Cross-firm empirical evidence · DExit reincorporation wave
Since the Tornetta ruling, scores of U.S. public companies have moved their state of incorporation — most to Nevada, the policy story to Texas. We measured the stock-price reaction the day each move was first disclosed. Across every test, every size cut, and every direction of move, the average abnormal return is statistically indistinguishable from zero. A null here is neutral evidence: it neither confirms a Texas penalty nor a Delaware-exit premium, and the design is honestly underpowered for the small effects the governance literature expects.
Every interval crosses zero. The highest-signal read — large, liquid firms where idiosyncratic noise is lowest — sits closest to the line. Figures hydrate from the canonical dataset; contested denominators are flagged in the data-lock panel at the foot of the page.
Each step removes noise, not signal — from every documented candidate down to the firms with clean, tradable price histories the model can actually measure. The same discipline a clinical trial uses to account for every enrolled patient.
Think of it like a clinical trial's enrollment chart. Start with every U.S. public company we can document that might have changed its state of incorporation since the Tesla vote. Confirm the ones that genuinely completed the move. Narrow to the ones with enough daily stock-price history to measure a reaction. What's left is the event-study cohort. Every company that drops out is accounted for, with the reason — click a grey routed-out segment to see exactly which firms and why.
What this chart provesNothing is silently dropped.
The study is a treated-versus-never-treated comparison. Comparing a mover to a near-identical Texas firm that stayed put is what isolates the effect of moving.
| Group | Role | n | What it is |
|---|
Most chose Nevada. Texas is the next-largest destination and the policy story behind SB 29. A handful went the other way — into Delaware — the counterflow that any honest tracker has to show.
What this chart shows
For each mover we line up its return on announcement day against what the market model says it should have done, using a long pre-event window to set the baseline. The gap is the abnormal return.
If the S&P is up 1% and Tesla — which usually swings about twice as hard — is up 1.8%, that's roughly expected. If Tesla instead jumped 4% the day it announced something, the extra ~2.2% is the "abnormal return": the part attributable to the news, not the market. If reincorporating mattered to investors, these gaps would cluster away from zero. Five different statistics ask the same question; all say no detectable signal.
What this chart showsA one-factor market model is fitted on the pre-event estimation window; an 11-day gap keeps any event-day data from leaking into the baseline. Day 0 is the first SEC-filed disclosure of the proposal; the test reads the abnormal return on Day 0 and over the two-day [0, +1] window. Hover any band for its definition.
The benchmark is the single-factor market model, estimated by OLS on the pre-event window:
where \(r_{i,t}\) is firm \(i\)'s daily return, \(r_{M,t}\) the market benchmark, \(t_0\) the announcement trading day, \(L\) the estimation length and \(g\) the gap before the event. The day-0 abnormal return is the realized minus the model-fitted return:
The cohort average abnormal return aggregates across firms; cumulated over an event window \([\tau_1,\tau_2]\) it is the CAAR:
Patell standardized statistic
with \(\widehat{\sigma}_i^2\) the estimation-window residual variance (Patell 1976; MacKinlay 1997). Because event-induced volatility inflates the standard error, the standardized cross-sectional test of Boehmer, Musumeci & Poulsen (1991) and the non-parametric rank test of Corrado (1989) are reported alongside the parametric \(t\) and sign tests; cross-event correlation is handled by the Kolari & Pynnönen (2010) adjustment and by Romano & Wolf (2005) step-down resampling.
Two one-sided tests (TOST) evaluate equivalence against a pre-set margin \(\pm\delta\). The null is non-equivalence; rejecting it requires both bounds to clear:
A non-significant TOST (large \(p\)) fails to establish equivalence — it does not affirm it. Minimum detectable effect at power \(1-\beta\) (Brown & Warner 1985; Cohen 1988):
One row per sample cut: the full cohort, the size screens, and the four directions of move reported separately. The dot is the average reaction; the whisker is the 95% interval; the shaded band is the equivalence margin (TOST). Hover a row for its full test detail; toggle the event window and the multiple-testing correction.
