RESEARCH DOSSIER
Marine Cargo Catastrophe Exposure and Loss Aggregation
Abstract. Marine cargo catastrophe exposure is the value of insured interests susceptible to one damaging event while goods move through geographically concentrated and operationally interconnected transport systems. Cargo catastrophes arise from hurricanes, floods, earthquakes, tsunamis, explosions, fires, container casualties, and other severe events whose physical consequences can reach cargo on vessels, at ports, inside terminals, in warehouses, or in inland transit. The resulting insurance loss depends jointly on the event footprint, the location and economic identity of each shipment, the mechanism of physical damage, the insurance period and conditions, and the contractual definition of a loss occurrence. This monograph examines that complete sequence. It differentiates cargo value, physical loss, direct indemnity, aggregated insured loss, and net reinsured loss; establishes practical methods for reconstructing exposure at a catastrophe timestamp; compares contractual and statutory controls in an international setting; evaluates public evidence from Tianjin, Beirut, world maritime transport, and container-loss reporting; and derives an auditable event-to-policy-to-treaty calculation. A fully specified hypothetical portfolio demonstrates the effect of coverage limits, shipment-presence uncertainty, hazard severity, and alternative occurrence definitions on insurer retention. The research assesses geolocation error, timing mismatch, incomplete declarations, changing port dwell times, correlated vulnerability, data censoring, and event catalog uncertainty. Its principal analytical finding is that geographic accumulation becomes a financially meaningful catastrophe accumulation only after cargo identities, temporally valid physical presence, insurable interests, policy response, and contractual grouping are reconciled. A single port inventory total or model-generated probability of exceedance cannot establish an insurer's aggregate catastrophe liability without those reconciliations. The findings are an original synthesis of verified legal, technical, empirical, and actuarial materials. No new authorial theory or proprietary catastrophe model is asserted.
Keywords: Marine cargo insurance; catastrophe risk; cargo accumulation; port risk; terminal storage; shipment exposure; event footprint; flood; explosion; container loss; catastrophe modeling; insured value; Institute Cargo Clauses; policy attachment; occurrence definition; maximum foreseeable loss; expected annual loss; reinsurance; sensitivity analysis; exposure data quality; marine insurance law; maritime logistics.
Introduction
Marine cargo insurers underwrite property in motion. Their portfolios include finished goods, bulk commodities, machinery, pharmaceuticals, agricultural products, electronics, vehicles, and other commercially distinct interests. A particular cargo may spend days at a terminal, a week aboard a vessel, time in a transshipment hub, and further time at an inland distribution location. The port is thus simultaneously a gateway, a place of storage, a concentration of goods belonging to numerous assureds, and a potential catastrophe epicenter. A storm surge or explosion affects this system through a physical footprint that has little regard for the categories by which insurance policies and logistics platforms record it.
The scale of the problem is apparent in international marine statistics. The International Union of Marine Insurance reported cargo premiums of USD 24.2 billion for 2025, accounting for approximately 57 percent of worldwide marine premium income in its 2026 global marine insurance report. This is an indicator of the economic size of marine cargo underwriting, rather than an exposure measure: premium volume cannot be converted directly into catastrophe values at risk. The UN Trade and Development Review of Maritime Transport 2025 describes rerouting, strained port performance, and congestion that change where cargo spends time. These movements matter because port inventory and catastrophe vulnerability vary across days and even hours.
The World Shipping Council's 2026 containers lost at sea report estimates 1,478 containers lost during 2025 out of approximately 280 million transported, compared with 576 in 2024. One major 2025 incident accounted for 640 containers. These are container counts and operational loss observations, not insured marine cargo claim counts or measures of declared cargo values. That distinction is fundamental: one container can contain low-value goods, exceptionally expensive cargo, multiple interests, or no covered interest under a particular policy.
Catastrophe modeling provides an established intellectual structure for studying rare, spatially dependent losses. The treatment developed in Grossi and Kunreuther's Catastrophe Modeling research volume separates hazard, exposure, damage, and financial response. The peer-reviewed analysis of simulation uncertainty in catastrophe models demonstrates why output variation cannot be reduced to a single source of randomness. Marine cargo adds three demanding features to that framework: exposures move during the hazard interval; cargo identities and insurable interests are not coextensive with containers or vessel units; and final claims may depend on multiple primary and reinsurance wordings.
This study follows a source-based legal, institutional, and quantitative method. It reviews original cargo-clause publication records, English marine insurance legislation, shipping safety requirements, international transport data, insurer and reinsurer catastrophe reporting, and supervisory and actuarial standards. Empirical publications are distinguished from illustrative calculations. The jurisdiction of each legal rule is identified, global statistics are described by their actual measurement unit, and hypothetical results are labeled as such. The six chapters proceed from concepts to event footprint, shipment exposure, claims aggregation, evidence criticism, and integrated interpretation. The analysis is designed for scientific documentation and institutional understanding rather than the evaluation of any particular insured's policy.
Conceptual and Historical Foundations
The insured object and five distinct loss units
The fundamental object of marine cargo catastrophe analysis is an insurable economic interest in identifiable goods during a specified period of exposure. For a research portfolio, the economically relevant unit may be a shipment line, purchase order, lot, vehicle, bulk parcel, pallet, container segment, or separately insured high-value item. The choice must preserve unique ownership and valuation while permitting physical location and damage to be reconstructed. A container is a useful logistics identifier but does not necessarily correspond to one insurance contract, one owner, one declared value, or one claim.
