Stress tests — the leakage finding does not hold

Eight tests at a glance

Each test attacks a different way the finding could have been ExxonMobil-specific.

If the −11.77 pp [T−10, +0] cumulative under FF6+BNO had been a real information-leakage signal, at least one of the eight tests below should have isolated ExxonMobil. None do.

Robust

Estimation-window sensitivity

Five estimation windows from T−120 to T−1,000. CAR range −11.1 to −15.9 pp; all p < 0.01. But β_BNO rises in shorter (more recent) windows from 0.42→0.57 — the opposite of what the stale-β interpretation would need.

Fails

Oil-instrument sensitivity

The signal exists only under BNO. Under FF6 alone (no oil control): CAR = −0.4 pp, p = 0.93. Under FF6+Chevron (peer-equity control): CAR = −2.3 pp, p = 0.39. The "leakage" is an oil-instrument-specific phenomenon.

Sector-wide

Peer-firm sanity check

21 oil & gas peers under the same FF6+BNO model. 18 had negative CARs. ExxonMobil ranks 7^th most negative, not first. SLB, Halliburton, Devon Energy, Occidental Petroleum, Baker Hughes, and Ovintiv all worse. Mean cohort CAR −7.95 pp.

Stable

Rolling β_BNO

60-day rolling β_BNO over 2 years. Mean 0.52 in the 60 days before the event, sd 0.05. The 14-days-pre vs 90-to-14-days-pre delta is −0.008. No ExxonMobil-specific break. β_BNO has trended upward over 2 years.

Apparent contradiction

Placebo-in-time

200 random pseudo-event dates. ExxonMobil's actual CAR falls in the top 0.5% of the historical distribution (empirical p = 0.005). Unusual by ExxonMobil's own pattern — but stress test 3 and 6 show the window was unusual for the whole sector, not for ExxonMobil.

Decisive

Sub-sector regression

Cross-sectional OLS over 21 firms: CAR = +1.54 − 18.59 × β_BNO, slope p = 0.0038, R² = 0.36. Refiners (inverse crack-spread, lowest expected oil sensitivity) had positive CARs of +1.1 pp. The pattern is mechanical.

Clean

ExxonMobil news / EDGAR audit

No ExxonMobil-specific material adverse event between Feb 1 and Mar 31, 2026. Q4 2025 earnings (Jan 30) were positive. EDGAR confirms zero public redomiciliation disclosure before March 10 — the DEFA14A, PRE 14A, and press release all dated March 10.

Cause identified

Oil-market geopolitical confound

U.S. and Israel launched joint air strikes on Iran on February 28, 2026. Brent crude rallied from ~$72/bbl on Feb 27 to over $100 within the window (peak ~$126 came later in the conflict). Strait of Hormuz disruption. Classic war-premium pattern: oil spikes, energy equities discount the war premium as transient.

Figure 1 · Headline scoreboard

The published Day-0 bounded null result is intact

Six inference frameworks applied to the announcement-day return. Five return null; one borderline result fails to survive multiple-testing correction and GARCH-corrected inference.

FF6 Day-0 AR

−1.07 pp

Patell p = 0.42 · null

FF6+BNO Day-0 AR (raw)

−2.01 pp

Patell p = 0.046 · borderline

FF6+BNO Day-0 (Romano-Wolf)

p = 0.119

Across 4-window family · null

FF6+BNO Day-0 (GARCH(1,1))

z = −1.02

p = 0.31 · null

TOST equivalence (±2 pp band)

p = 0.011

Affirmative equivalence claim

[T−10, +0] FF6+BNO (candidate)

−11.77 pp

p = 0.0002 raw · sector-wide artifact

Plain English. On the day ExxonMobil announced its Texas redomiciliation, the stock moved indistinguishably from zero under the standard six-factor model. Adding an oil-price control produces a borderline result that disappears under proper familywise-error correction (Romano-Wolf) and under volatility-clustering inference (GARCH). The longer leakage window initially looked large and negative, but every stress test below shows it is a sector-wide oil-equity dispersion phenomenon, not ExxonMobil-specific.

