This study examines the non-linear association between environmental, social, and governance (ESG) ratings and carbon total factor productivity (CTFP) among Chinese industrial firms, integrating organizational ambidexterity with legitimacy, resource orchestration, and absorptive capacity theories to delineate how exploitative and explorative pathways condition this association across ownership structures.
An interpretable machine learning framework combining LASSO selection, random forest prediction, and SHAP decomposition is applied to 7,791 firm-year observations from Chinese A-share industrial firms (2016–2022). Lagged specifications, sub-period partitions anchored to the 2020 Dual Carbon pledge, and industry-exclusion analyses assess temporal stability and measurement sensitivity.
The ESG–CTFP association follows a threshold-dependent trajectory: bounded compliance costs dominate below an ESG score of 4.5, giving way to accelerating returns once cognitive legitimacy is attained. Capacity utilization conditions this pattern through a resource orchestration threshold, while green technology innovation amplifies it via absorptive-capacity-driven acceleration yielding the largest marginal productivity contribution. Analyst coverage and executive green cognition bridge legitimacy for private firms but generate institutional friction in state-owned enterprises. The Dual Carbon mandate compresses the compliance cost regime, confirming institutional responsiveness.
The study resolves the ESG–productivity paradox by demonstrating that contradictory findings reflect positional differences along a threshold-dependent trajectory. It operationalizes two ambidextrous pathways, identifies ownership-contingent boundary conditions in a transitional economy, and deploys interpretable machine learning to recover non-linear configurational patterns that parametric specifications suppress.
