The Hedge Was Built for a Different Shock

Start with what happened, because it is the reason the question is being asked. We publish six months on deliberately: the next round of long-term capital market assumptions arrives around the turn of the year, and on this evidence that is where the answer belongs.

The strikes on Iran began on 28 February 2026. During March, shipping through the Strait of Hormuz was restricted and US crude oil rose from $67 to $103 a barrel. The geopolitical risk index we track recorded its highest daily reading of the episode, 500.8, on 2 March. Its reading for March as a whole, 330.2, is the sixth highest of the 500 months since January 1985. Collapse each crisis to its single worst month and March 2026 ranks fourth, behind September 11, the Gulf War and the invasion of Iraq, and ahead of the invasion of Ukraine at 319.0.

Here is what markets did from the last close before the strikes, 27 February, to the end of March.

The hedges moved with the risk, not against it. The ten-year Treasury yield rose 33 basis points, from 3.97% to 4.30%, in a month when a flight to safety should have pushed it down. Across the trading days of March, daily returns on intermediate Treasuries moved with equities at a correlation of +0.55, and gold at +0.44. A 60/40 portfolio of US equities and intermediate Treasuries lost 3.9%. Funding a 10% gold sleeve from the Treasuries made the month worse, at -4.8%.

What rose was what the shock ran through. Oil, and the dollar.

The equity loss was brief. US equities reached their low, 7.6% below the last close before the strikes, on 30 March, and were back above that close on 13 April, ten trading days later. By the end of August they were 12.4% higher, while gold was 15.6% lower.

The flows show what professional money reached for.

Over the five weekly readings dated 2 to 30 March 2026, US-listed exchange-traded funds recorded:

Government bond funds took in money at 1.5 times their 2025 pace while their prices fell. The buying continued, a further $25.3bn from April to June. That is the traditional reflex working exactly as designed, into an instrument that was losing value.

Gold funds lost $10.7bn in five weeks, after taking in $7.2bn over the six weeks before the strikes. Gold had risen 171.2% between February 2022 and February 2026. In the shock it was sold, not bought. A flow records what was done, not why.

Energy funds took in money in both wrappers, from a low base in 2025.

One month is an anecdote, so we went back through the whole record.

We define an episode as a month in which the index stood at least two standard deviations above its own average over the previous five years, with spike months that fall within two months of each other counted as one episode. That rule finds 13 episodes between January 1990 and August 2026, one every 2.8 years. For each, we measured markets over the month in which the episode began.

Split the episodes by what happened to oil and the pattern is plain. Three carried an oil shock: the Gulf War in 1990, Russia and Ukraine in 2022, and March 2026. Oil rose between 18.4% and 53.6% in those windows, and by at most 9.4% in any other. In all three, the ten-year Treasury yield rose, by between 27 and 50 basis points. Where we hold the data, long-dated Treasuries lost 3.9% and 4.2%, gold lost 1.7% and 11.1%, and equities lost 5.3% and 4.9%.

The other ten episodes rarely gave the hedge work to do. The median equity move in the month was +0.1%, and equities fell by more than 2% in two of the ten, September 11 and the attack on Israel in 2023. Yields fell in four of the ten. The three largest falls, 41 basis points in January 2020, 25 in September 2001 and 18 in August 2017, all came with oil falling. Gold rose in five of the eight episodes for which we hold its price.

The shocks that hurt equities were the shocks that hurt bonds. After September 11, the two largest equity losses in the record both came with an oil shock, and in both the bond sleeve fell alongside. The traditional hedge has a good record in the geopolitical episodes that mattered least, and a poor one in those that mattered most.

The third part of the question is whether history still fits. The playbook of buying duration and gold when geopolitical risk rises rests on two relationships. We measured both.

The first is between equities and bonds, and its change is well documented. We measured it on the same data because its size matters for what follows. From August 2002 to December 2021, the correlation of monthly returns between US equities and intermediate Treasuries was -0.31. From January 2022 to August 2026 it was +0.55. Measured over rolling three-year windows, it peaked at +0.67 after 2022 and stood at +0.44 at the end of August.

