The Illusion of Predictability: Deconstructing the ESG-Alpha Link with Robust Econometrics and Validated Machine Learning

The relationship between Environmental, Social, and Governance (ESG) performance and stock returns is widely debated, with many studies at risk of methodological flaws that can produce misleading results. This study aims to rigorously test the ESG-alpha hypothesis using a framework designed to prevent false discovery. A dual methodology was applied to a global dataset of multinational firms. First, a Two-Way Fixed Effects panel regression was used to control for unobserved firm and time effects. Second, a suite of machine learning models (XGBoost, Ridge, DNN) was tested for predictive power using a stringent walk-forward validation protocol to simulate real-world forecasting and avoid lookahead bias. The panel regression revealed no statistically significant relationship between a composite ESG score and excess stock returns. Critically, the machine learning models failed to generate statistically or economically significant out-of-sample performance, with backtests resulting in substantial capital loss. Strong evidence of a market leverage effect was also found. These findings challenge the narrative of ESG as a direct source of alpha, demonstrating that predictability is likely an illusion of less robust methods. This research highlights the paramount importance of methodological rigor in quantitative finance and offers a critical blueprint for practitioners and academics evaluating ESG investment strategies.