When something happens in the world, whether a war, a blockade or a shipping lane closing, it eventually shows up in oil prices, grocery bills, and the value of what you own. This page explains how we work out the size of that effect, how confident we are in each number, and what we still cannot tell you. No background in finance required.
A conflict on the other side of the world does not affect your savings by magic. It travels along particular routes: it pushes the oil price up, it makes shipping slower and dearer, it raises the cost of food, it moves currencies, it changes what governments spend on defence.
We map each situation as a set of those routes. For every route, we ask one question: when this route is under pressure, how much does each kind of investment typically move? Oil companies tend to gain when oil spikes. Airlines tend to suffer, because fuel is one of their biggest costs. Shipping firms gain when freight rates jump. We measure those typical movements from real market history, then add them up across everything you own.
Two situations are live on the platform. The Strait of Hormuz, the narrow sea passage most of the world's oil tankers pass through, works mainly through oil and shipping. The Taiwan Strait works mainly through computer chips, and has no oil route at all. They are deliberately different, because geopolitical risk is not one single thing that moves everything the same way.
If you enter what you own, the sums are done on your own device. Nothing is uploaded, stored, or sent to us. You can disconnect from the internet after the page loads and it will still work.
Six short explanations of what you see on the dashboard. The same text opens from the small “?” beside each of these on the dashboard itself.
Some of our numbers come from measuring what actually happened in a real crisis. Others are careful estimates, because markets have not priced the risk: either it has never happened, or it happened and share prices did not move with it. Those deserve very different levels of confidence, and we label which is which rather than presenting them as equally solid.
| Label | What it means | Used for |
|---|---|---|
| MEASURED | The event actually happened, and we measured how markets really responded | Hormuz / Iran |
| ESTIMATED | Share prices have not priced this risk. Either it has never happened, so there is nothing to measure (Taiwan), or it happened and share prices did not move with it (Red Sea). The figures are reasoned estimates, anchored to published research | Taiwan Strait · Red Sea |
| ROUGH DRAFT | A first pass that has not been checked against anything. Never shown to you | – |
Three rules keep these labels honest. A new situation never inherits another one's label. It has to earn its own. A label can only be upgraded by evidence, meaning something actually happened and we measured it. And every situation has a scheduled re-check date and a written list of things that would force us to redo the numbers, both shown on the page.
If you prefer the technical terms: these labels are empirical, unpriced and draft in the underlying model, and appear that way in the published source code.
The Strait of Hormuz was closed by war from 28 February to the ceasefire on 8 April 2026. Oil went from about $70.90 a barrel before the war to a peak of $138.20 on 7 April, then came back down. The fighting stopped; the crisis did not. Fewer ships pass through than before, insurance for those that do is still expensive, and our main working assumption treats today's uneasy truce as the normal state, rather than a hypothetical war.
Because this really happened, we did not have to guess. We took 41 investments, looked at their weekly prices from 2020 to 2026, and measured how each one moved during the crisis.
There is a trap here, and avoiding it is the most important decision in the whole model. Markets are always moving for reasons that have nothing to do with the crisis, including interest rate changes, an artificial-intelligence boom, ordinary business cycles. If you simply measure how much oil companies rose during the war, you capture all of that too, and you overstate the war's effect.
So for every measurement, we subtract what the whole stock market did over the exact same days. If energy shares rose 25% while the market as a whole rose 6%, we record the war's effect as 19%, not 25%. That single subtraction is what separates the crisis signal from everything else happening at the time.
| Type of investment | Affected through | We predicted | What really happened |
|---|---|---|---|
| Gulf oil producers | oil price | +4.9% | +3.3% |
| Energy companies | oil price | +19.6% | +19.4% |
| Oil tanker operators | shipping costs | +24.8% | +25.1% |
| Utilities | seen as a safe haven | +9.3% | +10.1% |
| Airlines | fuel costs, fewer flights | −8.7% | −9.9% |
| Luxury goods | people spend less | −17.9% | −19.7% |
We checked the model against the full 2026 war. On average it was off by about 2.5 percentage points per category. That is the honest margin of error, and it travels with every figure on the platform — but be clear about what it measures. This model was built on that war, so checking it against the same war tests how well it fits, not whether it can see ahead. A good result here is the floor, not proof it will do as well on the next crisis.
One more thing you should know about that figure. The model reads a crisis through seven pressure channels — oil, food, inflation, currency, shipping, growth, defence. Four were measured directly from market data at the time. The working notes for the other three did not survive, so we recovered them by fitting them back to the sector moves themselves. They are labelled in the code as reconstructed, not measured, and they are part of how the 2.5 was produced.
