Working Paper: NBER ID: w30562
Authors: Florian Berg; Julian F. Koelbel; Anna Pavlova; Roberto Rigobon
Abstract: Existing measures of ESG (environmental, social, and governance) performance ESG ratings are noisy and, therefore, standard regression estimates of the effect of ESG performance on stock returns are biased. Addressing this as a classical errors-in-variables problem, we develop a noise-correction procedure in which we instrument ESG ratings with ratings of other ESG rating agencies. With this procedure, the median increase in the regression coefficients is a factor of 2.1. The results are similar when we use accounting profitability measures as outcome variables. In simulations, our noise-correction procedure outperforms alternative approaches such as simple averages or principal component analysis.
Keywords: No keywords provided
JEL Codes: C26; G12; Q56
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| ESG performance (L25) | stock returns (G12) |
| noisier measurement of ESG performance (L25) | lower sensitivity of stock returns to ESG performance (G38) |
| attenuation bias (D91) | regression estimates towards zero (C51) |
| 2SLS regressions (C20) | stronger effect of ESG performance on stock returns (G17) |
| noise-correction procedures (C20) | true impact of ESG performance on stock returns (G38) |