Figure 2 · The smoking gun

Each firm's [T−10, +0] CAR is mechanically predicted by its oil-sensitivity loading

If the "leakage" were ExxonMobil-specific, ExxonMobil should be a clear outlier above the regression line. Instead, all 21 firms cluster around a single line with slope −18.59 per unit of β_BNO. Refiners with inverse crack-spread exposure (green) sit at the top with positive abnormal returns.

What this chart proves. If the −11.77 pp cumulative was ExxonMobil-specific information leakage about the Texas redomiciliation, the ExxonMobil point would sit far below the regression line. It doesn't — it sits roughly on the line, exactly where its β_BNO loading of 0.49 predicts. Refiners (green, with inverse crack-spread exposure to crude) had positive abnormal returns. Services firms (red, highest oil-tailwind sensitivity) had the most negative. The pattern is mechanical.

Figure 3 · Refiner deep dive — the cleanest evidence

Refiners — with inverse oil exposure — had positive abnormal returns over the same window

Refiners profit from crack spreads — the gap between crude input cost and refined product output price. When crude spikes via supply shock and downstream gasoline prices don't follow proportionally, refiner margins compress. But when refined product prices spike along with crude, refiner margins widen. During the Iran-strike window, refined-product prices spiked along with crude, lifting refiner stocks while the model's BNO control predicted them to fall.

Why this matters most. A genuine ExxonMobil-redomiciliation information signal would have no reason to differentially affect refiners, midstream operators, or services firms. Yet the cross-section is perfectly ordered by oil-sensitivity loading. Refiners — by far the lowest expected oil-tailwind sensitivity — were positive. This is the cleanest single proof that the [T−10, +0] pattern is mechanical model-misspecification under a war-premium oil shock, not information about a corporate-domicile change at one company.

Figure 4 · Peer-firm ranking

ExxonMobil ranks 7^th most negative of 21 oil & gas peers — not first

Same FF6+BNO model, same estimation window (T−260 to T−11), same [T−10, +0] event window applied to every oil & gas peer firm in the panel. If the leakage finding were ExxonMobil-specific, ExxonMobil should rank #1 in absolute magnitude. Six firms (SLB, Halliburton, Devon Energy, Occidental Petroleum, Baker Hughes, Ovintiv) had more negative cumulative ARs than ExxonMobil during the same window.

What this chart proves. Six firms in the same panel and the same window had cumulative abnormal returns more negative than ExxonMobil's. None of those six is reincorporating to Texas. The "leakage" pattern is not about ExxonMobil; it is about how energy equities responded to a war-premium oil shock relative to a model fit on calmer co-movement periods.

Figure 5 · The macro confound

U.S./Israel air strikes on Iran — February 28, 2026 — drove the entire window

Daily returns of Brent oil (BNO ETF) and ExxonMobil over the leakage window. Oil rallied roughly +30% as Strait of Hormuz disruption cut Gulf production by ~10 mb/d. ExxonMobil rose but lagged its in-sample oil-sensitivity prediction.

What this chart shows. The two largest BNO daily returns — +7.3% on March 2 (first trading day after Iran strike) and +9.2% on March 6 — were not matched by proportional ExxonMobil returns. ExxonMobil rose on March 2 (+1.1%) but nowhere near the model's prediction. Cumulatively, BNO rallied roughly 30% over the 11-day window while ExxonMobil gained only modestly. The FF6+BNO model, estimated on a calm period of co-movement, was forced to predict ExxonMobil up ~10pp; ExxonMobil was down 1.7pp raw cumulative. The residual is mechanical, not informational. On the announcement day itself (March 10, the right edge of the chart), ExxonMobil's raw return was −1.5% — but Chevron's was −1.66% the same day, and the matched-pair abnormal return (ExxonMobil minus Chevron, market-model-adjusted) is +0.04 pp (t = 0.05, p = 0.958). After sector adjustment, the Day-0 effect is statistically indistinguishable from zero.