One of the four “measured” ones was not what it said. Until 28 September 2026 the food channel was recorded as a 25.6% rise in “a global food index”, with no index or date named. We now measure it on the FAO Food Price Index, the same food index the strain index below uses. During the war it rose at most 2.2% over twelve months (April 2026). After the invasion of Ukraine it rose 34.5%. Our figure was more than ten times too high.
We corrected it twice in one day, and both steps are on the record. We first measured it on the International Monetary Fund’s food index, which showed 7.8%. The same day we moved the whole platform to FAO’s index, which publishes weeks sooner, so that every food figure we show comes from one source. The two indices agree about Ukraine (FAO 34.5%, IMF 31.4%) but not about this war (FAO 2.2%, IMF 7.8%), because they weight different foods. We use FAO’s and say so. Food affects only the agriculture and food category, which has no observation from this war to check against, so the 2.5 above does not change. We recomputed it after each step: 2.40 every time.
The same check corrected the food figures in the three scenarios, which had been written by hand. Each now rests on a measured FAO reading: Normalisation −0.9%, the pace just before the war, when food prices were edging down (February 2026); Armed Truce +2.5%, the latest reading under the truce (August 2026); Re-escalation +3.1%, the highest the first closure reached (May 2026). They were +5%, +12% and +30%. A second closure could push food higher than the first did, but nothing we have measured says how much, so we do not put a larger number on it. Holdings in the agriculture and food category show smaller moves as a result.
The previous version of the model was off by 19 points. It had been built by assuming the Ukraine war was a good stand-in for a Middle East oil crisis, and it dramatically overstated how badly everything would be hit. We replaced it, and the change is on the record.
The six predictions above are the record as published in July 2026, when this version of the model was calibrated. Running the same war through the exposure model as it stands today gives a different answer on two of them.
| Type of investment | What really happened | Published Jul 2026 | Current model |
|---|---|---|---|
| Utilities | +10.1% | +9.3% | +4.8% |
| Airlines | −9.9% | −8.7% | −12.6% |
On both rows the July print tracked the real outcome more closely than the betas now shipping. We are not adjusting those betas to close the gap. Changing a measurement so a later number agrees with it is how a model stops being evidence. The proper way to settle it is a re-fit, and a re-fit is parked until there is a test designed to earn one. Until then this stays on the page as an open question rather than a correction. Anyone can reproduce both columns: scripts/retest-hormuz.mjs in the repository recomputes them from the shipped betas.
That replacement is the rule, not the exception: when reality disagrees with the model, the model changes. We cut our airline estimate by more than three times when the real fuel-cost damage turned out milder than assumed. We had grouped commodity funds with gold, expecting them to behave similarly; in the war they rose 24.5% while gold managed 0.9%, so we separated them. We had assumed developing-country investments would suffer; in both wars we measured, the ones that export raw materials gained more than the ones that import them lost, so we changed the sign. Corrections like these are not embarrassments to be buried. They are how the thing gets more accurate.
Correction, 27 September 2026. The Workspace's holdings panel did not use this model. It used an older table of its own, and for Hormuz that table ran two to five times hotter than the calibrated figures: under the Armed Truce it showed RTX at +52% where the model says +21%, ExxonMobil at +38% against +10%, and Delta at −30% against −6%. It also scaled every figure by a scenario multiplier we could not source. The table is gone. Every portfolio figure on the site now comes from the same model as this page, and the Workspace no longer shows a 3-year column, because the model has no measured basis for one.
Correction, 27 September 2026 (Hormuz oil). Two things were wrong. First, our record said Brent averaged $69 in July and called it a round trip to pre-war prices. The EIA's daily prices say July averaged $83.80; oil never went back. Second, our Armed Truce case, which describes the present, said Brent would spike "toward ~$102". When drones shut Saudi Arabia's East-West pipeline, the route that carries Gulf crude around the strait, Brent reached $130.80 on 15 September. The Armed Truce's peak is now that measured day, the same convention we used for the war's $138 peak, so oil-linked exposures under it are larger: an energy producer's modelled gain rises from about 9.5% to 17.3%, an airline's loss from about 6.3% to 10.8%. The sector sensitivities themselves are unchanged. The forward path now starts from September's measured $113 and eases toward the EIA's September outlook ($77 in the second quarter of 2027), which was published before the pipeline strike.
