selection will be performed, by proposing a newly built methodology to deal with this task in panel data settings. Once described, the methodology is applied to a real panel dataset of 140 banks evaluated for a wide set of features over 2011-2020 time-span. The joint support of Resource Dependence and Internal Marketing theories reflect the used theoretical lens to question the power of governance-employees dynamics detains in predicting ESGC performance of banks, that later is confirmed as main finding. Finally, we provide implications for researchers, practitioners and regulators, further confirming the need for the rapid inroads that machine learning tools are actually making in the banking toolkit and in the regulatory technology