Mean-CVaR portfolio optimization under ESG disagreement

The ESG score of a company is a measure of its commitment to environmental, social and governance investing standards. ESG scores are produced by rating agen-cies using unique and proprietary methodologies. The complexity of measurement and the lack of widely accepted standards contribute to inconsistencies across agen-cies. Discrepancies in ratings issued by multiple data providers are particularly relevant in portfolio optimization problems that integrate ESG objectives into the classical risk-reward framework. In this work, we specifically study the impact on portfolio composition by examining Mean-CVaR-ESG optimal portfolios, where the objective function incorporates the portfolio's ESG score. To address ESG score discrepancies, we introduce a Distributionally Robust Optimization (DRO) reformulation of the Mean-CVaR-ESG model and assess its potential benefits. Our findings reveal a persistent divergence in optimal strategies across the investment horizon when ESG values from different rating agencies are used. We then apply the DRO approach by replacing a single provider's ESG score with a statistic de-rived from the scores of five different agencies. Our results show that, in this case, the DRO approach effectively mitigates score discrepancies by significantly reduc-ing optimal portfolio concentration while enhancing the ESG evaluation of optimal portfolios across all rating agencies.