Buy ESG Improvement, Not ESG Status

The evidence for consistently strong returns in environmental, social, and governance (ESG) investing remains mixed. Studies continue to produce conflicting results on the relationship between ESG ratings and stock returns, weakening confidence in the category.

Part of the problem is methodological: rating providers disagree sharply on what a top-rated ESG firm looks like. The larger problem is conceptual. The industry continues to treat ESG as a single, static firm characteristic. The evidence suggests it is a set of distinct and differently priced signals. For portfolio managers, asset allocators, and chief investment officers, composite ESG strategies may therefore miss the real source of alpha.

Governance delivers a consistent premium, environmental performance matters only where financially material, and year-over-year ESG improvement emerges as the strongest driver of excess returns. Markets may not consistently reward companies for having high ESG scores, but they do appear to reward companies that improve them.

This research examined US equity returns across all 11 Global Industry Classification Standard (GICS) sectors from 2009–2024, constructing long–short factors for each of the Environmental, Social, and Governance pillars separately, along with a year-over-year ESG momentum factor (ΔESG). The results challenge the conventional ESG investing playbook in four important ways.

1. Governance is the only ESG pillar with a clean, positive, cross-sectional premium. The G-factor delivers a positive, significant premium of approximately +18 bps per quarter (~72 bps annualized) in the full-sample baseline (t ≈ 2.3), and it remains significant in the fully extended model. This is consistent with a long lineage of governance research going back to Gompers, Ishii, and Metrick (2003). Agency costs are real, governance quality mitigates them, and the cross-section rewards firms that manage it well.
   
   Overweight well-governed firms directly rather than relying on composite ESG scores that may dilute the signal with weaker exposures from other pillars. A dedicated governance allocation provides a cleaner and more consistent return signal than a broad ESG overlay.
2. The Social pillar is negatively priced, and this is not an error. Firms with higher Social scores earned lower subsequent returns, with an S-factor coefficient of roughly –25 bps per quarter (~100 bps annualized; t ≈ –2.5) in the full sample. Two non-exclusive explanations fit the data. Either investors derive non-pecuniary utility from holding socially virtuous firms and accept lower returns as a result (Pástor, Stambaugh, and Taylor, 2021), or firms with weaker social profiles trade at a discount and earn a risk premium as compensation for reputational and regulatory exposure (Hong and Kacperczyk, 2009).
   
   For allocators, the implication is not that high-S companies are bad investments — it is that a portfolio tilted toward high-S names is likely to deliver lower risk-adjusted returns than an equivalent portfolio tilted toward high-G names. If your mandate is values-aligned, this is a price worth paying. If your mandate is to maximize risk-adjusted returns and you are running a generic ESG tilt, you may be implicitly paying the preference premium without realizing it.
3. Environmental performance pays where it is material, and only there. The E-factor is insignificant in the full sample but flips to positive and significant in Energy, Materials, Industrials, Utilities, and Real Estate — the sectors identified as environmentally material by SASB and the broader materiality literature (Khan, Serafeim, and Yoon, 2016). In Technology, Communication Services, Financials, and Consumer sectors, the E-factor is flat or negative.
   
   This is the empirical case against market-wide environmental tilts. A blanket “green” overweight averages across sectors where environmental performance is priced and sectors where it is not, delivering a signal diluted to near zero. The correct implementation is sector-conditional: overweight environmental leaders within materially exposed industries and accept that environmental leadership in low-materiality sectors does not deliver the same return signal.
4. ESG momentum is the factor that actually works. Of all the findings, the ESG momentum result is the one practitioners should internalize. Firms that improved their ESG scores year-over-year outperformed those with deteriorating or static profiles, robustly across every specification. The factor is economically meaningful: adding an ESG momentum tilt to a market baseline lifted the portfolio Sharpe ratio from 0.54 to 0.62, with lower volatility, lower maximum drawdowns, and higher Sortino ratios.

Three features make this finding investable rather than merely academic.

1. The signal is orthogonal to the standard factors. The ΔESG factor loads on a distinct principal component in the PCA decomposition — it is not repackaging size, value, or price momentum.
2. The premium is accompanied by improved risk characteristics, not compensated-for-risk.
3. The underlying mechanism is also coherent: ESG improvements diffuse slowly through markets because ratings lag, providers disagree, and investors remain skeptical of greenwashing. Those are precisely the conditions under which momentum-style premia tend to persist (Hong and Stein, 1999; Edmans, 2011).

1. Separate the pillars. Composite ESG scores combine a positive G signal, a negative S signal, and a conditional E signal into a single number with limited explanatory power. Decomposed factors carry cleaner information. If your factor model does not separate E, S, and G, it is discarding the signal.
2. Build an ESG momentum sleeve. Long the top 30% of year-over-year ESG improvers, short (or underweight) the bottom 30%. Rebalance annually to match the refresh frequency of the underlying ratings. The factor is orthogonal to existing exposures and improves mean-variance efficiency.
3. Apply materiality sector-by-sector. Use SASB/IFRS S1 materiality classifications to restrict environmental tilts to materially exposed sectors. Expect no E-premium in the low-materiality majority of the market.
4. Know what you are paying for. If the mandate is risk-adjusted return, a high-S tilt is a cost, not a benefit. If the mandate is values alignment, that cost may be acceptable, but it should be made explicit in client reporting rather than buried in a composite ESG score.

A portfolio manager running a generic “ESG integrated” strategy is likely running an unintentional and unbalanced combination of these factors. If the portfolio happens to tilt toward high-G, ESG-improving firms in materially exposed sectors, it will look like alpha. If it tilts toward high-S, static ESG leaders in low-materiality sectors, it will underperform, and the manager may struggle to explain why, because the attribution system does not decompose ESG.

Extending a factor-attribution framework to include a ΔESG factor and separate E, S, and G loadings is straightforward and illuminating. It converts ESG from a black box into a set of identifiable exposures, each with a defensible economic interpretation. Clients benefit from the transparency. Managers benefit from being able to distinguish genuine skill from unintentional factor bets.

These findings are based on US equities and Bloomberg ESG data from 2009–2024, and results may differ across providers or market environments. As ESG strategies become more widely adopted, the return premium associated with ESG improvement could narrow over time.

Still, the broader conclusion is clear: disaggregated ESG signals carry more information than composite ESG scores, particularly when ESG improvement is measured over time.

The past decade of ESG research may have been asking the wrong question: “Does ESG work?” treats ESG as a single characteristic. A better question is: “Which ESG signals are priced, in which sectors, and over what time horizon?”

Viewed through that lens, the evidence becomes clearer. Markets may not consistently reward ESG status, but they do appear to reward ESG progress.

Buy improvement, not status.