Update, 8 October 2026 (Hormuz oil). Two things moved, both measured. Brent set a new high since the ceasefire, $135.51 on 2 October, above the $130.80 of 15 September, so the Armed Truce's peak is now that day, by the same rule. Oil-linked exposures under it grow a little: an energy producer's modelled gain from 17.3% to 18.5%, an airline's loss from 10.8% to 11.5%. And the forward path now follows the EIA's October outlook, published on 6 October, after the pipeline strike, instead of September's, published before it: about $105 on average for the last quarter of 2026, easing to $74 by the end of 2027. The EIA's outlook stops there; holding $74 after that is our judgment. September is now the full month, $114.20. The sector sensitivities are unchanged.
Coverage, 27 September 2026. The model scores 116 stocks and funds directly; the Workspace lists them by sector. Any other ticker is now classified from its listed industry onto one of those sectors, and marked “classified”: an airline is scored as an airline, a bank as a financial. Where no sector of ours is close, as with health care or retail, it is scored like the broad market, which these events barely move, and we say so beside it. The mapping is our judgment, written down in one table; you can override it for any holding by picking a sector yourself.
Correction, 27 September 2026 (what-if panel). The dashboard's custom scenario builder was not this model. Its six sliders fed four formulas we could not source, it showed Brent near $145 while resting at our base case of $130.80, and one slider changed nothing. It is replaced by a single lever per structure: peak Brent for Hormuz, advanced chip capacity offline for Taiwan. Move it and the page shows how each sector's modelled exposure changes, computed by the same model as everything else here, with every other input left as in the chosen scenario.
Correction, 11 October 2026 (sector radar and risk matrix). Two more dashboard charts named sources they did not have. The sector “radar” said its 0–100 scores came from a sector correlation model with SIPRI and EIA inputs; the risk matrix said its probabilities came from GDELT event frequency; a GDP chart said it was built on IMF baselines and a conflict multiplier. Five of the Hormuz headline tiles did the same: shipping cited the Baltic Exchange, GDP cited the IMF and a multiplier, defence spending cited SIPRI, currencies cited the BIS, and duration cited a “duration model”. Each tile shows a scenario figure that is our judgment, and in the war we fitted to, shipping, GDP and defence were reconstructed by the fit, not measured. The tiles now say so. Both were tables we had typed. The matrix also contradicted our own model: under Re-escalation it gave a second Hormuz closure a 74% chance, while the model gives that scenario 25%. Both charts are now drawn from the model itself. The sector chart shows each sector’s modelled move under the scenario you are viewing, from the same engine as the Workspace. The scenario chart shows every listed crisis’s scenarios at the model’s odds and their effect on world output. The regional tables on the Continental tab (GDP and inflation by region, the impact matrix) are also our judgment; they now say so, and we are replacing them, figure by figure, with published estimates where those exist.
Correction, 11 October 2026 (stress heatmap). The dashboard's regional stress heatmap said its 0–10 scores came from a composite model of eight live dimensions, including news volume, GDELT tone and ACLED events. They did not. The scores are a fixed table we wrote: our own judgment of how hard each region is hit, year by year, with no measurement or published source behind them. The chart now says so. We also found that the AI commentary under the chart read its own, separately typed copy of the table, and for Taiwan the two disagreed (North America falling to 2.2 in the commentary, 1.5 on the chart). There is now one copy, and both read it. The values on the chart did not change.
Correction, 12 October 2026 (GDP tile). Under the Hormuz world-growth figure, the dashboard compared it with the 2003 Iraq war (“2003 Iraq = −0.5% · 1.4× that loss”). We could not find a source for the −0.5%, so the comparison is gone. The line now says what the figure is: a scenario figure, our judgment.
Correction, 27 September 2026 (intelligence map). The globe's regional stress has no scenario dimension, so something has to make a scenario visible on the map. It used a set of multipliers we could not source (×1.4 for the worse scenario, ×0.65 for the better one, and the same under both structures). The map now scales by the model's own scenario stresses relative to the base case: under Hormuz, Re-escalation shades at 2.6 times the Armed Truce and Normalisation at 0.4; under Taiwan, Invasion at 1.7 times the Quarantine and Gray-Zone at 0.1. Treating the regional table as the base case is a judgment, and we say so. Taiwan has no oil channel, so the oil-dependency layer is blank under it.
Taiwan has never been blockaded. The world's supply of advanced computer chips has never been cut off by conflict. So unlike Hormuz, there is no real event to measure, and rather than hide that, we made it the headline finding for this situation.
