Methodology · version 1.0.0
Every number, traceable to a row and a formula
Autonaly's exposure scores are deterministic: the same inputs always produce the same outputs, no model in the loop. This page is the whole method — including what it deliberately cannot see.
The formula
- DDR — direct dependency ratio. The share of a country's imports of a commodity basket sourced from the disrupted origin, computed from bilateral HS6 customs flows (CEPII BACI).
- HHI — supplier concentration. The Herfindahl–Hirschman index of the importer's supplier mix for that basket, recomputed at basket level. Concentrated dependence scores higher than the same ratio spread across many suppliers.
- Essentiality. A fixed weight per basket — staples, energy and critical minerals carry more than substitutable goods. Published in the basket catalogue; never tuned per event.
- Severity. Transit reduction × a duration factor that saturates at six months. For chokepoints, the reduction is measured from IMF PortWatch vessel transits — the observed collapse, not the headline's claim — and attenuated when a sea bypass exists.
Glossary
Every figure the product shows, and what it does and does not let you conclude. The same definitions travel inside exported documents.
Score
Exposure intensity on a 0–100 scale: 100 × dependency × concentration × essentiality × severity. Because it multiplies four factors that are each at most 1, real scores sit well below 100 — a score of 10 does not mean '10% affected', it means this country ranks at that intensity relative to the others in the same run. Scores are comparable within one scenario and not between scenarios, because severity differs.
Dependency (DDR)
Direct dependency ratio: of everything this country imports in the affected commodity basket, the share sourced from the disrupted origins. Computed from bilateral HS6 customs flows. 51% means half of that basket's imports are exposed.
Concentration (HHI)
Herfindahl–Hirschman index of the importer's supplier mix for the basket, from 0 (many equal suppliers) to 1 (a single supplier). It scales the score between one half and one — never to zero, because a diversified buyer still suffers when a supplier stops; it simply has somewhere to turn.
Value at risk
The dollar value of imports from the disrupted origins, scaled by severity. This is magnitude, not intensity: a large economy can carry the biggest dollar exposure while barely registering on Score, and a small one can be devastated proportionally while losing far less money. Both orderings are reported because reading only one produces a misleading brief.
% of GDP
The value at risk expressed as a share of the country's GDP (World Bank, latest year). Two figures already published, divided — and the one that turns a dollar amount into a judgement: $7.6bn is 1.96% of Egypt's economy and a rounding error against Germany's. It sizes the shock; it does not predict the response.
Largest absolute
The single country with the greatest dollars at risk. Marked on the map with a bright outline. Frequently a large economy whose Score is low.
Most dependent
The top of the intensity ranking — the country for which the disrupted origins represent the largest share of supply. Often small, and often invisible in a dollar ranking.
Severity
How much supply stops and for how long. Duration saturates at six months, because buyers re-contract and a twelve-month disruption is not twice as bad as a six-month one. Where a sea bypass exists the score is attenuated, so a closure becomes a cost-and-delay shock rather than a cutoff. For real events severity is measured from observed vessel transits, never assumed.
Computed vs curated
Every briefing carries one of these. Computed means the deterministic engine produced rankings: countries, dependency ratios, dollars at risk. Curated means it did not, and the briefing carries the reason instead of a number. That happens when severity cannot be established honestly — a vessel-transit baseline that is itself degraded, or an event whose mechanism is invisible in customs data, such as a banking crisis. A curated briefing is not a failed one: the written analysis is there, and so is a statement of what would change the answer. It is the desk declining to publish a figure it cannot defend.
Essentiality
A published weight per commodity group: staples and energy carry the most, then fertilizers, critical minerals, and industrial goods. Fixed in advance and never tuned per event, so it cannot be adjusted to produce a desired ranking.
Materiality floors
A ranking answers "where is the global supply impact", not "who is most dependent". Importers below a floor — the greater of $100m or 5 basis points of the basket's world trade — are excluded, so micro-importers at 100% dependency do not outrank major economies at 70%. Value at risk in dollars is always shown beside the intensity score, because the two orderings answer different questions.
The provenance guard
Narratives are drafted by Gemini around the computed numbers, then checked mechanically: any figure in the text that does not exist in the engine's output rejects the draft. The model can phrase; it cannot invent. Briefings the desk cannot score honestly — for example when AIS data looks degraded — are filed unscored with the reason quoted, and a human approves everything before publication.
What the model deliberately cannot see
Latest-year trade weights
Exposure uses the most recent full year of bilateral customs data (BACI 2024). It measures dependence on today's network, not how trade would rewire under stress. Post-2022 rewiring — such as Europe's pivot from Russian seaborne energy — is already embedded in the weights.
First-order effects only
A country is exposed through its own imports from the disrupted origin. Second-order chains (neon → fabs → autos) are not computed; where they matter historically, they are curated narrative, clearly labelled.
No inventories, no substitution dynamics
The model assumes no stockpiles, no domestic production ramp, and no supplier switching within the scenario window. Real adjustment makes real outcomes smaller than modelled exposure — the score is a ceiling on first-order exposure, not a prediction.
No probabilities, ever
The desk publishes exposure, never likelihood. Hypothetical scenarios open with the word 'Hypothetical' and refuse probability by design; historical probability belongs to the record, not the model.
Only 24 commodity baskets are visible
Grains, energy, fertilizers, and critical minerals — 72 HS6 codes chosen for supply-shock relevance. Services, autos, machinery and most manufactures are outside the model, so a crisis in a diversified manufacturing economy is understated here.
Financial crises are refused
Banking, currency and debt crises transmit through capital flows and confidence — invisible in customs data. The engine will not score them; the desk routes them to a curated, unscored briefing that says why.
Data and attribution
- Trade flows: CEPII BACI (Etalab 2.0) — 11.25m bilateral HS6 rows, 2024.
- Chokepoint transits and port shares: UN Global Platform; IMF PortWatch.
- Country context: World Bank WDI (CC BY 4.0); currency names via CLDR.
- Crisis history: curated from the historical record, reviewed in version control — never generated.
Informational, not advice. Exposure, never probability. About Autonaly →