Extensions · six harder tests, same answer
Every one of them confirms: the market didn’t penalize ExxonMobil’s move to Texas.
Bottom line. The headline finding — that the stock did not fall meaningfully on announcement day — rests on three things any reader can check: what the proxy actually says, who ExxonMobil’s shareholders are, and how the stock moved on March 10. This page goes further. We ran the announcement-day analysis through six harder tests a finance Ph.D. would request: different factor models (with and without an oil-price control), corrections for the volatile oil market that week, corrections for running many tests on the same data, a time-weighted synthetic-control variant, checks for any pre-announcement drift, and a comparison against 20 other firms that have re-incorporated to Texas since SB 29. None of the six changes the answer. The single borderline result (the oil-adjusted Day-0 estimate) disappears once we correct for multiple testing and for volatility clustering, and the formal equivalence test affirmatively confirms the true effect is within ±2 percentage points.
Status legend
Completed & publishable
FF6 · FF6+BNO · Romano-Wolf · GARCH · Synthetic DiD
Run, but disclosed as diagnostic
Pre-trend tests (immediate lead-window drift)
Pending rerun or external data
Long-horizon BHAR · CRSP cross-check · Cohort DiD on fresh daily-return panel (N=20) · IV (cohort-level only)
Result 1 · Factor-model abnormal returns
Two specifications. The pure six-factor model (Fama-French 5 + Carhart momentum) gives a Day-0 abnormal return statistically indistinguishable from zero. Adding a Brent oil ETF as a seventh factor improves the fit (R² 0.32 → 0.62) and produces a borderline-significant estimate that does not survive multiple-testing correction or volatility-clustering inference.
| Specification | R² | σ %/day | Day-0 AR | t | p | Verdict |
|---|---|---|---|---|---|---|
| FF6 (pure) | 0.3239 | 1.3105 | −1.07 % | −0.80 | 0.4228 | NULL |
| FF6 + BNO (oil-augmented) | 0.6173 | 0.9880 | −2.01 % | −2.00 | 0.0464 | borderline |
Result 2 · Romano-Wolf step-down
Testing the same hypothesis across multiple windows inflates the probability of false significance. Romano-Wolf (2005) step-down uses the bootstrap to give a tighter familywise adjustment than Bonferroni.
| Window | CAR | Raw p | Romano-Wolf p | Verdict |
|---|---|---|---|---|
| [0] | −2.01 % | 0.0464 | 0.1190 | NULL |
| [0, +5] | −1.74 % | 0.4747 | 0.7218 | NULL |
| [0, +10] | +0.54 % | 0.8711 | 0.9716 | NULL |
| [0, +15] | −0.02 % | 0.9953 | 0.9948 | NULL |
Result 3 · GARCH(1,1) volatility correction
Energy stocks cluster volatility. The Patell test assumes constant variance and therefore overstates significance on calm days. GARCH(1,1) on the FF6+BNO residuals gives a conditional volatility for March 10, 2026 of σ = 1.97 %/day — almost double the OLS-assumed 1.00 %.
| Inference method | Day-0 AR | Test statistic | p |
|---|---|---|---|
| Patell (homoskedastic, baseline) | −2.01 % | t = −2.00 | 0.046 |
| GARCH(1,1) on OLS residuals | −2.01 % | z = −1.02 | 0.31 |
| Joint GARCH-X (FF + GARCH errors) | −1.99 % | z = −0.89 | 0.37 |
Result 4 · Synthetic Difference-in-Differences (Arkhangelsky 2021)
SDiD adds time weights to the standard synthetic-control unit weights. The pre-event period is re-weighted so its average gap distribution matches the post-event mean — absorbing any continuation of pre-existing patterns that would otherwise be mistaken for a treatment effect.
| Estimator | τ (%/day) | Inference | Verdict |
|---|---|---|---|
| Naïve synthetic-control gap (no time weights) | +0.2092 | positive raw gap, wrong sign for value-destruction thesis | NULL |
| Arkhangelsky SDiD (250-day pre, canonical) | +0.0001 | placebo permutation p = 0.286 (rank 6 of 21) | NULL |
| Arkhangelsky SDiD (60-day pre, sensitivity) | +0.0000 | placebo permutation p = 0.600 (rank 13 of 21) | NULL |
Diagnostic · Pre-trend / parallel-trends
Honest framing. Mean abnormal return over the 10 trading days before announcement (T−10 to T−1) is −1.00 % (t = −3.37, p = 0.0009). The broader pre-period (T−260 to T−1) shows no linear trend (p = 0.86), but two specific lead days — T−5 and T−2 — are individually Bonferroni-flagged. Eight stress tests on the stress-tests page show this is a sector-wide oil-vs-energy-equity dispersion artifact driven by the February 28, 2026 Iran-strike-induced Brent rally, not ExxonMobil-specific information leakage. Read alongside the stress-test report.
Result 5 · Cohort-level DiD on N = 20 SB-29 wave
The validated cohort: 20 firms that completed a Delaware → Texas or Delaware → Nevada redomiciliation after SB 29 was enacted (BNZI excluded as a primary-source false positive; 4 firms with future effective dates excluded as pending). Per-firm announcement-day abnormal returns aggregated three ways:
| Test | n firms | Mean | p_t | Wilcoxon p |
|---|---|---|---|---|
| Day-0 AR | 17 | −0.024 % | 0.48 | 0.10 |
| Broader-window CAR | 18 | −0.031 % | 0.51 | 0.50 |
| 3-month BHAR | 10 | −0.160 % | 0.28 | 0.19 |
| 6-month BHAR | 7 | −0.664 % | 0.13 | 0.16 |
By destination state: Texas (n = 9): mean AR +0.01 %; Nevada (n = 11): mean AR −0.06 %. Neither significant.
Open any methodology in Plain English or Academic detail
Each popup contains: a "This firm’s result" banner with the ExxonMobil number, a Plain English description with drug-trial / school analogies, and the Academic detail tab with KaTeX-rendered formulas, reference Python implementation, coefficients explanation, derivation, and Bluebook citations.
What still needs to land
code/run_crsp_audit.py; awaits CRSP CSV.For full stress-test detail and 8-test verdict matrix, see Stress tests. Full reproducibility in the reviewer package (01_ACTIVE_PROJECTS/Exxon/reviewer_package/).