Five units require permanent separation. First, the exposure unit records value potentially subject to loss at a time and place. Second, the damage unit records physical consequences for the goods, including total destruction, partial impairment, contamination, or recoverable deterioration. Third, the claim unit records a legally and contractually responsive demand for indemnity after exclusions, deductibles, limits, and valuation conditions. Fourth, the occurrence unit groups related claims according to the governing contract and its event definition. Fifth, the annual financial unit assembles occurrences across the insurer's reporting period, including reinsurance and reinstatement effects.
Their relationship is directional but conditional. A cargo physically present in a flood zone may escape damage. Damaged cargo may fall outside policy attachment or be excluded by an applicable term. Several distinct cargo claims may belong to a single reinsurance occurrence. Conversely, one meteorological system may generate several contractually separate occurrences where wording and governing law so provide. A claim also may comprise general average or mitigation expenses, depending on the insurance contract, rather than only the diminished market value of goods. The IUA's published Institute Cargo Clauses (A) record establishes the identification of the CL382 edition dated January 1, 2009; the particular policy schedule, subsequent amendments, and governing law determine its application.
The same event can affect vessel hull, cargo, terminal property, port equipment, environmental liability, and interruption of business. Cargo is an insured-property category; hull is an interest in the ship; protection and indemnity (P&I) typically addresses specified liabilities of ship operators; terminal property and business interruption depend on their own wordings. Reinsurance can link these portfolios, but the gross physical impact of a casualty is not automatically the cargo indemnity paid by one insurer. A correct catastrophe analysis maintains separate ledgers for these categories and reconciles shared event identities without counting the same economic loss twice.
How cargo accumulates through the transport chain
Cargo concentration has several physical forms. A loaded ship may contain thousands of separate shipments and interests. A terminal may contain boxes awaiting customs release, goods unloaded but still under carrier custody, export freight not yet on board, and containers temporarily stored after operational disruption. A free-zone warehouse may serve multiple insurers and assureds. Inland depots, container freight stations, river terminals, rail interchanges, and airport cargo facilities can concentrate similar goods outside the core port boundary.
Commercial concentration may be less visible than physical concentration. One marine open-cover policy can insure many separate consignments in transit. One multinational enterprise may insure shipments issued by numerous subsidiaries. One reinsurer may accumulate participations via direct policies, facultative placements, quota-share treaties, and multiple cedents. Therefore, a map of vessels is an incomplete proxy for marine cargo exposure, just as a port manifest is an incomplete map of insured interests.
Temporal accumulation is central. A shipment's presence indicator changes from one location to another, and the indicator may change within the footprint of a single windstorm or flood episode. The UNCTAD maritime transport assessment links route disruptions with higher waiting times and altered shipping networks. The actuarial implication is a change in dwell-time distributions, which changes the probability that insured goods will be present in a vulnerable location when the peril materializes. This is a reasoned inference from transport operations, not a quantified global cargo-loss effect derived by UNCTAD.
Catastrophe experience and the evidentiary problem
Catastrophe events reveal the importance of coincidence in place and time. A large explosion at a terminal can destroy locally concentrated cargo and adjacent buildings; an extensive storm surge can affect several terminals and off-site warehouses simultaneously; a vessel fire can create a shipboard cargo concentration. The stochastic dependence is produced by common peril exposure. It persists even if individual policyholders are unrelated and their shipments originate in different countries.
The 2015 explosions at the Port of Tianjin provide the most recognizable modern warning. Swiss Re's contemporaneous 2016 sigma catastrophe release provisionally estimated USD 2.5–3.5 billion of insured property losses from the event, emphasizing the difficulty of assessing many concentrated assets, including vehicles in transit, and the complications of an exclusion zone. This is an estimated multi-line property loss figure, not a verified global cargo-only settlement total. Its significance here lies in demonstrating the dependence of ultimate loss estimation on reconstructing asset positions, coverage and damage after a severe common event.
The World Bank, United Nations, and European Union's Beirut rapid damage assessment estimated USD 3.8–4.6 billion in physical-stock damage and USD 2.9–3.5 billion in economic-flow losses following the August 4, 2020 port explosion. These figures cover much more than marine cargo, and economic-flow losses are not the same as insured business-interruption claims. They illustrate why citywide disaster estimates cannot be imported into a cargo portfolio as insured sums.
A historical event can establish the existence of a concentration mechanism without identifying its full event-level insurance liability. Published industry totals may cover multiple lines, multiple jurisdictions, or different loss development stages. This monograph therefore uses cases as evidence of failure modes and scale while reserving quantitative insurer-level conclusions for models with transparent shipment and policy inputs.
Event Footprint
Physical event, peril and loss causation
An event footprint is a spatial-temporal representation of a causally connected occurrence, including the location, timing, and intensity of the physical process. For a tropical cyclone it may combine wind speed, surge depth, precipitation and river flooding; for an earthquake it may combine shaking, liquefaction and tsunami inundation; for a terminal explosion it may combine blast pressure, thermal effects, fire spread and contaminated debris. A useful footprint preserves the distinction between processes because cargo susceptibility and insurance coverage may differ among them.