What this chart proves
| Sample cut | n | Mean AR | 95% CI | t-test p | Patell Z | Inside ±MDE? | Verdict |
|---|
A null is only as strong as the test's ability to find a true effect. This curve shows the probability of detecting an effect of a given size — and the shaded floor below which any real effect would slip past undetected.
Like a pregnancy test with a sensitivity limit: a clear negative is meaningful only if the test could have turned positive at the level you care about. Drag the sample-size slider. With the cohort we have, the smallest reaction we could reliably catch is around a full percentage point. The governance literature expects effects a third to a fifth of that size — which sit squarely inside the grey "undetectable" zone. That is why we call this a power-limited null, not proof of no effect.
What this chart provesThe literature's expected band (20–50 bp) lies inside the grey undetectable zone at the cohort's sample size. The data are consistent with no effect and with a true effect too small for this design to resolve. Honest framing requires reporting both.
Power is computed for a two-sided one-sample test at \(\alpha=0.05\) against a true mean \(\mu\), with panel residual dispersion \(\sigma=\) percentage points (the cohort estimate flagged for reconciliation in the data-lock):
The slider recomputes the MDE live as \(n\) changes; the curve is \(1-\beta\) as a function of \(\mu\) at the selected \(n\).
Every firm the model can measure, with its direction of move, the event date, the Day-0 abnormal return, and the EDGAR accession that fixes the disclosure. Sort any column. Where a value is not yet pinned to a primary source it is marked source pending — never invented.
| Firm | Ticker | Direction | Event date | Day-0 AR | EDGAR accession |
|---|
This is the cohort's equivalent of a complete peer table: every measured firm shown, every disclosure traceable. Rows without a sourced return are held open, not filled with a guess.
Texas's headline reforms are opt-in. A firm can move to Texas and adopt the derivative-suit threshold (§21.552), the shareholder-proposal threshold (§21.373), both, or neither. The choice — not the destination — is where the governance content lives. Every cell below is verified to the firm's own operative or proposed instrument on EDGAR.
§21.552 lets a Texas company require that whoever sues the board on the company's behalf actually own a meaningful stake — up to 3% of shares. §21.373 lets it require a similar stake before a shareholder can force a proposal onto the ballot. Both raise the bar for small activists. Moving to Texas doesn't impose either one automatically; the company has to write it into its charter or bylaws. Most movers took neither.
| Firm | Status | §21.552 derivative | §21.373 proposals | Source instrument |
|---|
§21.552(a)(3) sets a derivative-standing floor measured on outstanding shares; the statute caps the electable threshold at 3% (a ceiling on the corporation's election), and the elected figure becomes the plaintiff's standing floor. §21.373 sets a shareholder-proposal threshold measured on voting shares — the lesser of \$1,000,000 in market value or 3% of voting shares, with a holding period. Different denominators, different bills (SB 29 vs SB 1057), different elections. The real-world drafting convention is to elect §21.419 (business-judgment codification) and then state "at least 3% of the outstanding shares"; a §21.419 election alone, without the 3% language, is not a §21.552 threshold adoption — a category error the verification pass specifically caught and reversed.
A null result invites the sharpest scrutiny. Each card states the strongest challenge a Journal-of-Finance or JFE referee would raise, our response, and the honest status of the fix — run, queued, or retracted. We do not claim a fix is done before it is.
A centerpiece earns trust by stating its own limits first. Three columns, in the discipline of the ExxonMobil brief.
The inference caveats, and an open data-reconciliation flag a reviewer should see before citing any denominator.
Flagged values are contested across the live page, the 2026-06-14 fresh run, and the adjudicated canonical record. Every number on this page is bound to a single object; resolve the flags in the canonical workbook and all surfaces above update at once. No figure here is hand-typed into the layout.