Liu and Huang's environmental-risk screen looks plausible for an oil-chokepoint shock, but it is too early to trade it as a hedge. The evidence rests on a single cross-section at the 28 February onset. The coefficient reaches significance over seven sessions and disappears over three. At re-escalation it is -0.00003. And the energy sample, where the hedge question matters most, contains eight stocks.
Why should environmental risk predict Hormuz losses?
The Strait of Hormuz usually carries about one-fifth of global seaborne oil trade, with roughly 100-130 ships passing each day. A closure should hurt firms less if they rely less on fossil inputs. ESG risk scores might capture that reliance, especially through the environmental pillar, which measures environmental risk. Liu and Huang test the idea against the 2026 crisis: onset on 28 February, a memorandum of understanding on 17 June and renewed attacks on 7-8 July. They ask whether pre-crisis ESG risk predicted which large US firms lost least.
The event study follows a familiar design. Each firm's abnormal returns are measured against a market model using SPY, fitted over about 120 trading days ending 15 days before each event. The scores come from a publicly shared S&P 500 file dated 3 March 2021 and follow the Sustainalytics-style risk convention: lower means better ESG. Matching leaves 225 firms, with onset CARs available for 222. The authors regress CARs on the total score, controlling for size, P/E, dividend yield and sector fixed effects. They also divide firms at the median score and examine the E, S and G pillars separately. For onset, they repeat the analysis using Fama-French five-factor returns.
Over [-3,+3] around onset, the low-risk half earned +0.53%, against -1.38% for the high-risk half. That makes a gap of 1.91 percentage points (N = 118 and 104). Using five-factor CARs, the respective returns were +1.26% and -0.57%. With size controlled, the total-score coefficient is -0.00110 (p = 0.009). Adding sector fixed effects takes it to -0.00153 (p = 0.002). The environmental pillar alone yields -0.00202 (p = 0.005); social and governance are insignificant. The authors describe all these results as associations and acknowledge that unobserved operational flexibility could confound them.
Reading the 1.91-point gap
The halves average seven-session CARs. Each firm's CAR is calculated against its own fitted SPY market model, so the gap measures relative performance against an estimated benchmark. The paper gives no trading rule and studies no other crisis. Even the high-ESG half's +0.53% is relative to each firm's market-model prediction; the sample's mean CAR was -0.37%.
The authors keep the magnitude in proportion. A standard deviation of ESG risk, about 8.6 points, corresponds to 0.95 percentage points of onset CAR. That is one-sixth of the 5.7% cross-sectional SD. The regression controlling for size has an R² of 0.047, while the raw correlation of total score with CAR is -0.17.
The environmental pillar has more to prove
The pillar results are the paper's most interesting evidence, and the easiest to overstate. E's coefficient is -0.00202, compared with -0.00110 for the total score; their raw correlations are -0.23 and -0.17. Those figures fit an energy-security explanation. They do not establish lower energy exposure. There are reasons to hold back.
Each pillar enters a separate regression. Governance's point estimate is -0.00139, with a standard error of 0.00152, too imprecise to treat as zero. Its raw correlation with onset CAR is -0.01. We found no test of whether the E and G coefficients differ.
The controls leave a larger question. In Table 5, the pillar regressions control only for log market cap. Elsewhere the paper reports an E coefficient of -0.00205 (p = 0.005) with P/E and dividend controls. We found no pillar result that clearly includes sector fixed effects. Sector means cannot explain away the total-score finding: adding sector fixed effects strengthens its coefficient from -0.00110 to -0.00153 (p = 0.002).
E still needs that test. Energy and Utilities have the highest total risk scores in the paper's sector chart. Without a sector-FE estimate, the environmental pillar could simply sort firms by sector or oil beta. The authors acknowledge the missing evidence. Their data contain no direct energy-exposure measures such as refining throughput, Middle East sales or equity oil betas.
Splitting out energy firms does little to resolve this. The non-energy coefficient is -0.00139 (p = 0.001, N = 214). For energy firms it is +0.00123 (p = 0.571, N = 8), with an interaction term of +0.0027 (p = 0.208). The hypothesis section calls the split exploratory. Yet the conclusion describes the evidence as "consistent with H1 and H2," with H2 being the energy prediction.
Seven sessions carry the result
A hedge should pay when the shock arrives.
Over [-1,+1], the ESG coefficient is +0.00014 (SE 0.00017): insignificant and signed the wrong way. Widening the window to [-5,+5] produces -0.00116; over [-10,+10] it is -0.00124 (p = 0.067). At July re-escalation, the coefficient falls to -0.00003 and the portfolio gap is 0.35 points. The onset date, 28 February 2026, was a Saturday. We did not find an explanation of how day 0 is mapped to trading sessions.
The authors disclose where the result is concentrated. Their abstract says the strongest findings come from multi-day windows around the initial restriction. They argue that significance fading away from onset supports an event-concentrated effect rather than a permanent return difference. Pre-event drift remains unresolved, though. The [-3,+3] window includes three pre-event sessions, and the wider windows include more. A placebo using only pre-event days, such as [-10,-2], would distinguish those explanations. The authors identify that test and defer it to the replication package.
The standard errors also warrant a discount. HC1 treats the 222 firms as independent even though they all experienced the same shock on the same date. Sector fixed effects remove differences in sector means. They leave residual correlation within sectors intact.
Before using a 2021 snapshot
The score snapshot is dated 3 March 2021, about five years ahead of the event. The authors acknowledge that it may miss changes made in 2024-2025. Its 225 firms amount to less than half the index, and Information Technology is relatively well represented.
We could not rerun the test. Our data contain neither the 2021 snapshot nor comparable pillar scores. Our price history also stops in 2025, before the events in 2026. Substituting a fundamentals-based or news-based sustainability score would test a different signal.
Two findings would move this from a curiosity to a candidate: a [-10,-2] placebo near zero, and an E-pillar coefficient that holds up with sector fixed effects and an oil-beta control.