For every event e, let H_e(x,t,k) denote the intensity of physical mechanism k at location x and time t. For shipment i, let X_i(t) be the best available location and S_i(t) the operational state, such as onboard, stacked outdoors, inside a warehouse, in a reefer unit, or already delivered. Its affected status cannot be inferred solely from the vessel's coordinates where the shipment is no longer on board. Damage modeling conditions on the intersection of H_e with X_i(t), the period of hazard exposure, and the vulnerability of goods in state S_i(t).
The official NHC best-track archive supplies six-hourly tropical cyclone locations, maximum winds, central pressure and, for the relevant periods, wind radii, together with storm reports and GIS products. These observations provide event evidence but are not a complete local cargo-hazard field. The maximum wind at a storm center does not specify inundation at a particular berth, and a six-hour observation grid may omit a critical short exposure interval. For other regions or perils, locally authoritative meteorological, hydrographic, seismic and port authority sources must be selected independently.
A causal chain can cross several mechanisms: wind destroys power supply; a refrigerated container loses temperature control; cargo deteriorates; a policyholder claims. The physical sequence must be established, and then the wording must determine whether the damage mechanism is insured, whether deterioration is a covered physical loss, and whether a delay, temperature-control, inherent-vice or other exclusion applies. Event geography establishes exposure; it does not resolve proximate cause or coverage by itself.
Spatial, temporal and intensity characterization
Ports require a finer geospatial model than a point-coordinate representation. Relevant locations include marine terminal polygons, yard blocks, berth lines, bonded warehouses, external logistics parks, road access, inland freight stations, and ship anchorage positions. Vertical elevation, flood barriers, warehouse floor levels, pile height, and packaging conditions affect vulnerability. An event that overtops one terminal may leave another unaffected despite their shared port code.
Time is equally granular. A useful exposure record carries the effective-from and effective-to timestamps for each state change, preferably with time zone and source-system timestamps separately. Cargo delivered at 10:00 but scanned into a warehouse system at 16:00 can generate a false six-hour aggregation if administrative event time is mistaken for physical movement time. Likewise, a terminal's gate-in time, discharge time, customs release, and insurer declaration time need separate fields. The correct presence state is the one supported by the best operational evidence at the damage interval, with ambiguity retained where records disagree.
The IMO maritime single-window requirement became mandatory for electronic ship-port reporting from January 1, 2024, under amendments to the FAL Convention. Such standardized electronic exchange improves the administrative environment for transport data, but it does not create a globally complete insurance inventory or prove a shipment's final warehouse bay at an instant. This limitation is important when system interoperability is mistaken for universal exposure observability.
Exposure geometry can be expressed as a collection of uncertain trajectories rather than fixed pins. When the time and route are uncertain, a defensible model assigns probability to alternative physically possible locations and records the evidence supporting those probabilities. A shipment cannot simultaneously be counted in two mutually exclusive places. Spatial uncertainty should propagate into a distribution of event loss instead of disappearing through an arbitrary selection of the most severe location.
Operational evidence and event attribution
An evidentiary reconstruction should compare the bill of lading, cargo declaration, terminal operating-system gate transactions, vessel manifest, container number, booking number, vessel and voyage identifiers, customs records, warehouse receipts, transport waybills, invoices, GPS or AIS data where relevant, survey reports, photographs, and salvage documents. Each source answers a different question. A bill of lading addresses carriage and goods descriptions; a terminal transaction indicates an operational movement; a commercial invoice informs value; a cargo policy defines covered interest; AIS principally reports vessel movement, not the ownership or contents of each unit on board.
A normalized event dossier includes the source and UTC timestamp for hazard observations, the movement-event ledger, geospatial transformations, insured values, policy wording editions, loss surveys, claims adjustments, and the methodology for attaching records to an occurrence. The evidence chain should retain original identifiers and every subsequent transformation so that a later reviewer can reproduce the exposure at the event timestamp.
IMO safety instruments provide contextual information about preventable physical mechanisms. The IMO/ILO/UNECE cargo transport unit code addresses packing, securing, and related intermodal handling as a non-mandatory global code of practice. The IMO's verified gross mass requirements, effective July 1, 2016 under SOLAS, require verification of packed-container mass before loading. These instruments concern operational safety and do not themselves settle a cargo insurance claim.
Dangerous goods create another source of common-event severity. The IMO IMDG Code 2024 edition, incorporating amendment 42-24, became mandatory from January 1, 2026 for its defined scope of packaged dangerous goods carriage. The exact edition and application date matter when examining historical events. Compliance evidence can inform causal assessment and loss prevention, while insurance coverage continues to turn on contractual language and applicable law.
Legal and contractual boundaries of covered damage
Under English law, the Marine Insurance Act 1906, section 55 states the statutory default for loss proximately caused by an insured peril, subject to the policy's terms, and addresses specified excluded loss mechanisms, including delay unless the policy otherwise provides. The modern Insurance Act 2015 duty of fair presentation governs non-consumer insurance risk presentation within its relevant statutory scope. Neither provision establishes a uniform worldwide marine cargo coverage rule. Their relevance is concrete for English-law placements and comparative analysis of disclosure and claims causation.
