Methodology

What the model does.
How we know it works.

Written for risk professionals who want to understand what is under the hood before a conversation. This document covers model architecture, outputs, validation, and the scope of what the model does and does not do.

Risk corridors modelled 22corridors
Trade value in scope $1.8TUSD
Coverage rate 89.6%
Importing countries 85countries
Commodity classes 7classes
Model version 0.1May 2026
What the model is

A bilateral trade exposure model. Not a monitoring tool.

The Narrows model quantifies the financial exposure of importing countries to disruptions in specific trade corridors. It answers a different question from AIS-based monitoring tools, which measure what is happening to throughput. Narrows measures what a disruption costs — in dollars, by country, by commodity, under a scenario you define.

The model is built on validated bilateral trade flow data matched to corridor-specific dependency scores. It produces three structured outputs: value at risk, rerouting penalty, and fleet at risk. Every parameter the model uses is substitutable with your own assumptions.

The model produces dependency scores and exposure estimates. It is an input to your analysis — not a replacement for it.
How it works

Five stages. Each produces a discrete, auditable output.

The model runs in five sequential stages. The transformation at each stage is where the analytical value sits — not in the underlying data, which is available to anyone, but in how it is structured, matched, and scored.

01
Flow ingestion
Bilateral trade flow data for 7 commodity classes is ingested, cleaned, and standardised across reporter countries and reference years.
02
Corridor assignment
Each bilateral flow is assigned to its most probable maritime corridor using a proprietary routing model. Unmatched or ambiguous flows are flagged and excluded.
03
Dependency scoring
Country-corridor-commodity dependency scores are calculated for 85 importing countries. Each score states the share of a country's import of a given commodity that transits a given corridor.
04
Scenario simulation
A user-defined disruption — corridor, severity, and duration — is applied to the dependency matrix. Exposed trade value is calculated against the relevant time window.
05
Output generation
Three structured outputs: value at risk by country and commodity, rerouting penalty by vessel class, and fleet exposure estimate for the disruption window.
What it produces

Three outputs. Each answers a specific professional question.

Every output connects to a downstream use case. The model is designed to produce numbers that slot into existing analytical workflows — accumulation models, trading desk pricing, credit stress tests — not numbers that require translation before they can be used.

Output 01
Value at risk

Dollar-denominated trade exposure by country and commodity under a user-defined disruption. Severity (0–100%) and duration are variable. Capacity sensitivity is adjustable per corridor.

Malacca 100% / 14 days Japan: $1.6B iron ore · South Korea: $1.1B · China: $3.8B crude
Output 02
Rerouting penalty

Per-voyage cost overrun by vessel class: detour days, bunker uplift, charter rate differential, and inventory carrying cost. Tagged by audience — underwriter, lender, cargo owner. All rate assumptions substitutable.

Suez closure — VLCC via Cape +14 days · $1.4M voyage overrun · $230k carrying cost per vessel
Output 03
Fleet at risk

Estimated vessel count in-window at event onset, cargo value on water, and portfolio-level rerouting loss across the fleet. The accumulation number for P&I and cargo underwriters.

Hormuz 100% / 30 days 28 VLCCs in-window · $3.4B cargo at risk · $28M portfolio overrun
Sample output

Strait of Malacca — 100% closure, 14 days.

The following is a live model output, not a hypothetical illustration. Parameters shown are the defaults for this scenario. All can be substituted in a demo session.

Corridor
Strait of Malacca
Severity
100%
Duration
14 days
Capacity sensitivity
0.85
Base data
2023
Total trade value at risk
$62.4BUSD
All matched commodity corridors transiting Malacca, 14d at 100% closure
Countries with material exposure
18nations
>5% import dependency threshold. Includes China, Japan, South Korea, India, Taiwan
Rerouting delay — Sunda Strait
+4days avg
Partial alternative only. Depth-limited for laden VLCCs and Capesizes
CII-adjusted Capesize system cost
$21M/yr
IMO CII speed constraints add ~0.7 days to Lombok detour. ~600 Japan voyages/yr

Country-level exposure — iron ore

Dependency percentage is a fixed output of the dependency matrix. Duration and severity are the variables you control in the simulator.

China
68% via Malacca
$19.2B
Japan
82% via Malacca
$4.8B
South Korea
78% via Malacca
$3.1B
India
41% via Malacca
$1.4B
Taiwan
74% via Malacca
$0.9B

Values are 14-day simulation outputs from 2023 base data. Dependency percentages are model-derived from bilateral trade flow analysis.

Validation

Route inference outputs are validated against independent aggregated datasets.

The model's corridor assignment logic is validated by checking whether predicted chokepoint shares are directionally consistent with observed vessel traffic distributions from independent sources. Validation is described as directional rather than exact — trade data and vessel traffic data do not map to the same time period or granularity.

Source 01
Port call and transit counts

Weekly vessel call counts at major ports and chokepoint transit volumes derived from aggregated AIS data. Used to validate total traffic distribution across corridors.

Source 02
Energy corridor flows

Weekly and monthly aggregated tanker flow data from official energy statistics, covering Hormuz, Bab el-Mandeb, Suez, and Malacca. Primary validation for energy commodity assignments.

Source 03
Annual shipping statistics

Annual statistics on container flows and bulk cargo through major corridors from international shipping bodies. Used for annual cross-check of overall corridor share estimates.

Scope and limitations

What the model is, and is not.

These limitations are stated because the model is intended to support serious risk analysis. A professional who understands where the model is uncertain is better positioned to use it appropriately — and to substitute their own assumptions where they disagree with ours.

Exposure estimates, not valuations. The model produces dollar-denominated exposure estimates and dependency scores. It is not an insurance valuation tool and does not produce premium recommendations, loss estimates, or probability-of-disruption assessments. Those remain with the analyst.
Annual base data, not live flows. Trade flow data reflects annual bilateral patterns from the most recent complete reference year. The model measures structural dependency, not current cargo positions. Voyage count estimates are model-derived decompositions, not live vessel tracking.
No seasonal adjustment. Duration-based calculations use linear scaling rather than seasonal adjustment. The model does not capture monthly variation in commodity flows or periodic vessel traffic patterns.
Country-level, not vessel-level. The model does not track individual vessels, flag states, or policy-specific cargo. Portfolio accumulation analysis requires mapping individual policies to corridor assignments — a data integration step that sits outside the model.
Capacity sensitivity scores are calibrated parameters. The 0–1 sensitivity scores applied per corridor are informed judgements calibrated against historical disruption events. They are defensible and adjustable — not statistical estimates with confidence intervals. Commercial clients may substitute their own scores.
Coverage gaps at thin bilateral flows. Some commodity-corridor pairs with limited bilateral data coverage are excluded from the dependency matrix. The 89.6% coverage rate reflects value-weighted coverage — smaller flows by value are disproportionately underrepresented.
Land-based and pipeline flows excluded. The model does not distinguish transport mode within bilateral trade data. Flows that are primarily overland or pipeline-routed are excluded where corridor assignment is ambiguous, but the exclusion is not exhaustive.

Full documentation is available
under NDA for serious counterparties.

This page describes the model at a level sufficient to evaluate its credibility and fit. Full documentation — including corridor definitions, the complete dependency matrix, simulation API specification, and integration notes — is available on request to qualified counterparties under a standard NDA. Request a demo and we will send the documentation pack before the call.

fysh@narrows.io · A person will reply.