Stress tests · eight adversarial attacks on the headline finding
If the headline result were a real effect, at least one of these would have caught it.
Bottom line. The eight stress tests below are designed to falsify the headline event-study finding by attacking different ways the finding could have been an artifact (sector confounding, estimation-window selection, oil-instrument mis-specification, family-controlled two-class anomalies, etc.). For Dillard's specifically, three additional firm-specific stress tests apply: (1) sector-confound separation, (2) post-conversion re-ratification interaction, (3) department-store-shock isolation. All ten tests are pending the per-firm event-study run.
Eight canonical stress tests (pending per-firm run)
| # | Stress test | Attack vector | Verdict |
|---|---|---|---|
| 1 | Estimation-window sensitivity | Vary pre-window: 60d, 120d, 250d (canonical), 500d. If the headline depends on pre-window length, identification is unstable. | [PENDING] |
| 2 | Oil-instrument sensitivity | Vary BNO substitute: WTI, Brent futures, XLE, no oil instrument. For Dillard's specifically, oil instrument likely not load-bearing (auto firm). | [PENDING] |
| 3 | Peer-firm sanity check | Day-0 AR for every department-store peer. If most peers also moved similarly, Dillard's's AR is a sector signal, not a firm-specific announcement effect. | [PENDING] |
| 4 | Rolling β diagnostic | Dillard's's β to FF6 factors rolling over the pre-window. Unstable loadings invalidate the constant-β assumption. | [PENDING] |
| 5 | Placebo-in-time | 100 pseudo-event dates from the pre-window. Dillard's's actual Day-0 AR's rank among the placebo distribution. | [PENDING] |
| 6 | GARCH-corrected [T−10,+0] | Leakage-window inference under GARCH(1,1)-conditional variance. Standard test assumes constant volatility. | [PENDING] |
| 7 | Sector / sub-sector split | For Dillard's: specialty-retail-specific peer subset (Nordstrom, Ross, TJX) vs ICE-incumbent peer subset (Macy's, Kohl's, Burlington). Sub-sector dispersion test. | [PENDING] |
| 8 | Cross-sectional regression | Day-0 AR regressed on firm-level controls (market cap, β to market, β to sector ETF). Dillard's's residual against the regression line. | [PENDING] |
Dillard's-specific stress tests
- sector-confound separation. The January 30, 2024 Chancery ruling rescinding the Dillard family's two-class capital structure (Class A / Class B) is a within-window confounding event for any analysis using July 22, 2025 (post-after-close T+1) as the canonical event date. The stress test separates sector-driven price movement from redomiciliation-anticipation movement by re-running the Day-0 test with the canonical event date set to August 31, 2025 (effective date), then comparing the two specifications. [PENDING]
- Post-conversion re-ratification interaction. The August 19, 2025 annual meeting approved both the redomiciliation AND a re-ratification of the 2018 compensation plan. The stress test tests whether the meeting-day price movement (June 13 / June 14) can be cleanly attributed to one decision or the other. [PENDING]
- Auto-industry-shock isolation. The first half of 2024 included material department-store events (July 2025 sector earnings dispersion, tariff exposure, consumer-discretionary spending shifts, broad-line vs department-store sub-sector splits, Dillard's's own delivery-miss announcements). The stress test isolates Dillard's's AR from contemporaneous department-store shocks via the sector-ETF residualization on Day-0. [PENDING]
Founder-controller stress test
Standard event-study identification assumes the marginal trader sets price. For Dillard's, the Dillard family's ~N/A% voting stake may make this assumption non-standard: a family-controlled two-class could deliver a credible threat to leave the market entirely (take-private, redomicile again), and the marginal trader prices that threat. The stress test compares Dillard's's announcement-window AR to the AR of (a) other founder-controlled large-caps and (b) other DE→TX movers without family-controlled two-class status. [PENDING per-firm run]
Primary sources
See /methodology for the canonical inference framework. Per-firm data: /replication. Full Bluebook-format citations in SOURCES.md.