The dated Institute Cargo Clauses (A) CL382 are widely recognizable market clauses, but they must be read with their exclusions, duration provisions, schedule and any amendments. Their recognizable all-risks formulation does not imply that every catastrophe consequence is insured. In the published 2009 wording, delay and specified inherent or ordinary properties of the cargo are addressed in exclusions; special war and strike cover is a separate contractual inquiry. The IUA separately identifies the Institute War Clauses (Cargo) CL385, also dated January 1, 2009. The title of a loss event in a catastrophe database cannot substitute for examination of the particular peril, clause edition and endorsements.
Transit duration and storage can be decisive. A catastrophe striking a warehouse after normal transit has terminated may fall inside or outside the insured period depending on agreed extensions, storage terms, attachment rules and governing law. Conversely, a shipment temporarily awaiting onward carriage may still fall within an appropriate transit wording. One must identify when risk attaches, where the goods actually were, who had an insurable interest, what policy covers that interest, and whether the proximate cause is insured.
Carriage liability is a distinct legal inquiry. The Hamburg Rules treaty record identifies a convention in force from November 1, 1992, while the Rotterdam Rules status record shows that the latter had not met its required entry-into-force threshold as of this study. These treaties concern contracts of carriage and potential carrier responsibilities; their relevant jurisdictions, dates, and private-law application must be evaluated separately from cargo insurance coverage and recovery rights. A liability recovery or subrogation prospect may influence the net ultimate claim without changing the event's initial physical damage.
Shipment Exposure
Ownership, insurable interest and policy identity
The exposure inventory begins with a unique economic item, not an estimate of the monetary contents of every container visible in a port. A shipment may be insured by a manufacturer, seller, buyer, financier, logistics provider or other party with a relevant interest, and the transfer of risk under the sale agreement may differ from transfer of ownership or physical custody. The research record must therefore hold several identifiers: goods or lot, commercial contract, principal assured, other interested parties, declaration, direct insurance policy, carrier, shipment and voyage, and the relevant time periods.
For each economic item i, define V_i(t) as the contractually relevant value at time t, expressed in a specified currency and valuation basis. A sum insured may include invoice cost, freight, anticipated profit or other components under the wording. The cargo's declared value may differ from its market replacement value, salvage value or the cost of reconstructing an unfinished manufactured item. Cargo valuation should preserve the policy basis rather than mechanically multiplying every container by an industry average.
Double insurance and layering require a separate record. Two policies may respond to the same interest, but the insurer's aggregate financial loss is determined under their actual contribution and other-insurance provisions. A reinsurer may hold multiple shares that trace back to the same cargo. Counting every underwriting record as a distinct physical asset inflates a modeled catastrophe inventory, whereas eliminating duplicates without tracing legally distinct insured interests can understate the financial system's obligations. Physical asset identifiers and contractual coverage identifiers must remain linkable but distinct.
Location at the event timestamp
Let P_i(x,t) denote an exposure-location distribution for cargo item i at time t. With complete and reliable operational tracking, its probability is concentrated at one observed state. With missing scans or disputed transfer times, it may assign weights to mutually exclusive feasible locations. The total weight across the feasible states equals one. The modeling object is a probability distribution over physical states, not a license to count one cargo twice.
A port inventory snapshot is reconstructed by joining the goods ledger with time-valid movement events. The join must distinguish a ship's arrival from the physical discharge of a box; an export booking from terminal gate-in; a customs clearance from removal; and a warehouse's booking record from actual receipt. The port code is not a sufficiently precise vulnerability location. For flooding, elevation and storage-floor information may be essential; for explosion or fire, proximity to the ignition source and compartment barriers may be decisive.
A useful reconciliation hierarchy starts with verified physical transactions, incorporates corroborated terminal and warehouse records, tests manifest consistency, and uses vessel-position information to narrow alternative states. Contradictions should be retained with a recorded confidence level and reason, such as late data entry or missing intermediary. Administrative updates that occur after a catastrophe do not necessarily show that the cargo moved after the event. An audit trail therefore preserves both event time and system-recording time.
The rule for a missing shipment is disciplined uncertainty, not an automatic zero. If the declaration shows a shipment covered during the week but there is no contemporaneous location record, it contributes to an exposure uncertainty set. Location probabilities may be estimated from the transport schedule and available checkpoints. A catastrophe assessment reports both a central estimate and material bounds arising from that uncertainty, especially when a high-value shipment could materially alter the tail.
Cargo value, storage, routing and concentration
A transparent aggregation register stores at least the item identifier; container or conveyance; beneficial or insured interest; coverage policy; voyage; currency and valuation basis; quantity; physical origin and destination; all time-valid stops; warehouse and terminal zone; peril-sensitive characteristics; packing and storage state; deductibles and limits; any agreed location or conveyance sublimit; and source timestamps. Currency conversion requires a defined valuation date and exchange-rate source. Values in different currencies must not be added without a documented convention.
Operational concentration evolves with arrivals, discharge, inspections, storage, onward movements and transshipments. This produces occupancy peaks that a quarterly stock average can conceal. The port's nominal throughput is a flow measure; cargo accumulation is an inventory at a point or over an event window. The two are linked through dwell time but are not equal. An unusually long dwell time may increase snapshot concentration while also altering the route by which an extreme event can affect goods.
Consider an illustrative terminal inventory with USD 100 million of opening goods, USD 25 million of recorded inward movements and USD 20 million of outward movements during a day. If all movements are correctly timestamped and unique, the closing physical inventory is USD 105 million, not USD 145 million. Adding opening stock and all movements as if each were a different exposed asset creates false accumulation. The illustrative figures are not drawn from an observed port; they show the inventory-identity principle.
