When raters disagree: ESG rating uncertainty in Thailand stock market

Abstract

This study investigates how ESG rating uncertainty impacts the cumulative abnormal returns and stock volatility of firms listed on the Stock Exchange of Thailand. Using the constituents of the SET100 Index over the 2015–2024 period, ESG ratings from three major providers (LSEG, Bloomberg, and S&P Global) are combined into a normalized overall ESG score and three individual pillar scores. Rating uncertainty is measured as the cross-sectional disagreement among these raters, from which a Low ESG Uncertainty dummy variable is constructed, taking a value of 1 if a firm's rating disagreement falls within the bottom quintile (indicating high rater consensus), and 0 otherwise. Cumulative abnormal returns are estimated from the Carhart four-factor model and stock volatility from a joint Carhart–GARCH(1,1) model. The relationships are tested using firm- and year-fixed-effects panel regressions with industry-clustered standard errors. The results do not support the unconditional "greenium" prediction that higher ESG ratings lower abnormal returns; instead, the overall ESG score and, most notably, the Governance pillar are associated with higher abnormal returns. The more consistent finding concerns information quality: firms with low ESG rating uncertainty earn significantly lower abnormal returns, an effect that persists when macroeconomic factors are added. The interaction between the ESG score and low uncertainty, however, is statistically insignificant once the score, the dummy, and their product are estimated together, for the overall ESG score and for every pillar, so the return analysis does not support a conditional, interaction-based channel. For stock volatility, the ESG, Social, and Governance scores carry the predicted negative sign but are statistically insignificant, and the low-uncertainty dummies are likewise insignificant. The only exception is the Governance pillar, whose interaction with low governance uncertainty is negative and significant, offering limited and suggestive evidence that any conditional volatility-reduction channel is confined to governance. Macroeconomic variables generally have weak direct effects, as neither geopolitical risk nor corruption perceptions are consistently priced when introduced into the models.