What we found: we tested five past scare episodes to see whether markets had ever priced this risk in. They had not. During the December 2025 Chinese military escalation, which was the most serious in decades, chipmaker shares rose 20.6% relative to the market. The artificial-intelligence boom did not merely drown out the fear, it reversed it. The August 2022 Pelosi visit produced the only genuine fear response we could find, and it was small: chipmakers fell 8.3%. A major earthquake that physically damaged chip factories in 2024, and the real chip shortage of 2021, left no usable trace at all. Every test, with its dates and verdict, is published on the platform.
What we estimated instead: our figures are anchored to published research from Bloomberg Economics, dated February 2026. It puts a year-long blockade at a 5.0% hit to the world economy. A war it puts at 9.6%, or roughly $10.6 trillion in the first year. We then worked out how each kind of investment would likely fare, based on how much of its business depends on Taiwan. These are reasoned estimates of something that has never occurred. They are useful for comparing your exposure and for thinking through what would matter. They are not a forecast, and we label them that way everywhere they appear.
What we got wrong first. We had treated the January–February 2026 “return to Suez” as a second test of how shipping shares respond to the route. Daily ship-tracking data (IMF PortWatch) shows the return never happened: container traffic through the strait stayed at its post-attack level every month since January 2024. Carriers announced it; ships did not come back. We withdrew the claim on 3 October 2026 and kept the original entry, with the correction beneath it. An earlier reading of the 2026 blockade window was also wrong: half of the “liner” move came from ZIM, which has traded as a cash takeover since February 2026. And our first draft of the sector figures gave utilities a borrowed “safe haven” boost that made them the biggest winner, gave shipping shares a size taken from a model, and gave carmakers a size nobody had published. All three were removed before this page went live.
What happened: from December 2023, after attacks on ships, container lines stopped using the Bab el-Mandeb strait and sailed round the Cape of Good Hope. About 19 container ships a day used the strait before the attacks, and 4 to 7 a day in every month since. Container traffic round the Cape roughly tripled. We publish the monthly series on the dashboard.
What moved: container freight, measured from Drewry’s weekly index, rose 122.7% over the reroute. Oil did not: Brent fell 9% over the same weeks, and world food prices fell too. Seaborne trade fell most in sub-Saharan Africa, Oceania, the Middle East and South Asia; Europe’s trade volumes held, because the cargo still arrived, just later.
What did not move: the shares of the container lines. Measured the same way we measured Hormuz, they did not follow the route in any way that survives a change of start date or a second company, and a shipping line that does not use the route at all rose under every starting date we tried. Gold showed no flight to safety either. That is why this situation is labelled estimated although the event is real: share prices did not price it, so there is no market response to fit.
How the figures are set instead: the scenario figures come from measurements (freight, trade, food and currencies) and from published research: OECD (February 2024) and the ECB (March 2024) on how freight costs feed into prices, the IMF on the countries along the Red Sea, and Allianz Research and the Federal Reserve Bank of St. Louis on world growth. Sector figures are the same growth sensitivities the Taiwan situation uses, unchanged and labelled that way. None is a Red Sea market measurement, none is shown as a winner, and shipping shares carry no figure at all. Oil and food are counted once, in the Hormuz situation, so the same barrel is never scored twice. Where no source gives a number, the dashboard says “not sized” instead of filling one in.
The two situations above go quiet for long stretches. Nothing is exploding this month, so those pages sit still. The DELAX Global Strain Index exists for exactly those weeks. It answers a different question. Has background pressure been unusually tight, complete through the last day all five inputs exist? A new reading appears when the slowest of those inputs has landed, not every morning. Between prints the figure does not change, and that is the index behaving correctly.
Zero means normal. Not zero pressure. Normal pressure, meaning the level that has been typical across the years we have measured.
Above zero means tighter than usual. Below zero means easier than usual. A reading of +1 means conditions sit one ordinary-sized swing above normal. That is noticeably elevated, but the sort of thing that happens from time to time. A reading of +2 or beyond is genuinely unusual, the kind of pressure seen only a handful of times in a generation.
It is deliberately not a mark out of 100. A score out of 100 implies there is a known worst case, and there isn't one. Conditions can always get tighter than they have ever been before.
On 28 February 2026, the day the strait closed, this index printed −0.33 (on the current food source; also −0.33 before it changed). Brent was still about $71 and barely above its own normal. The reading was below zero mainly because the dollar's twelve-month pace sat well under its own average, and consumer prices and food prices were both slightly soft. The war lives on the Hormuz card. This number is the background average, and on that day the background had not moved.
On 2 April 2026, with Brent at $127.61, the index printed +1.09 (it read +1.18 before the food source changed on 28 September 2026). Two of the five inputs were beyond the cap that day: oil and conflict attention. Consumer prices, food prices and the dollar's twelve-month pace were not, and the dollar was pulling the other way. A reading of +2 or beyond means pressure seen only a handful of times in a generation. The 2026 war never reached it on this gauge, because two loud legs averaged against three quiet ones is what this recipe is built to produce. That is the design working, not a miss.