Cargo susceptibility can be highly heterogeneous: bulk grain may suffer water or contamination damage; automobiles may be damaged by hail, inundation, or blast debris; electronics can fail after saltwater exposure; refrigerated cargo depends on temperature-control conditions; chemicals can generate secondary fire or toxic-release events. A single mean damage ratio applied to the entire port understates differences among locations, packaging, duration and commodity-specific vulnerability. The IMO cargo packing and securing guidance provides a source-based operational context for these differences, without supplying insurance-indemnity vulnerability curves.
Exposure registers and reconciliation
The fundamental exposure quantity for event e is not the sum of all goods ever routed through the port. It is the value of unique insured interests present within relevant physical mechanisms during the hazard interval. In simplified notation:
where I_i(e) equals one if item i is present in the defined exposure footprint and period, and zero otherwise. For uncertain presence, the expected exposure replaces the indicator with a calibrated presence probability. This is a value-at-risk inventory measure, not a prediction of damage or paid claims.
For underwriting, at least three views are needed. The physical view shows total unique goods exposed to a given hazard. The direct insurance view attributes covered interests and shares to each insurer. The ultimate risk-carrier view traces the amounts to reinsurance and retrocession, retaining the underlying identity so common interests can be recognized. These views can legitimately have different totals because they answer different financial questions.
An operational accumulation control is only as reliable as its update frequency and completeness. The most consequential failures include absent storage declarations, blanket open-cover reporting at insufficient granularity, incorrect insured values, duplicate shipment identifiers, movement records entered late, and confusion between warehouse stock policies and transit policies. Regular reconciliation against the declaration ledger, shipment milestones and recent claims is therefore part of catastrophe exposure management rather than merely a data-cleaning exercise.
Loss Aggregation
From physical damage to insured indemnity
Event loss calculation has at least four transformations: exposure to physical damage, damage to policy-eligible loss, eligible loss to direct indemnity, and direct indemnity to the insurer's net retained share. Let d_i(e) be the actual or simulated physical loss fraction for goods i in event e, bounded between zero and one, and V_i the appropriate valuation basis. The ground-up physical loss for the item is G_i(e)=V_i d_i(e), adjusted as necessary for salvage or expenses when consistent with the chosen definition. This is a modeling identity; it does not resolve legal coverage.
Let C_i(e) be an indicator or more detailed assessment of whether the peril, time, location, insurable interest and policy conditions yield covered loss. Let D_i represent an applicable deductible, L_i a relevant limit, and q_i the insurer's participation share where applicable. In a simplified single-item indemnity with no other constraints, the insurer amount is:
Actual contracts may use deductibles by shipment, conveyance or occurrence, franchise conditions, percentage deductibles, location sublimits, aggregation across declarations, valuation clauses, contribution provisions and distinct treatment of salvage, general average, or sue-and-labor expenses. The simplified function is useful precisely because each omitted contractual feature can be named and tested.
The event gross insured loss is S_e=\sum_i Y_i(e) after eliminating duplicate economic interests and observing each insurer's appropriate policy share. An occurrence or catastrophe limit may operate on the aggregate after individual items have been calculated, changing the expression. The order of operations is essential. The sum of capped item indemnities differs from the cap applied to total physical loss. Direct policy response is calculated before reinsurance coverage, since reinsurance has its own terms and loss basis.
A reported catastrophe's economic loss, insured property loss and marine cargo direct indemnity therefore occupy different statistical populations. The global market premium statistic published in IUMI's 2026 report is a premium flow; the cargo asset values insured at a port are exposures; and settled catastrophe claims are ultimate financial outcomes. Comparisons across these measures require explicit denominators, periods, currencies and scope.
Event-by-event and annual calculations
For a stochastic catastrophe analysis, a year contains a random number N of relevant events, with occurrence losses S_1,...,S_N and annual aggregate loss A=\sum_{e=1}^{N}S_e. Expected annual loss equals E[A]; with independent identically distributed event severities and N independent of severities, E[A]=E[N]E[S]. The independence assumption may be inappropriate where events cluster seasonally or share a latent climate or operational condition. Cargo is particularly sensitive to exposure occupancy at the actual event time.
Occurrence exceedance probability at threshold z is the probability that the largest single event during a year exceeds z. Aggregate exceedance probability at threshold z is the probability that annual summed loss exceeds z. The distinction matters when several moderate events produce an annual aggregate above a threshold that no single event reaches. Gross, net, expected and conditional tail measures should be labeled by loss basis and the layer of cover.
As a fully hypothetical frequency example, assume independent Poisson processes for annual events producing insured severities of USD 11 million and USD 20 million, at respective rates 0.06 and 0.02. The expected annual insured loss is 0.06×11+0.02×20=USD 1.06 million. The occurrence probability of at least one loss greater than USD 10 million is 1−exp(−0.08)=7.6884 percent. The occurrence probability of an event greater than USD 15 million is 1−exp(−0.02)=1.9801 percent. The annual aggregate probability of loss greater than USD 15 million is 1−exp(−0.08)×1.06=2.1497 percent, because two USD 11 million events also exceed that annual threshold. These are mathematical consequences of explicit illustrative assumptions, not empirical port catastrophe estimates.
