Missing value imputation in environmental, social, and governance data: an impact on emissions scores

This paper addresses the challenge of missing value imputation in environmental, social, and governance (ESG) data from The London Stock Exchange Group (LSEG) and its impact on category scores. First, using simulations, I compare traditional and machine learning (ML) imputation methods and show that ML methods consistently outperform traditional imputation approaches. Applying these methods to real-world missing ESG data, I recalculate emissions scores and uncover notable discrepancies from LSEG-reported values, suggesting that LSEG’s methodology may unintentionally favor firms with more complete disclosures. Moreover, I identify a pattern that companies with larger market capitalization tend to have lower rates of missing data and receive higher emissions scores. These results show a potential missing data bias in ESG data that favors larger firms and the importance of imputing missing data with ML techniques to reach category scores that more closely capture actual sustainability performance.