Marine war risk is one of the few exposures where the unit of risk is not a policy, an obligor or a location, but a voyage. A vessel that is entirely unremarkable on Monday becomes a concentrated peak exposure on Wednesday because its routing takes it through a narrow, contested waterway for a handful of hours. The Strait of Hormuz and the southern Red Sea / Bab el-Mandeb corridor are the canonical examples: geographically tiny, economically enormous, and capable of turning a diversified marine book into a single-event book without a single new policy being written.
This article sets out how that exposure should be measured and carried through to capital. It is a modelling argument, not a market commentary: it deliberately avoids quoting current premium levels, cancellation notices or live rating movements, all of which change faster than any published note can track and none of which are necessary to get the model structure right.
The unit of exposure is the transit, not the policy
A conventional property or casualty exposure model asks what is insured and where is it. A marine war model has to ask where will the insured thing be, and for how long. Three consequences follow immediately.
First, exposure is time-weighted. A vessel's contribution to a chokepoint scenario depends on hours inside the defined area, not on its annual premium. Two hulls of identical value contribute very differently if one transits weekly and the other transited once.
Second, exposure is stochastic in a way the underwriter does not control. Routing decisions are made by the assured, often at short notice, and often in response to the same threat environment the insurer is trying to price. Diversion around the Cape rather than through Bab el-Mandeb removes war exposure and adds voyage duration, fuel and delay exposure — a substitution effect, not an elimination.
Third, the exposure is declared rather than continuously observed in many structures. The modelling implication is that a war book carries a data-latency risk on top of its underwriting risk, and any accumulation figure should be reported with an explicit statement of how stale the underlying transit data is.
Hull, cargo and P&I are three different risks wearing one label
"War risk" is a peril, not a line of business, and the three principal covers behave differently enough that modelling them jointly without decomposition is a mistake.
Hull war attaches to the vessel itself. Severity is bounded above by the insured value, the distribution is strongly bimodal (light damage or total loss with little in between for weapon-type events), and the exposure is single-asset and unambiguous. This is the most tractable of the three.
Cargo war attaches to goods. The insured value on a single hull can be a large multiple of the hull value, ownership is fragmented across many assureds and many policies, and the same physical event therefore triggers a large number of separate claims. Cargo is where the accumulation problem is at its worst, because the insurer's own systems often cannot see, at bind time, that two hundred certificates will end up on the same deck.
P&I and liability attaches to consequences: crew injury and death, wreck removal, pollution, and third-party liabilities arising from a blocked or damaged waterway. Severity here is not bounded by an insured value in any natural way; it is bounded by policy limits and pooling arrangements. The tail is longer, the development is slower, and the correlation with hull events is high but not one-to-one — a near-miss that injures crew is a P&I loss with no hull loss at all.
Any capital model that treats these as one lognormal with a shared severity curve will understate cargo frequency-per-event and understate P&I tail duration simultaneously.
Accumulation control: the real modelling problem
The failure mode in marine war is not mispricing an individual transit. It is discovering, after an event, that the book's true exposure to a single incident was several multiples of the figure being monitored.
A workable accumulation framework needs at minimum:
- A geographic definition of each accumulation zone that is stable, auditable and consistent with the areas used in the underlying contract wordings. Model zones that do not line up with wording zones produce numbers no one can reconcile after a loss.
- A time-slicing rule — the maximum exposure over any rolling window, not the average. Peak co-presence is what matters, and an average will hide it.
- A per-event loss definition that spans hull, cargo and P&I on the same incident, including cargo written through delegated or facility business.
- A blockage or closure scenario in addition to a per-vessel scenario. A waterway that becomes impassable generates trapped-vessel exposures, delay and consequential claims across a very large number of risks that were never simultaneously "in" the zone in the transit sense.
The blockage scenario is the one most often missing. It is structurally different: exposure is defined by vessels unable to leave rather than vessels choosing to enter, and the loss driver is duration rather than intensity.
Stress testing without inventing numbers
The right output of a marine war stress test is not a point estimate of loss. It is a small set of internally coherent scenarios with explicit, documented assumptions, each carried consistently through gross loss, reinsurance recoveries, net loss, liquidity and capital.
A defensible scenario set separates the drivers that actually vary independently:
- Event type — single-vessel attack, multi-vessel campaign over weeks, or waterway closure.
- Duration — how long elevated conditions persist, which drives the number of exposed transits and, for closure, the trapped-asset count.
- Behavioural response — the diversion rate. This is the single most important assumption and the one most often left implicit. A high diversion rate truncates the loss but shifts exposure to delay, cargo condition and charter-party disputes.
- Recovery and dispute friction — the proportion of claims contested on war-versus-marine causation, which drives reserve development and expense, not just quantum.