This simple example also shows why expected loss alone is inadequate for concentration control. Two portfolios can have the same expected annual loss but radically different event severities and thus different liquidity, reinsurance and capital needs. A defensible pricing and solvency analysis examines the entire material portion of the loss distribution and the assumptions governing its upper tail.
Illustrative port catastrophe portfolio
Consider four cargo interests insured by one hypothetical direct insurer. All currency amounts are in USD millions, values are agreed claims valuation bases, and the catastrophe is assumed to consist of one eligible physical peril during a covered transit/storage interval. Each claim is subject to a per-interest deductible and indemnity limit, with no coinsurance, salvage adjustment, or other insurance. The assumptions are deliberately simplified to make every calculation auditable.
| Cargo interest | Physical state at event | Insured value | Damage fraction | Physical loss | Deductible | Item limit | Direct indemnity |
|---|---|---|---|---|---|---|---|
| A: vehicles | Exposed terminal quay | 12.00 | 0.60 | 7.20 | 0.50 | 8.00 | 6.70 |
| B: electronics | Inland warehouse reached by flood | 9.00 | 0.40 | 3.60 | 0.20 | 5.00 | 3.40 |
| C: machinery | Vessel outside affected footprint | 6.00 | 0.00 | 0.00 | 0.20 | 4.00 | 0.00 |
| D: consumer goods | Outdoor container stack | 5.00 | 0.20 | 1.00 | 0.10 | 2.00 | 0.90 |
| Total | Four unique interests | 32.00 | — | 11.80 | — | — | 11.00 |
The table shows why physical value at risk of USD 32 million, ground-up loss of USD 11.80 million and insurer direct indemnity of USD 11.00 million are all correct but answer separate questions. The absence of damage to cargo C is a feature of the event footprint. Adding C's declared value to the losses would be a valuation category error. Likewise, summing gross physical loss without deductibles exaggerates this insurer's contractual direct loss.
Now assume an illustrative occurrence catastrophe excess-of-loss reinsurance layer covering USD 5 million in excess of USD 5 million of eligible direct loss, with no additional exclusions, coinsurance or reinstatement effects. For one occurrence S=11.00, the layer pays min(5,max(11−5,0))=USD 5.00 million, leaving USD 6.00 million retained. This result depends on the assumption that A, B and D form one covered treaty occurrence.
If the governing reinsurance wording and applicable facts instead produced two separate occurrences, with A and D grouped at USD 7.60 million and B at USD 3.40 million, the same layer would pay USD 2.60 million on the first and zero on the second. The total insurer retention would be USD 8.40 million. The difference of USD 2.40 million arises solely from the occurrence grouping assumption; the cargo and the event's total physical damage are unchanged. This is a scenario analysis, not a prediction about how a particular catastrophe would be construed by any court.
Next suppose warehouse records cannot establish whether B was physically inside the flood footprint when the loss mechanism occurred, and two mutually exclusive feasible reconstructions receive illustrative weights of 0.5 each. With B present the direct loss is 11.00 and retained loss 6.00; with B absent the direct loss is 7.60, reinsurance recovery is 2.60 and retained loss is 5.00. The expected retained loss over these two reconstructions is USD 5.50 million, with a USD 5.00–6.00 million scenario range. The weights are invented assumptions for a teaching calculation and have no empirical basis in a real warehouse.
Finally, apply uniform illustrative damage-severity multipliers of 0.75 and 1.25 to the three damaged interests while holding values, deductibles and item limits constant. At 0.75 the indemnities are 4.90, 2.50 and 0.65, totaling 8.05; the layer recovers 3.05 and the insurer retains 5.00. At 1.25 the item A claim reaches its USD 8.00 million limit, while B pays 4.30 and D pays 1.15, totaling 13.45; the layer recovers 5.00 and the insurer retains 8.45. Both damage ratios and layer recoveries are nonlinear in this interval because of the item cap and the occurrence-layer attachment and exhaustion.
| Hypothetical calculation | Direct insured loss | Reinsurance recovery | Insurer retained loss |
|---|---|---|---|
| One occurrence; central damage | 11.00 | 5.00 | 6.00 |
| Two occurrences; central damage | 11.00 | 2.60 | 8.40 |
| B physically absent | 7.60 | 2.60 | 5.00 |
| Damage severity ×0.75 | 8.05 | 3.05 | 5.00 |
| Damage severity ×1.25 | 13.45 | 5.00 | 8.45 |
These calculations establish three interpretive results. Exposure uncertainty changes the existence of a loss item; physical severity changes its damage amount; occurrence wording changes financial grouping. Their effects can move in different directions and cannot be summarized by one universal percentage error. In particular, an insurer may experience the same total covered direct loss yet retain substantially more if the treaty recognizes multiple separate occurrences.
Reinsurance occurrence and aggregation
In practical treaty analysis, the terms of catastrophe coverage, hours clauses, event or cause language, geographic limits, attachment, exhaustion, reinstatements, aggregate deductibles, annual limits and exclusions must be checked one by one. The relationship of an initial explosion and subsequent fire, or a prolonged storm and multiple flood episodes, may depend on whether the relevant contract groups by originating cause, occurrence, related events or a defined time window. A catastrophe model's internal event identifier has no automatic contractual authority.