A reading near zero with one loud leg means the average is still normal and one channel is not. Read both.
Every change, dated. Each change to how this index is built has its own entry in the Strain Index changelog: what changed, when, why, and how many figures were first shown under it.
How it is built. We take five things that get measured constantly, and for each one we ask the same question. Compared with its own history, is today's figure unusually high or unusually low? Then we average those five answers into a single number.
Until this date the two household-cost measures entered as price levels rather than as rates of rise. That was a mistake, and worth explaining plainly because it changed the published number.
A price level almost never falls. It climbs year after year, so comparing today's level against its own past will always say today is high, and will say so more emphatically every year that passes, whatever is actually happening to inflation. On 25 August 2026 the consumer price input sat so far above its own past that it made up forty-four percent of the entire published figure, while carrying no information about whether prices were rising quickly or slowly. It was measuring the passage of time.
Both now enter as the change over twelve months, which is what people mean when they talk about inflation, and which can fall as well as rise. This lowered the published figure, and that drop reflects the correction rather than any change in the world. Readings published before this date came from the older definition and are not directly comparable with readings after it.
Until this date the food input was the International Monetary Fund’s food price index, which we received through the Federal Reserve’s data service. It arrived there 17 to 24 days after each month ended, and on 27 September August’s figure had still not appeared. The food input is now the FAO Food Price Index, read directly from the UN Food and Agriculture Organization. FAO publishes on a fixed calendar in the first week of each month: August 2026 came out on 4 September.
It enters exactly as before, as the change over twelve months, measured against its own history back to 1991. Only the food input changed. We deliberately left the other four inputs’ history untouched, so every difference below comes from the food source alone.
One trap we found and avoided. The download address that search engines return for FAO’s index still works, but it serves a file last updated in May 2018 that ends in April 2018. Built on that, the index would have carried 2018 food prices under a 2026 date without any error. We read the address FAO’s own page links to, and the index refuses any food file whose newest month is more than a hundred days old.
| IMF food (to 27 Sep) | FAO food (at the switch, 28 Sep) | |
|---|---|---|
| 18 September 2026 | +0.69 | +0.66 |
| 2 April 2026 (the war) | +1.18 | +1.09 |
| 28 February 2026 (the strait closed) | −0.33 | −0.33 |
| Peak (8 March 2022, Ukraine) | +2.264 | +2.203 |
| Trough | −0.871 (30 Apr 2020) | −0.766 (10 May 2020) |
Computed across all 4,278 days before the switch. The average day moved by 0.06; the largest move was 0.23. Readings published before 28 September 2026 used the IMF source and are kept as they were first shown.
Until this date the fifth market print was the US natural gas price (Henry Hub). That is a regional American fuel price, driven by American weather and American storage. It moved a great deal from one day to the next and told us very little about the world.
Measured across the eleven years of reconstructed history, it accounted for roughly seventy per cent of the index's day-to-day movement and under a fifth of the level — the loudest of the five inputs and the least informative. During the March 2022 Ukraine shock it stood about one ordinary swing above its own normal, while oil, food and conflict coverage were all between two and three. It was elevated. It was simply far less elevated than everything else, and averaging it in pulled the headline down.
It is replaced by the Federal Reserve's emerging-market dollar index (DTWEXEMEGS), which tracks the US dollar against the currencies of emerging-market economies. When the dollar strengthens against those currencies, every country and company that borrowed in dollars finds its debts harder to service and its imports dearer, food and fuel among them. That is a channel through which pressure reaches most of the world at once. A regional American gas price is not.
It enters as the change over twelve months, not as the level, and the reason is the same one set out in the panel above. The dollar index does not swing around a settled centre; it climbs. Measured against its own 2015–2026 average it would have scored about −2.6 in early 2015 and about +2.1 in early 2025 — a swing of more than four and a half ordinary swings, produced by nothing except the passage of ten years. Read as a rate of change instead, it answers the question that matters: is the dollar rising against these currencies unusually fast right now? What strains a borrower is the currency moving against them, not a level they have had years to arrange their affairs around.
