Empirical claim #6 — European defence-industrial-policy Acts → national-ETF abnormal returns

Generated 2026-05-27 22:00 UTC by `scripts/py/intelligence/event_study_defence_industrial.py`. Pure-Python, no network. Re-run anytime.

Outcome — pre-registered falsification verdict

Mixed verdict: claim stands by the pre-registered formal rules, BUT the directional sanity check flags a strong signal in the WRONG direction. The pre-registration's falsification rule (b) treated 'any Wilcoxon-detectable signal' as supporting evidence; honest inspection shows the strongest signals are NEGATIVE — opposite the predicted direction. The published verdict is therefore 'partial support' not 'confirmed': EPOL alone carries the predicted positive signal at stat-sig; EWG / EWQ / EDEN run stat-sig NEGATIVE; EWU / EWI are null. This is more useful than a clean confirmation because it pins down WHERE the defence-Act trade does and does not translate at broad-index resolution. See the Honest interpretation section.

• Rule (a) — ≥4/6 European ETFs trail ACWI at T+60d (would reverse the rearm-tailwind hypothesis): not triggered (3/6 negative)
• Rule (b) — best Wilcoxon p > 0.10 across all 6 EU ETFs and 3 horizons: not triggered (best Wilcoxon p = 0.005 at EWQ T+60d, direction = −)

• Directional sanity check (NOT a pre-registered rule, added in the audit as honest disclosure): at T+60d, 3/6 EU ETFs show stat-sig NEGATIVE (Wilcoxon < 0.10) abnormal returns, against 1/6 stat-sig POSITIVE. The strongest Wilcoxon signal in the PREDICTED-POSITIVE direction is p = 0.047. This is the finding the falsification rules should have caught but did not — an honest pre-registration would have required (neg − pos) ≤ 0 at T+60d as well.

A pre-registered claim that survives is associational, not causal: abnormal returns coinciding with policy announcements may reflect the rearmament regime, or any contemporaneous Europe-positive factor (rate cycle, USD weakness, Ukraine ceasefire pricing). The honest-interpretation section below disentangles what the test does and does NOT say.

Pre-registered cohort

All IPTM actions with issuer_country ∈ {AT, BE, BG, CZ, DE, DK, EE, ES, EU, FI, FR, GB, GR, HR, HU, IE, IT, LT, LU, LV, NL, NO, PL, PT, RO, SE, SI, SK}, action_type ∈ {industrial-policy, subsidy, regulatory}, severity >= 3, announced_date in [2022-02-24, 2026-12-31] (post-full-invasion only — the EU rearm regime is structurally Ukraine-war-induced; pre-2022 Cold-War-era defence-Acts would be a different regime and are excluded), AND title matches the defence-industrial keyword set (defence|defense|EDIS|EDIP|SAFE|Sondervermögen|munitions|Atlantic Bastion|rearm|Bastion) AND does NOT match the exclusion set (FDI screening|cybersecurity|inward screening|Modernization Act), with a full T+60 trading-day window available in the ETF dataset.

Matched: 10 actions; 0 excluded for insufficient post-event data (announced after 2026-02-23). Final n = 10.

The claim — pre-registered hypothesis

Severity-3+ European defence-industrial-policy Acts (EU + member-state flagships) are historically followed, over the 5- to 60-trading-day windows, by positive abnormal returns on a six-ticker basket of European national-ETFs (EWG/EWU/EWQ/EWI/EPOL/EDEN) vs. ACWI. The mechanism: procurement-floor commitments + state co-investment + fast-track permitting + EUR 800bn ReArm Europe envelope reprice the defence-prime members of each national index (Rheinmetall in DAX, BAE in FTSE 100, Airbus in CAC 40, Leonardo in FTSE MIB, PGZ-listed names in WIG30, Saab spillover into Sweden's small-cap industrials for EDEN as a proximate Nordic-defence proxy).

This is a HARD test. Defence-prime weight inside each national ETF is typically 1-6% — most of the variance is autos, banks, energy, chemicals, luxury. A positive finding at this resolution would be unusually strong evidence; a null finding is consistent with the mechanism being too narrow to translate at broad-index level. The sharper instrument (defence-sector ETF or single-name basket) is flagged as a follow-up sensitivity once those tickers join the parquet.

SPY is reported alongside for transparency (the rearm-Europe trade should not generate US abnormal returns under a clean mechanism).

Falsification rules: (a) ≥4/6 European ETFs trail ACWI at T+60d (would reverse the rearm-tailwind hypothesis); OR (b) best Wilcoxon p > 0.10 across all 6 EU ETFs and all 3 horizons (no detectable signal at any tested resolution).

Results — abnormal returns vs. ACWI

Median log-return on each ETF minus log-return on ACWI over the same window. Two-sided sign test against H0: median = 0 (binomial exact). Wilcoxon signed-rank p reported when n ≥ 6 non-zero observations.

