ESG Pillar Composition and One-Year Accounting Loss Onset: Evidence from Bloomberg-Covered Firms, 2020–2024

Abstract

This study examines whether the internal composition of an environmental, social and governance (ESG) rating contains incremental information about one-year accounting-loss onset after the overall ESG level is held constant. Bloomberg environmental (E), social (S) and governance (G) pillar scores are reparameterised as ESGLevel = (E + S + G)/3, GovernanceHalo = G − (E + S)/2, and ESContrast = E − S. The FY2020–FY2024 panel contains 13,745 firm-years for 2751 relatively large listed firms; the principal model uses 8029 eligible transitions and 472 loss-onset events. A one-standard-deviation increase in GovernanceHalo has an average marginal effect of −0.02 percentage points (95% CI −0.87 to 0.83). Firm-blocked cross-validation provides essentially no incremental one-year predictive gain. Reviewer-driven sensitivity analyses using within-year standardised pillars, a firm fixed-effects linear probability model, Firth penalised logit, stabilised inverse-probability weighting for complete-case selection, and dynamic pre-trend controls do not reveal a stable loss-onset association. A separate 30% annual market-capitalisation contraction outcome is negatively associated with GovernanceHalo and remains significant after false-discovery-rate adjustment (q = 0.024), but it is a valuation-contraction proxy rather than an accounting-loss or crash-risk measure. The previously observed E/S score build-out is no longer evident after controls for prior score changes, current E/S level and historical score volatility. The evidence therefore rules out large average one-year loss-onset effects in this Bloomberg-covered sample while leaving open smaller, heterogeneous, longer-horizon and provider-specific effects.