Each of these should be a stated parameter with a stated range, not a number buried in a spreadsheet. The general discipline is the same as that described in Beyond Historical VaR and in the state-transition treatment in Geopolitical and War Risk: A Quantitative Modelling Framework: the model's job is to make the judgement auditable, not to replace it.
Frequency and severity when the history is thin
Marine war events are rare, clustered and non-stationary. Fitting a frequency distribution to a long historical series implicitly assumes the future threat environment resembles the average of the past, which is precisely the assumption a chokepoint crisis violates.
Two adjustments are standard and defensible. The first is conditioning frequency on a stated threat state rather than modelling it unconditionally, so the parameter changes explicitly when the assessed environment changes rather than drifting silently. The second is separating attack frequency from success frequency: the number of incidents and the proportion producing an insured loss are driven by different mechanisms, and collapsing them into one rate destroys the ability to reason about either.
On severity, the practical constraint is that hull severity is bounded and cargo severity is aggregation-driven, so a single fitted curve is the wrong object. The methods in Frequency–Severity Modelling apply, with the caveat that the exposure base must be transits rather than policy-years. Where parameter uncertainty is material — and here it always is — a Bayesian treatment with an honest prior, as discussed in Bayesian Climate Risk Modelling, communicates the uncertainty better than a point estimate with a footnote.
Reinsurance structure and the capacity question
Marine war is a market that manages severity through structure rather than through diversification, because the diversification is not available: the peril concentrates by design.
On 19 June 2026, Lloyd's announced a Chubb-led marine war-risk consortium providing up to USD 200m of capacity separately for hull and P&I risks, together with a dedicated USD 200m of cargo capacity, subject to underwriting criteria, sanctions and regulation (Lloyd's press release, 19 June 2026). Two features of that structure are directly relevant to modelling and should be read alongside Lloyd's own market guidance rather than inferred from market commentary.
The separation of cargo capacity from hull and P&I capacity reflects exactly the decomposition argued above: cargo accumulation on a single hull is a distinct problem from hull loss, and pooling them into one limit would let a single event consume capacity intended for a different exposure. The explicit conditionality — underwriting criteria, sanctions and regulation — is also modelling-relevant, because it means available capacity is state-dependent. A capital model that treats a consortium limit as unconditionally available on the day of loss overstates recoveries.
The general lesson generalises beyond this specific structure: where reinsurance recovery is conditional on eligibility, sanctions screening or notice provisions, the recovery should be modelled with a haircut reflecting the probability that conditions are not met, and that haircut should be documented and challenged rather than set to zero by default.
Carrying the result into capital
The final step is the one most often done informally. Marine war losses reach the balance sheet through several routes at once, and each needs its own treatment:
- Underwriting risk capital — the net-of-reinsurance loss distribution for the war peril, with its own correlation assumption against the rest of the marine book and against wider catastrophe exposures. Independence is not a safe default; a closure event correlates with commodity, credit and market stress.
- Counterparty default risk — reinsurance recoveries concentrated in a small number of counterparties on a peril that stresses those same counterparties.
- Liquidity — large gross claims are paid before recoveries are collected, and P&I development is slow. The gross-to-net timing gap is a funding requirement, not an accounting one, and belongs in the framework described in Liquidity Risk: LCR, NSFR and ILAAP Stress Testing.
- Reserve risk — causation disputes and slow-developing liability claims mean the reserve is uncertain long after the event window closes.
For insurers operating under Solvency II, the natural home for the scenario set is the ORSA rather than the standard formula alone, since the standard formula's marine module is not designed to capture a correlated chokepoint closure. The treatment of scenario-driven capital add-ons in Solvency II, ORSA and Economic Capital applies directly, and the governance expectations for scenario-based models are those set out in Model Validation Lifecycle and Independent Challenge.
What good looks like
A marine war framework is in reasonable shape when four things are true. Accumulation is measured on peak co-presence and reconciles to contract wordings. Hull, cargo and P&I are modelled separately and recombined on a per-event basis. Every scenario assumption — diversion rate, duration, capacity availability — is written down as a parameter with a range and an owner. And the path from gross loss to net loss to capital and liquidity is traceable in a single document that a validator can follow without asking for a verbal explanation.
None of that requires a proprietary model. It requires the discipline to treat a voyage as the unit of exposure and to refuse to average away the concentration that makes the peril worth modelling in the first place.
Sources
- Lloyd's, press release, 19 June 2026: Chubb-led marine war-risk consortium — up to USD 200m capacity separately for hull and P&I risks, plus USD 200m dedicated cargo capacity, subject to underwriting criteria, sanctions and regulation.
- Lloyd's market guidance and insights for the applicable underwriting and regulatory framework.