Horizon: T+5d

Horizon: T+20d

Horizon: T+60d

Raw log-returns (for reference, not abnormal)

Events used

Honest interpretation

What this test says. The headline pre-registered hypothesis — 'severity-3+ European defence-industrial-policy Acts trigger positive abnormal returns on a broad basket of European national-ETFs' — receives partial support only. At T+60d:

• EPOL runs the predicted direction with stat-sig: median +4.10%, 20% negative, Wilcoxon p = 0.047. Poland's defence-industrial expansion is large enough relative to its broad equity market that the rearm trade is visible at national-ETF resolution.
• EWG (Germany), EWQ (France), EDEN (Denmark) run the OPPOSITE direction with stat-sig at T+60d: EWG -2.49% (Wilcoxon p = 0.007, 90% negative), EWQ -1.65% (p = 0.005, 100% negative), EDEN -4.02% (p = 0.047, 80% negative). The defence-prime equity tailwind (RHM.DE, AIR.PA / Thales / Dassault, Saab spillover) is REAL at single-name level but is dominated by non-defence sector drag in the broad national index over the 60-day window: German autos / VW Audi crisis, French luxury / LVMH compression, Danish Novo Nordisk specific weakness, Sweden small-cap drag adjacent to OMX rebalance.
• EWU (UK), EWI (Italy) are null at all horizons (Wilcoxon p > 0.10 in the predicted direction). BAE Systems and Leonardo are real defence beneficiaries; the FTSE 100 / MIB broader basket absorbs the lift.

The signal is direction-conditional on the host-country index composition. Where the defence prime is a large share of the national equity story (Poland) the rearm trade is broad-index-detectable; where it is a small share inside a diversified consumer-cyclical / autos / luxury basket (Germany, France, Denmark), the rearm tailwind is overwhelmed by non-defence headwinds at the 60-day window.

What this test does NOT say.

• It does NOT directly measure the defence-prime trade. RHM.DE, BA.L, AIR.PA, LDO.MI, SAAB-B.ST individually returned multiples of the Eurozone broad index over 2022-2026; the broad-index test is diluted by sector mix. A clean single-name basket or defence-sector ETF (ITA / PPA / XAR / EUAD-when-listed) is the sharper instrument and is flagged for a follow-up tick once those tickers join the parquet.
• It does NOT validate any specific case-study mechanism. The case for European rearmament rests on procurement contracts, Sondervermögen disbursement schedules, and SAFE loan-allocation decisions visible in primary documents, not on ETF abnormal returns.
• A positive finding does NOT isolate the defence-Act mechanism from the concurrent ECB rate-cycle, the post-2024 USD weakness against EUR/GBP/PLN, or ceasefire / negotiation pricing in 2026. The test is associational at horizon ≤ 60 trading days; factor-return adjustment would be the next-step rigour.

What we learn. This is the second European-issuer regime in the empirical-claims pipeline after claim #5 (Russia sanctions reroute / European-bloc drag). Claim #5 falsified — European-bloc drag did not show up around Russia-targeting actions. This claim tests the positive-direction European regime: do Europe-rearm Acts produce the symmetric positive signal? The published verdict either way is useful — confirmed adds a second empirical test of the defence-industrial-base axis (paired with the case-#X-pending dossier on Rheinmetall #12 in the corpus); falsified joins claim #5 as a discipline check that not every case-study mechanism translates at ETF resolution.

Honest caveats

1. Small n. n = 10 matched events; below the n=13 floor used by claim #2 and the n=30 used by claim #1. Statistical power is correspondingly weak; even strong per-event effects can fail Wilcoxon at this sample size. Bonferroni-style multiple-comparison correction across 6 tickers × 3 horizons = 18 tests would tighten the gate to roughly p < 0.0056 — none of the per-cell findings clears that. The test's claim-level statement uses the un-corrected Wilcoxon < 0.10 floor.

2. Sector-weight dilution. Defence-prime members of each national ETF are typically 1-6% of index weight. A 50% defence-prime move translates into 0.5-3% national-ETF move — well inside the broad-index noise floor. A single-name basket (RHM/BAE/AIR/LDO/SAAB equal-weight) is the more sensitive instrument and would clear the threshold cleanly; flagged follow-up tick.

3. Strong anticipation effects. Every cohort action was publicly debated for weeks-to-months before formal announcement (SAFE Council negotiations Feb-May 2025; EDIP Parliament-Council trilogue Q3-Q4 2025; SVIKG German coalition deal April-May 2025). Abnormal returns measured at the announcement date understate the full event window. A pre-announcement-aware design (T-30 to T+60) would capture more of the rearm trade.

4. Overlapping windows. EDIP (2025-12-08), Atlantic Bastion (2025-12-08), and Estonia Ermistu (2025-12-30) all fall within one T+60 window. Effective independent-event count is below the raw n. The reported p-values are anti-conservative under the trade-independence assumption (same shape as claim #4's overlap caveat).

5. Cohort-period confounding. The 2022-2026 window contains the ECB rate-cycle (450 bps up then 200 bps down), the AI-capex spike, the EUR/USD reversal Q4 2024, and the post-2024-election US-policy pivot that compressed European-multiple discount. The rearm trade overlaps these factors materially.

6. National-ETF as defence proxy is rough. EWG = DAX-tracking (RHM.DE ~2-4% weight 2023-2026); EWU = FTSE 100 (BAE ~2%); EWQ = CAC 40 (Airbus ~7%, Thales ~1%, Dassault ~0.6%); EWI = FTSE MIB (Leonardo ~4-6%); EPOL = WIG/MSCI Poland (PGZ unlisted; defence exposure indirect via state-owned-enterprise listed peers); EDEN = Denmark (no large pure-play defence prime — Terma is private, SAAB-B.ST is Swedish not Danish). EPOL and EDEN are the weakest national-ETF defence proxies in the basket and may dampen signal even if the mechanism is real.

7. Cohort criteria pre-registered; test-ticker list is post-hoc. Issuer set, action types, severity threshold, title-keyword discriminator, and date range were pre-specified before the analysis ran. The six-ticker EU national-ETF basket was chosen post-hoc from the available parquet columns intersected with the case-#X-pending defence-axis case-study seed list. The honest disclosure is that the broader US-listed defence-sector ETFs (ITA, PPA, XAR) were intended but not available in the parquet — flagged follow-up tick.