A negative correlation is what makes a bond sleeve insurance. A positive one makes it a second holding of the same risk. For two decades a balanced portfolio could take that insurance for granted. For the last four and a half years it could not, and March 2026 was that change arriving in a geopolitical shock.

The second is between gold and real yields. Since the end of 2021 the ten-year real Treasury yield has risen from -1.04% to 2.44%, 3.5 percentage points, and gold has risen 138.9%. That fact needs no model. For scale, from 2005 to 2021 the real yield explained 73% of the variation in the level of gold, and each percentage point rise came with gold roughly 27% lower. On that relationship gold's implied level fell 67%, and at the end of August gold traded at 6.5 times it. A single-factor relationship describes an asset with several drivers, so read the multiple as a measure of distance from the old anchor, not as a valuation.

Gold still responds to real yields month to month, but its level has come loose. The correlation between monthly gold returns and changes in the real yield was -0.50 before 2022 and -0.41 since. The sensitivity survived; the anchor for the level moved.

The move in the anchor is consistent with a change in who buys gold. Across the same period, central banks and other official holders have been widely reported as sustained buyers, adding gold to reserves for diversification rather than on a view of real yields. A large buyer that does not price gold off real yields changes what sets its level. Official-sector purchases sit outside the data behind this note, so we offer this as the explanation most consistent with what we measured, not as a finding of our own.

Gold entered March 2026 after a 171.2% rise in four years, and in the shock it was sold.

Timing a risk requires repetition, and this record is thin.

Thirteen episodes in 36.6 years cannot support a trading rule. Move the threshold between 1.5 and 3 standard deviations, or the look-back between three and ten years, and the count ranges from 7 to 18 across the twelve combinations. A rule estimated on a dozen events, three of which behave differently from the rest, ends up fitted to whichever kind happened most recently.

The equity loss usually reverses quickly. US equities fell in 11 of the 12 episodes we can measure; in June 2025 they rose through the episode month. In 7 of those 11, equities regained their level from the close before the episode within 20 trading days of their low, with a median of 10 trading days. The four that took longer were September 11, the Paris attacks, January 2020 and January 2022, and two of those four ran into a pandemic and a hiking cycle. Three months after the episode month, equities were higher in 8 of 12, by a median of 6.4%.

Trade shocks are a separate series. April 2025, the month of the largest US tariff announcements, is the highest of 500 months on the trade policy uncertainty index we hold. On the geopolitical risk index it ranks 59th of 500. A tactical process that treats every political headline as one risk is combining two shocks that the data records separately.

If geopolitical risk belongs anywhere, it belongs where portfolios are designed. So we examined the tools used to design them.

The stress library. A standard library we hold carries three scenarios: two recessions and an inflation-driven rate spike modelled on 2022. Each shocks 18 or 19 asset lines, and gold is the only commodity among them.

The scenarios describe losses over a full downturn and March is a single month, so compare the direction of each line, not its size.

Two of the three scenarios assume the flight to safety the energy episodes did not deliver. The third gets the direction of bonds right and gives gold a small gain. The shape observed in 1990, 2022 and 2026 is a fourth one: equities, bonds and gold falling together while oil and the dollar rise.

Long-term assumptions leave a similar gap. Of the seven providers with 2026 long-term capital market assumptions in our data, one publishes a gold assumption, three publish a commodities line, two publish an energy line and four publish an inflation-linked bond line. A portfolio built on those inputs can hold a gold hedge that six of its seven assumption sets leave unpriced.

What strategic work looks like. Three pieces, each measurable.

1. Two geopolitical scenarios, defined by channel. A growth scare, calibrated to episodes such as September 2001 and January 2020, in which yields fall and duration hedges. An energy-supply shock, calibrated to 1990, 2022 and 2026, in which yields rise and duration adds to the loss.

2. A hedge sleeve tested against both. Treasuries, gold, the dollar, inflation-linked bonds, energy and cash each behave differently across the two shapes. A mix that survives both is a strategic answer. A mix that survives one is a bet on which kind comes next.

3. Assumptions reviewed through the channel. Where a structural change raises the odds of energy-driven shocks, the place to express it is in the inflation, real-asset and rate assumptions that feed the allocation, reviewed with each annual revision, not in a new asset bucket.