The predecessor series is not deleted. It is frozen, 4,261 days, under the table name dgsi_daily_v1_archive. Anyone who cited the old March 2022 peak of +2.55 can still find that number, and can see why the live series now prints a different one.
| Old five | Four, gas removed | New five (to 27 Sep) | |
|---|---|---|---|
| Peak (March 2022, Ukraine) | +2.550 | +2.857 | +2.264 |
| Trough | −1.392 | −1.547 | −0.871 |
Middle column: the archived values with gas removed and nothing put in its place. Right column: the series rebuilt 4 September 2026 — peak 8 March 2022, trough 30 April 2020. The food source changed on 28 September 2026; see the panel above for the current figures.
We expected the Ukraine peak to rise, and it fell. Removing gas alone lifts it to +2.857, which is why the middle column exists. Adding the dollar brings it back down to +2.264, below the old +2.55, because the dollar was close to its own normal in March 2022 while the other four were between two and three. An earlier draft of this page carried +2.857 as the new figure. That was the middle column mislabelled, and it was wrong.
The trough is the larger change, and it is the clearer argument for the swap. The old series bottomed at −1.392 on 30 April 2020. US gas had collapsed in the pandemic demand shock, and as a raw price level it dragged the index to an extreme the other four inputs never supported. A third of that reading was American gas storage. The new series bottoms at −0.871 on the same day — still the calmest moment in eleven years, but no longer overstated by one regional fuel price.
| What we look at | Where it comes from | Updated | Share |
|---|---|---|---|
| Oil price (Brent crude) | US Energy Information Administration | Weekly batch | One fifth |
| How fast US consumer prices are rising | US Bureau of Labor Statistics, via Federal Reserve economic data (FRED) | Monthly | One fifth |
| How fast world food prices are rising | FAO Food Price Index, read from FAO (since 28 September 2026; International Monetary Fund food index via FRED before) | Monthly | One fifth |
| US dollar against emerging-market currencies | Federal Reserve Board, H.10 release | Weekly batch | One fifth |
| Conflict in the news | Caldara & Iacoviello Geopolitical Risk index (two Federal Reserve Board economists; not an official Fed release) | Weekly | One fifth |
One input is American, not global. The consumer-price input is the United States’ own figure, published by the Bureau of Labor Statistics. The page labels it “US consumer prices” wherever it appears, so a global index is never read as carrying a world inflation figure it does not have.
No single one of the five can dominate. Each is capped, so even an extraordinary move in one input cannot swamp the other four.
Each input is capped before it is averaged. When that cap binds, that component is marked as capped and the average uses the capped figure, while the page still shows the uncapped one. That is how an extraordinary print stays visible without owning the headline. On 2 April 2026 it bound twice: oil's uncapped reading was 3.07 and conflict attention's was 4.43, both entering the average at 3.00. Most weeks it binds on nothing at all.
The headline is the average. The page also names the loudest input, and the input that is holding the completion date. Those two facts are how you read a week when oil has moved and food has not.
“Weekly batch” means a daily reading that arrives once a week. The oil price and the dollar index are recorded every business day, but the bodies that publish them release those days in a batch, once a week. The conflict measure is published on a Monday. Two of the five change once a month. So the date a figure carries and the date we can first see it are different dates, and none of the five arrives every day. That is why this index completes weekly rather than daily, and why we say what it is complete through rather than claiming it is current.
We let a late monthly figure go unmarked for weeks, and we have fixed that. A monthly figure stands for its whole month, and we keep using it until the next month's figure is published. We had no rule for when that next figure was late. On 28 September 2026 the index was still using July's world food figure from the International Monetary Fund, with August's still not published, and the page listed it as an ordinary carried input. (The same day we moved the food input to FAO, which had published August on 4 September. That is the next panel.) The rule now: each month's figure is due by the 25th of the month after it. If one is late, the page says which month is missing and the index keeps using the last figure, because one slow publisher should not switch the whole index off. If two are late, we stop publishing new days until the figure arrives. We do not quietly average the other four instead. We first set that at the 20th, as a guess. The releases we could date say it was too early: the food index arrived 24 days after February ended and 17 days after July, so the 20th would have called February late when it was not. The 25th covers both. Two releases is not much evidence, so we are recording when each figure actually arrives, and the date will be reset from a full year of that record.
We compare each thing against the longest run of trustworthy history we have for it, rather than forcing them all into the same window. For Brent oil that means back to 2015, which is as far as the series we use will go, and which is therefore where the composite itself can start. For how fast the dollar is moving it means back to 2007. The index itself starts in 2006, and a twelve-month change needs a year of history before its first reading. For how fast consumer prices are rising it means back to 1983, and for food prices back to 1991: FAO’s index starts in 1990, and needs a year before its first twelve-month change. For the conflict measure it means back to 1985.
