The Hedges That Don't Hedge

Executive summary · free edition · September 2026

Complete in itself. The finding is here, not held back to make you buy the report.

The question, and why the usual answer is worthless

"What does another war or tariff round do to my money?"

It is the best-evidenced question in our demand research — six independent sources agree on it, including two surveys with opposing sponsors. It is also the question this industry answers worst, because the standard exhibit is an artefact of its own window length.

What we tested, and killed

We expected the shocks that mattered would be the ones that changed the interest-rate path. They were not. Splitting 29 events by whether the policy rate moved 100 basis points in the following year produces a dead heat: −9.41% against −9.22%, nineteen basis points apart, with a permutation probability of 0.99 across 200,000 reshuffles.

What the measurement kept producing instead

The market's reaction to the event itself was trivially small. What made the following year bad was almost always something with nothing to do with the event.

Across 29 events since 1970, the shock's own drawdown and the following year's drawdown are almost uncorrelated: R² = 0.029. The event explains under three per cent of the year that came after it.

And in 20 of 29 events, the index had already regained its pre-event level before the one-year low even arrived.

Two examples that make it unarguable

Any report quoting twelve-month drawdowns after a geopolitical shock is, in the main, quoting recessions and bubbles with a war attached.

These are not cherry-picked. The event list was built mechanically, by a stated rule applied without exception from 1970, precisely because a hand-picked list is how this subject usually produces whatever answer the author wanted.

And we replicated it rather than caveating it

Most of the price history had a single source. Instead of noting that as a limitation, the research re-ran every event on an independent index: median event drawdown −1.28% against −1.30%, and 16 of 29 events up on day one on both series. The conclusions replicate.

Now the products sold for this

Eight funds, all sold into the same idea. Eight, not nine — the source table has nine rows because one fund appears twice, and this report said "nine" until the count was checked.

And the cash line means three different things. In five funds it is genuine residual money. In one it is securities-lending collateral — $535.1m on loan to 26 named borrowers, offset by a matching liability, so neither idle nor risk-free. In one it is swap collateral, and that one is the finding of the report.

The fund that is two-thirds synthetic

The 62.40% is real. The fund holds only 35.5% of its European exposure directly; the rest is eighteen single-name total-return swaps, filed with no counterparty and no notional field, and carried at mark-to-market rather than at exposure.

A reader of that fund's published weights understates its largest position by about six times and another by about three times. The holdings page does not show what the fund owns — and that, not a data error, is why an earlier screen could not reconcile two published cash figures for it.

What a "defence fund" actually holds

• One fund's top ten is 77.42% of the fund and 29.63 percentage points of defence revenue. Its largest holding, at 21.95%, is about 10% government by its own filing.
• Another fund's largest holding sold its defence segment in February 2024.
• A third holds 8.32% in an unlisted private company, inside a daily-dealing fund, classified by the issuer as "Diversified Telecommunication Services".
• A fund named for the alliance holds the stub of a conglomerate left after its aerospace arm was spun off, and a Hong Kong listco controlled by a Chinese state aerospace group whose US subsidiary the Department of Defense named a Chinese military company in June 2026.

None of this is a criticism of any index. It is what index rebalancing looks like from outside, and it means the fund name describes the index's intent rather than this month's contents.

The order books, and the gap between a pledge and a payment

Backlog coverage runs 1.83 to 4.35 years across the major listed names — but the spread is not a demand signal. One builds submarines on decade-long contracts and another sells radios on short cycles. A ratio built from two different production physics is not a ranking.

The more useful measure is what share is actually funded: 43.89% to 76.27%.

And one headline is 59% commercial: of a $289bn backlog, $170bn is jet engines and cabin interiors. Its defence backlog alone is $119bn — 1.27 years, not 3.09.

A pledge is not an order, and an order is not a payment. Under 1.5% of the announced headline has moved as cash.

Every growth rate, on both bases

One company publishes both bases itself, which is why we can show the wedge rather than assert it: +30.1% reported against +16.9% like-for-like — about €5.9bn was bought. Another: +39% against +30.5%. A third reports +17% organic and +11.7% reported, because a divestiture reverses the direction.

What we cannot tell you

• 29 events over 56 years is a small sample, and the events overlap in regime. Every statistic carries its permutation probability rather than an assertion.
• The event rule is ours. A different rule gives a different list, so the rule is printed in full before the list.
• Swap exposure cannot be fully measured when the counterparty and notional fields are empty. The six-times figure is a floor, not a measurement.
• One company could not be read at all — its site blocked automated access from both routes — and is marked not verified rather than estimated.
• The absence of a relationship is not proof of absence. We say the event's own reaction explains 2.9% of the following year in this sample. We do not say geopolitics cannot matter.

Verification. 2,575 automated checks across two research passes, none failing. 59 planted mutations, all 59 caught. The event-study checker is a full second implementation that does not import the measurement code, matching exactly rather than within tolerance bands. Thirteen errors were found and corrected inside the packs, and ten holes were found in the packs' own checkers across four mutation rounds — documented rather than repaired quietly.

Disclosures. The author holds no position in any fund or security named here. No issuer payment, no advertising. Impersonal research, identical for every purchaser, tailored to no one. No statement about what any reader should hold, no forecast. Finnacl Research is not an investment adviser.

The full report adds the eight-fund snapshot series ordered by cash weight, the complete event table, the backlog chapter, the adviser chapter and the evidence appendix.