The U-Shaped ESG Performance Paradigm: Gradient-Boosted Machine Learning Evidence Challenging Sustainability-Financial Orthodoxy

Abstract

This dissertation tests whether ESG risk scales linearly with investment returns. It does not.
Using LightGBM on 500 S&P 500 companies over the 2019-2024 period, this dissertation identifies a distinct U-shaped pattern. High-ESG-risk stocks materially outperformed, delivering cumulative returns roughly double those of the other groups. Low-risk stocks provided stronger downside stability, maintaining drawdowns under 30% while preserving competitive Sharpe ratios. Medium-risk stocks underperformed on both dimensions.
To explain this result, the dissertation proposes a Governance-Driven Risk Premium. High-ESG-risk firms may attract activist investors whose governance interventions unlock value that markets price only partially in advance. Medium-risk firms, by contrast, may be insufficiently differentiated to attract either activist pressure or sustainability-premium capital, leaving them structurally disadvantaged.
The practical implication is direct. Investors should not treat ESG risk as a single monotonic screening variable. Instead, ESG categories should be evaluated as distinct risk factors with different return and protection profiles.
Keywords: ESG investing, machine learning, portfolio performance, gradient boosting, sustainable finance, governance risk, activist investors, U-shaped relationship JEL Classification: G11 (Portfolio Choice), G23 (Pension Funds), M14 (Corporate Culture), Q56 (Sustainability)