Abstract
This study examines associations between ESG scores and financial performance in
a secondary, provider-processed cross-sectional snapshot of large-cap U.S. firms from
the sburstein/ESG-Stock-Data repository (Burstein, 2021). The analysis combines a prespecified
baseline OLS association screen, adjusted OLS specifications, sensitivity checks,
and machine-learning predictive benchmarking under a fixed protocol. In this covered
snapshot, adjusted OLS specifications show directionally positive ESG-pillar associations
with EPS, the baseline S1 screen indicates a positive total ESG-EPS association, and random
forest shows lower predictive error than benchmark models under the stated protocol. The
contribution is a decision-oriented evidence structure for non-specialist readers: pairwise
orientation, adjusted directional readout, and exploratory predictive benchmarking are
separated, each core statement is mapped to one artifact, and interpretation boundaries are
stated explicitly for cross-sectional design, correlated predictors, provider-specific ESG
measurement, and split-specific predictive uncertainty. All conclusions are association-safe
and non-causal, and findings are restricted to this provider-covered one-date snapshot.