Judging how fast prices are rising against only the last ten years would leave out every earlier bout of inflation, and a decade that was mostly quiet makes any later flare-up look more extreme than it deserves. We start in 1983 rather than going further back because the way housing costs are counted changed that year, and the measure before then is not the same measure.
One honest limitation. Each monthly reading of how fast prices are rising compares that month with the same month a year earlier, so consecutive readings overlap by eleven months and are not independent of one another. Forty years of monthly readings is nowhere near forty years’ worth of separate evidence. It is closer to a dozen distinct episodes of prices speeding up and slowing down. That makes the normal range for these two inputs the least precisely known of the five, which is another reason we use as much history as we can get.
The reason matters. Judging today's conflict level against only the last ten years would leave out the September 11 attacks, the 2003 invasion of Iraq and the Gulf War. Those are the events that tell you what a genuinely high reading looks like. Leave them out and the recent past looks calmer than it really is, which makes every new flare-up appear more extreme than it deserves. We would rather each input be judged against everything we reliably know about it, and say plainly that the windows differ, than use one short window that flatters the numbers.
It measures pressure, not danger. Four of the five inputs are market prints that have already been recorded: an oil price, and three rates of change — consumer prices, world food prices, and the dollar against emerging-market currencies. The fifth counts how much of the world's news is about conflict. Together they describe the strain that exists today.
It does not predict. Nothing in it looks forward. The index describes conditions now. It does not tell you what happens next, and we will never claim it does.
Parts of it move slowly, and the number is not daily. Two of the five are published once a month; the other three arrive weekly. A new reading appears when the slowest of them has landed, not every morning. Between those, the figure on the page does not change, and that is the index behaving correctly rather than failing.
It is a strain measure, not a conflict ticker. Four of the five are financial and price measures. One counts conflict coverage, and it is the slowest-moving of them. If you want a fast read on a specific unfolding situation, that is what the individual risk structures on this site are for. This number describes background pressure, and it changes on the calendar its sources keep.
Where the conflict measure comes from, and why it is not ours. We spent weeks building our own. It counted how many news articles mentioned conflict, and we tested it before trusting it. Those tests taught us two things, and both of them ended the project.
We tracked news coverage of the Ukraine invasion from February to December 2022. Coverage peaked at 32.7% of all news being monitored, then fell to 7.6% by the end of that period, under a quarter of its peak. The war was still fully underway the whole time.
So a measure built on how much coverage a war is getting would have reported that Ukraine risk had mostly gone away by the end of 2022, when nothing on the ground had improved. The world stops reporting on a war long before the war stops.
The second problem was ours alone. Our version searched for words like Hormuz, Iran and oil tanker. That is a list about one waterway, not about the world. It would have gone quiet during a crisis in Taiwan or the Red Sea, and it would have counted the Gulf twice over, because trouble in the Gulf already reaches this index through the oil price. An index called Global cannot be built on a search for one strait.
So we use an established measure instead. The Geopolitical Risk index is built by two economists at the Federal Reserve Board, Dario Caldara and Matteo Iacoviello. It reads ten major newspapers going back to 1985 and counts articles across eight kinds of trouble, including threats of war, military build-ups, nuclear threats and acts of terrorism. It is published openly, free for anyone to use, and the method behind it was reviewed and published in one of the leading economics journals.
It counts shares, not totals. The world publishes far more news than it did in 1985. A measure that simply counted conflict articles would show the world getting steadily more dangerous purely because more news exists. This one reports conflict coverage as a proportion of all coverage, which cancels that out.
We use its thirty-day average, not its daily figure. The daily figure can sit seventeen percent away from its own monthly average, which would let a single news cycle visibly swing our headline number. The rest of the index moves slowly, and we would rather the conflict input moved at a comparable pace than have one input drive nearly all the movement on its own.
Source: Caldara, Dario and Matteo Iacoviello (2022), “Measuring Geopolitical Risk”, American Economic Review 112(4), pp. 1194–1225. Data published under a Creative Commons licence and downloaded from matteoiacoviello.com.
If any input is missing, we publish nothing. The index is not quietly recalculated around a gap. If one of the five cannot be read on a given day, the number is withheld and the space stays empty rather than showing a figure built on four inputs while describing itself as built on five.
Out of date counts as missing, and the conflict file sets the date. The conflict measure is published by someone else, roughly once a week. It is one of the inputs that bound the completion date of the print. The date on the headline and the date the chart line ends are the same date: the oldest complete observation among those bounding inputs. There is no exemption and no carrying a stale conflict reading forward so the page can look current. If the Monday file slips a holiday, the index goes quiet instead of showing a current-looking headline over a stale conflict leg.