Full data

Complete peer-firm table

Every firm in the panel, FF6+BNO estimation T−260 to T−11, [T−10, +0] window. Ranked from most negative cumulative AR to most positive. Column 5 reports the per-firm geometric cumulative abnormal return (compound ∏(1+AR_t)−1), the convention used for ranking cross-sectionally and for the β_BNO-vs-CAR regression. The headline −11.77 pp figure cited elsewhere on this page is the Patell-standardized arithmetic sum of daily ARs over the same 11-day window, the convention used for parametric significance testing. ExxonMobil’s rank (7th of 21) and the sector-wide pattern are identical under either normalization.

Rank	Company	Sub-sector	Oil sensitivity (β_BNO)	[T−10, +0] CAR (pp)	Patell p
1	SLB	Services	0.495	−16.59	0.0007
2	Halliburton	Services	0.703	−15.73	0.0073
3	Devon Energy	Pure E&P	0.739	−15.44	0.0009
4	Occidental Petroleum	Pure E&P	0.750	−14.93	0.0013
5	Baker Hughes	Services	0.400	−14.41	0.0087
6	Ovintiv	Pure E&P	0.859	−14.20	0.0064
7	ExxonMobil ← subject firm	Integrated supermajor	0.486	−12.49	0.0002
8	Diamondback Energy	Pure E&P	0.721	−12.41	0.0065
9	APA Corp.	Pure E&P	0.979	−9.75	0.1401
10	ONEOK	Midstream	0.372	−8.90	0.0504
11	ConocoPhillips	Pure E&P	0.628	−8.87	0.0303
12	Chevron	Integrated supermajor	0.402	−7.59	0.0259
13	Targa Resources	Midstream	0.453	−7.20	0.1604
14	Energy Transfer	Midstream	0.231	−7.05	0.0636
15	EOG Resources	Pure E&P	0.571	−7.03	0.0621
16	Phillips 66	Refiner	0.441	−2.45	0.5946
17	EQT	Pure E&P	0.259	−2.12	0.7557
18	Williams	Midstream	0.142	−1.78	0.6940
19	Valero	Refiner	0.429	+0.92	0.8732
20	Marathon Petroleum	Refiner	0.412	+4.72	0.3388
21	Cheniere Energy	Midstream	0.255	+6.25	0.1855

The takeaway from this table

If ExxonMobil's pre-announcement drop reflected insider knowledge of the Texas redomiciliation, ExxonMobil should rank #1 most negative. It doesn't — six other energy firms had more negative cumulative abnormal returns over the exact same window (SLB, Halliburton, Devon Energy, Occidental Petroleum, Baker Hughes, and Ovintiv), and none of those six is reincorporating to Texas. Refiners (Phillips 66, Valero, Marathon Petroleum) had positive abnormal returns — the opposite of what a market-wide governance shock would produce. The pattern is mechanical (oil-sensitivity-driven), not informational (ExxonMobil-specific).

Conclusion

What stands. What gets disclosed.

What stands

The published Day-0 bounded-null headline is intact. The announcement-day return is statistically indistinguishable from zero under FF6 (p = 0.42), under FF6+BNO with Romano-Wolf step-down across the multi-window family (p_RW = 0.119), and under GARCH(1,1) volatility correction (z = −1.02, p = 0.31). The TOST equivalence test at ±2 pp affirms equivalence at p = 0.011. The cohort-level result on N = 20 SB-29-era completed-mover firms shows mean Day-0 AR −0.02% (p = 0.48). The published one-paragraph summary does not change.

What gets disclosed

The [T−10, +0] FF6+BNO CAR of −11.77 pp now appears in the verification log and methodology popups, framed as a sector-wide oil-vs-energy-equity dispersion artifact caused by the February 28, 2026 Iran-strike-induced Brent spike, not ExxonMobil-specific information leakage. The eight stress tests above are cited in support. The cross-sectional regression and the refiner-positive-CAR finding are the headline evidence for that framing.

What still needs verification before final submission

CRSP cross-check (WRDS query pending). Long-horizon BHARs (Ken French end-of-May and end-of-June refreshes). Calendar-time portfolio alpha at 3, 6, 12 months post-event. Workbook integrity fixes (BNZI flag flip; EQT, ONEOK, and Schlumberger added to the validated cohort panel). All are queued in the reviewer package.