The International Association of Insurance Supervisors' ICP 13 reinsurance guidance emphasizes the need to assess effective risk transfer, risk concentration, reinsurance terms, documentation, recoverables, and the possibility of mismatches between original policies and reinsurance. It also recognizes reinsurance counterparty and basis risks. The consequence for cargo accumulation is that gross port-level concentration and net retained port-level concentration require distinct calculation and governance.
Facultative participation can produce additional complexity when the same catastrophe affects several large declared shipments. Quota-share treaty recoveries, per-risk excess covers, catastrophe excess layers and retrocession can interact. The financially correct calculation follows the actual sequence of contracts and shares; an undifferentiated 'reinsurance percentage' can misstate recoveries near attachment thresholds. A single severe port event may also exhaust more than one cedent's layer with the same reinsurer, creating risk accumulation at the reinsurer level that direct insurer reports alone do not reveal.
The occurrence analysis should preserve documented alternative readings where contract interpretation is genuinely unsettled. A model can test conservative and less conservative outcomes, identifying which contract words or facts determine the difference. It must not represent the highest or lowest computed result as an adjudicated entitlement. Contractual uncertainty has a concrete numerical consequence for retained capital and liquidity, and the claim file must show how that uncertainty was handled.
International case evidence
The 2015 Tianjin explosion is analytically valuable because the initiating hazard was spatially concentrated while the affected insured assets and coverages were heterogeneous. Swiss Re's Tianjin accumulation discussion linked the provisional USD 2.5–3.5 billion multi-line insured estimate with limitations on access to damaged assets and uncertainty over exposure concentration. For cargo modeling, the lesson concerns vehicle storage, hazardous materials, terminal yard location and post-event asset verification. The source does not justify assigning that whole dollar range to marine cargo.
The 2020 Beirut event demonstrates a different measurement gap. The World Bank-led rapid assessment of physical damage distinguished destroyed assets and economic activity losses at the city scale. Insurance line, cargo-owner interest and policy response cannot be extracted from those aggregate estimates. The source is nevertheless relevant to the event footprint and the dependence of cargo disruption on damaged port infrastructure. Publicly reported broker and market estimates of insured loss varied substantially soon after the explosion; an academic comparison must identify estimation date and line-of-business scope before juxtaposing figures.
Natural-hazard accumulation has a wider footprint. A hurricane may affect on-board cargo at berth, outdoor stock, warehouse inventories, road access, and inland depots in one period. The NHC hurricane data archive supports event reconstruction at basin-specific scope. A model must separately map wind, flood and surge mechanisms to each cargo state; treating a best-track line as a universal affected-zone boundary is insufficient. Detailed cargo claim counts, limits and exposure location histories are seldom available in public event summaries.
The WSC container loss statistics provide an important complementary measure. Their reported 2025 concentration, with one incident accounting for about 43 percent of estimated annual lost containers, shows that isolated major casualties can dominate observed container loss frequency. It does not estimate the insured severity of the lost goods; casualty concentration in container numbers is not a calibrated distribution of cargo insurance catastrophe losses. Conversely, port catastrophes can affect goods never lost at sea. A scientifically honest aggregation analysis preserves this difference.
International comparison therefore works best at the level of mechanisms, measurement protocols and documented legal regimes rather than producing synthetic universal loss ratios from incomparable case estimates. The relevant evidence includes marine and property loss categories; insured versus economic amounts; early versus developed claims; exposure unit; catastrophe footprint; policy wording; date; and currency. If one of these elements is unavailable, the comparison should show the gap explicitly.
Evidence Appraisal and Sensitivity
Data provenance and missingness
Three distinct data families enter marine cargo catastrophe modeling. Hazard observations describe events and intensity. Operational data describe the position, contents and state of goods. Insurance data describe values, insured interests, policy terms and realized claims. These families are produced by different organizations for different purposes and are often collected on incompatible spatial grids, timestamps, identification systems and financial bases. Linking them is a substantive inferential operation.
The EM-DAT database limitations statement recognizes incomplete event reporting and data inconsistencies. Its economic impact variable documentation further warns that disaster economic damage is often underreported and insured damage entries tend to be available for more visible events and well-covered jurisdictions. These are dataset-wide limitations, not a calibrated bias adjustment for any particular marine cargo claims series. They forbid treating every missing insured-loss observation as a zero.
Official catastrophe records can also be revised. The NHC best-track archive documents post-analysis tracks, winds and radii, with historical revisions. For financial backtesting, an analyst must identify whether the hazard data represent information available at the underwriting date, immediately after the event, or after subsequent scientific revision. Using hindsight data to evaluate operational forecasts without labeling the distinction can overstate predictive skill.
Industry claims data also change through the settlement cycle. Immediately after a port catastrophe, access restrictions, damaged records, salvage uncertainty, litigation, and delayed cargo declarations can cause substantial loss development. The time stamp and maturity of an incurred-loss estimate should therefore accompany every historical amount. Premiums and model losses must have aligned underwriting and accounting periods before a ratio is interpreted. A model fitted only to publicly reported large disasters will otherwise learn a reporting-selection process alongside catastrophe severity.
Hazard, vulnerability, temporal and contractual uncertainty
Hazard uncertainty concerns the number, location, intensity, timing and physical processes of events. For an explosion, the ignition location and severity may be relatively well known after investigation while secondary fire paths and stored hazardous inventory remain disputed. For a coastal flood, surge height, rainfall, defenses, drainage and building penetration can differ across small distances. For cyclone modeling, uncertainty in local wind and inundation can persist even after best-track analysis.