If the conflict reading behind a given day is more than seven days old, that day is never written at all. The check happens where the series is built, not where it is displayed, so a reading assembled around a stale conflict input cannot exist to be shown — on this page, in an export, or anywhere added later. Seven days is roughly one missed weekly update: enough to ride out a public holiday, not enough to let a frozen conflict leg sit through a real move in the other four.
The first print is kept. The first time a day's reading is served on the site, that figure is recorded and never rewritten — not by a nightly rebuild, not by a change of method. The headline you see is the current reading, rebuilt against the current baseline; the first print is what the same day said when it was published. When the two differ, the page says so underneath the number rather than quietly replacing it. Days that were never served, including the whole reconstructed history before this record began, carry no first print, because nobody was shown one.
A restated history is never silent. Changing an input definition rebuilds the baseline first. Reconstructing the daily series against an older baseline is refused, because that would draw a new line and call it the old one. Charts drawn under the current definition are not joined, without a seam, to charts drawn under the previous one.
The headline number stays free, permanently. The most respected gauges of this kind, including the Chicago Federal Reserve's financial conditions index and the Geopolitical Risk index we now draw on, are all free to look up, and that is exactly why people trust and cite them. A paid tier adds the breakdown by component, the full history, and alerts when the number crosses a threshold.
Free to cite. Licensed to reuse. Quote, chart or report any DGSI reading, in the news, in research or in your own notes, with the credit “Source: DELAX Global Strain Index (DGSI), delaxcom.org”. Bulk downloads, data feeds, and use inside a product, fund, index or trading model need a licence from us. This page explains how the index is built so that you can check it. It does not give anyone the right to publish their own index under the DGSI or DELAX name. The DGSI and its published values are © 2026 DelaxCom LLC. The five inputs remain their publishers’, as credited above. Full terms are in our terms of use.
Open your holdings in the Workspace and enter what you own. You will get a figure for how your particular mix would be affected, broken down by which route the damage travels along and which holding is hit hardest. Everything is calculated on your own device.
On the dashboard, switch between Hormuz and Taiwan with the same holdings loaded. An oil crisis barely touches a technology-heavy portfolio; a Taiwan blockade transforms it. Then switch between the outcomes. Each one shows how likely we think it is, in numbers and in plain words.
Ask DELAX, the speech bubble on the dashboard and the Workspace, takes plain questions like “how exposed am I if Taiwan is blockaded?”. Here is how it answers. Our server first gathers the facts the question needs from this model: each crisis’s odds and trust label, each scenario’s figures with their sources, your book’s modelled moves, the latest published DGSI reading. Each fact carries its date and where it came from. An AI writes the answer from those facts only. Before you see it, we check it: every figure must match a fact we gave it, it must show which facts it used, it must name how sure the model is, and it must not tell you to buy, sell or do anything with your money. An answer that fails is held back and you are told why. The same facts and checks now stand behind the dashboard’s other AI explanations. Your questions are not stored.
Every figure carries a label. MEASURED means it came from a real event. ESTIMATED means it is reasoned, because the event has never happened. On the Taiwan view you can see every past scare we tested and what each one showed. Beside each label is the date we last reviewed it and when it is next due.
The 2026 Hormuz war lasted six weeks. A closure lasting a year could damage company profits in ways a six-week shock never did. Our measurements describe what has happened, not every possibility that hasn't.
There is no free, constantly-updated source for shipping rates, real-time global economic output, or defence budgets. Where we have had to estimate, we say so on the figure itself. Our weakest area is the effect on developing-country currencies: in both wars we measured, gains and losses largely cancelled out. One figure broke this rule until September 2026: the inflation forecast chart credited “IMF + BIS inflation transmission models” it was never built from. Its numbers are our judgement, and the chart now says so, with the IMF’s own rule of thumb (a 10% oil rise held for about a year adds roughly 0.4 points to global inflation) shown for scale.
The Taiwan figures describe something that has never occurred. They are built for comparing exposure and thinking through consequences, not for precision. Treat the extreme scenarios as rough magnitudes.
We measure and explain exposure. We do not tell you what to buy or sell, and nothing here should be read as a personal recommendation. If you want advice about your own situation, speak to someone qualified and regulated to give it.
A number you cannot question is a number you are taking on faith. Everything above can be checked: the events are public, the investments are widely traded, the dates are stated, the outside research is named, and the core method is one subtraction, namely what this investment did minus what the whole market did. Where we have been wrong, the correction is written down. If you think one of our figures is still wrong, this page gives you what you need to make that argument.