Illustrative. Observed ranges, not a forecast. Equity, Treasury and gold ranges cover 2022 and 2026, the two energy episodes inside those price histories.

Four questions, answerable against your own book this week.

1. What did your hedges do in March 2026? Measure it line by line, from 27 February to 31 March. If the answer resembles section 01, your hedge sleeve is a growth-scare hedge. That can be the right design, and it should be a deliberate one.

2. How many of your stress scenarios have bonds and equities falling together while oil rises? That count is your coverage of the shock that has now happened twice in five years.

3. What correlation between equities and bonds do your risk numbers assume? A risk model estimated mostly on 2002 to 2021 carries a negative figure into a period in which the measured one has been +0.55. It will understate the risk of a balanced portfolio in precisely the shock this note describes.

4. What will you tell a client whose hedge lost money in a war? Section 02 shows allocators buying the traditional hedge while it fell. A manager who can explain the channel before the client asks turns a disappointing month into a conversation about portfolio design.

The risk measure. The Geopolitical Risk index (Caldara and Iacoviello, American Economic Review, 2022), built from the share of newspaper articles covering wars, military threats and terrorism, as held on our platform: monthly from January 1985 to August 2026, and daily to 8 September 2026. It measures coverage of geopolitical events, not their economic cost. We use its level, its rank within its own history, and its distance from its own five-year average.

Episodes. A month in which the index stood at least two standard deviations above its trailing 60-month mean, the current month excluded. Spike months within two months of each other form one episode. The first month the rule can test is January 1990. Across thresholds of 1.5, 2, 2.5 and 3 standard deviations and look-backs of 36, 60 and 120 months, the count ranges from 7 to 18. Episode names describe the principal event of the month and are ours.

Windows. From the last trading day before the episode's first month to the last trading day of that month, and from there to three month-ends later. A monthly window captures everything else that happened in the month: January 2022 includes the repricing of US rates ahead of the hiking cycle, and the three months after January 2020 include the onset of the pandemic. Energy episodes are those in which crude oil rose more than 15% in the window; the next largest rise in any episode was 9.4%.

Markets. US equities and short, intermediate and long-dated Treasuries are total returns on exchange-traded funds tracking each market. Gold is a physically backed gold fund. The listing currency and primary price source were verified for every series. Equity history begins in 1993, Treasury funds in 2002, gold in 2004 and the broad dollar index in 2006, which is why earlier cells in the episode table read n/a. The ten-year nominal and real Treasury yields and crude oil are public market series. Correlations use monthly total returns. The gold relationship is a regression of the logarithm of the gold price on the ten-year real yield, monthly, January 2005 to December 2021, carried forward unchanged.

Flows. Weekly net flows for US-listed exchange-traded funds by category, the five readings dated 2 to 30 March 2026 and the six dated 19 January to 23 February. The comparison is the 2025 average weekly flow multiplied by five. US dollars.

Stress library and assumptions. A standard three-scenario stress library, shocks as specified. Long-term assumptions from the seven providers with a 2026 vintage in our data; a provider counts as publishing a line if any of its 2026 asset classes names it.

What we did not do. We have not forecast the next episode or the oil price, recommended an allocation, or modelled costs and taxes. We have not tested inflation-linked bonds, energy equities or currencies as hedges across the episodes, which is the natural next step.

Where this could be wrong. Three energy episodes are a small sample; the mechanism and the regime evidence support the reading, and a fourth energy shock arriving in a period of falling inflation could behave differently. A news-based index sees what newspapers cover. The gold relationship is a single-factor description of an asset with several drivers.

Run section 08 on your own book. The four questions can be answered from your own records.

Or send us one portfolio and we will return a single page: what its hedges did in March 2026, how it fares in a growth-scare and an energy-supply scenario side by side, and what stock-bond correlation its risk numbers assume.

Or let us map the channel across your product range. Which products carry energy-shock exposure their clients believe is hedged, measured on the same data.

Every figure below is read from the same single derivation as the note. n/a marks a price history that begins after the episode.

A1. Markets in the month each episode began

A2. US equities through each episode

A3. Episode count under twelve rules

The rule used in the note is 2.0 standard deviations over 60 months, 13 episodes.

A4. The standard stress library, full shocks