Exploring the predictive power of ESG Information for bank risk: a machine learning approach

This empirical study investigates the relationship between ESG information and bank risk for an international sample of banks from 2015-2022.

The study aims to address two main questions:

1. Does a non-linear association exist between ESG information and bank risk?
2. Can ESG information serve as a useful predictor for bank risk?

This analysis responds to the need for integrating machine learning into the modern financial analysis landscape. By building six distinct random forest models, each incorporating specific, or no, ESG indicators and employing several measures of risk, a comprehensive comparative analysis is conducted.

The main findings indicate that there is no strong evidence to support that ESG information is a useful predictor for bank risk. Further there is clear evidence of non-linear behaviour, underscoring the necessity for sophisticated analytical techniques to capture the association of EGS information and bank risk. Moreover, the study raises concerns regarding the reliability and maturity of ESG information.

Consequently, this implies prior research to be flawed and underlines the need for its re-evaluation and highlights the need for policymakers and supervisors to establish and implement consistent regulations governing the disclosure of non-financial information such as those relating to ESG activities.