The conditional value relevance of ESG and innovation disclosures: a multi-method analysis of European firms combining panel econometrics and machine learning

This study examines the value relevance of accounting fundamentals, social and governance performance, and innovation-related disclosures for non-financial European firms. Building on the Ohlson (1995) framework, the analysis combines panel-data econometrics with Random Forest machine learning to evaluate both inferential and predictive dimensions of value relevance. We find that non-financial disclosures are not uniformly value relevant. Social-governance performance and environmental innovation exhibit no direct linear association with firm value, while SDG 9 innovation orientation is negatively priced. Environmental innovation, however, emerges as the strongest non-financial predictor in the Random Forest models, revealing predictive content that fixed-effects specifications miss. Value relevance is further conditional on signal interaction: environmental innovation attenuates the valuation effect of social-governance performance, while stronger social-governance performance reduces the negative pricing of SDG 9 disclosure. Overall, our results indicate that disclosure type, signal interaction, and nonlinear predictive structure jointly determine whether sustainability and innovation information is priced by markets.