Vulnerability uncertainty concerns how a given hazard damages a particular type of cargo in a particular physical state. A flood depth of one meter has different consequences for palletized electronics, bulk mineral cargo, sealed metal drums, vehicles and goods held above floor level. The appropriate vulnerability model may require thresholds and categorical damage states rather than a smooth continuous loss fraction. Correlated vulnerability matters when a common warehouse condition or packaging defect exposes many goods simultaneously.
Temporal uncertainty concerns whether the goods were present during the relevant hazard interval and whether insured transit had attached or terminated. An event that strikes at a terminal gate transition can make a small timestamp error financially decisive. The model should not convert location uncertainty into random independent assignments for every box when many boxes were moved by the same convoy or unloaded by the same operation. Common movement and processing uncertainties induce dependence between exposure states.
Contractual uncertainty concerns coverage interpretation, valuation, exclusions, occurrence grouping, attachment and limits. The primary Institute Cargo Clauses (A) edition and the English statutory causation rule demonstrate that physical damage is not equivalent to covered loss. A different legal system, manuscript clause, warranty or extension can change the answer; legal application cannot be inferred from commodity category alone.
Model uncertainty concerns the appropriateness of statistical structure, event clustering, dependence, loss severity tails and simulation. The research on catastrophe simulation uncertainty offers an analytical warning against treating model outputs as exact. Marine cargo adds fast-moving insured inventory, so distributional assumptions estimated from annual property portfolios may miss short-lived concentrations. Scenario ensembles should vary exposure and contract inputs jointly when the underlying failures could occur together.
A practical sensitivity matrix varies at least event intensity, footprint, cargo presence, item valuation, vulnerability, deductibles, policy cover, occurrence grouping and the reinsurance layer's interpretation. One-variable changes isolate mechanisms, while correlated stress scenarios test plausible combinations. The numerical example above demonstrates that a modest intensity increase may push an item to its limit and a treaty into exhaustion, producing discontinuities in retained losses. Such discontinuities justify explicit stress tests around each significant financial threshold.
Comparative regimes and governance
Marine cargo insurance is international in commerce and diverse in law. English-law marine insurance statutes and London market clauses are influential but cannot be projected unchanged onto every jurisdiction. Local mandatory insurance rules, contract interpretation, claims practices, carriage liability conventions, customs procedures, storage regulation and dangerous-goods requirements may change the relevant exposure and recovery. A scientific comparison therefore identifies the law chosen for the policy, the jurisdictions connected to the event and the dates of each operative instrument.
Carriage conventions illustrate this constraint. UNCITRAL's Hamburg Rules status information identifies participating states and entry-into-force dates; the Rotterdam Rules participation record identifies signatures and ratifications while recording that required entry into force had not been reached. A contract-of-carriage analysis must determine which regime actually governs before estimating carrier recovery or liability. Neither regime establishes the cargo insurer's own catastrophe occurrence definition.
Safety requirements must likewise be dated. The 2014 CTU Code implementation context is non-mandatory unless implemented elsewhere; SOLAS verified gross mass rules became effective in July 2016; the IMDG 42-24 amendment is mandatory from January 2026 within its specified scope. Historical claims analysis should use the operative editions on the event date rather than present-day rules retrospectively.
Professional risk governance requires independent model challenge and traceable communication. The U.S. Actuarial Standards Board's ASOP 38 catastrophe modeling standard, effective for work performed on or after December 1, 2021, addresses model understanding, appropriateness, output validation, reliance on others and documentation within its professional scope. IAIS ICP 13 risk-transfer requirements provide a supervisory basis for reconciling gross concentration, net retention and reinsurance risk. Their common practical implication is that a plausible-looking catastrophe result is not a substitute for documented assumptions, validations, contract checks and a clear statement of uncertainty.
Validation and editorial reproducibility
Validation should operate independently on each material stage. Exposure validation reconciles cargo declarations and movement ledgers to terminal stock and exceptional high-value items. Hazard validation compares modeled spatial intensity with authoritative observations and engineering or forensic evidence. Vulnerability validation compares predicted damage with surveyed outcomes for comparable commodities and storage conditions, adjusted for claim selection and censoring. Financial validation replays the actual policy and treaty wordings against documented sample claims and reconciles gross and net amounts to insurer records.
An end-to-end stress test should ask whether the model can reconstruct a known single catastrophe without using facts that became available only later as if known before the event. It should also challenge a high-value shipment with a missing location, a record duplicated under two container numbers, a policy ending during storage, a cargo loss and hull loss from the same casualty, and a treaty with alternative occurrence definitions. These are specific test cases in which semantic and financial errors can dominate statistical error.
Backtesting needs a stable claim definition. A model predicting cargo physical damage cannot be assessed by a dataset of all-line property insured claims without a bridge between the measures. Nor should it be evaluated against publicly reported uninsured economic disruption. Where only a historical catastrophe's event date and total citywide damage are known, the evidence supports qualitative scenario plausibility, not numerical calibration of a marine cargo insurer's tail.
For the present work, the hypothetical portfolio has been recalculated independently from its disclosed values, deductibles, limits and assumed treaty. The Poisson example is analytically derivable. No proprietary port inventory was obtained, no confidential cargo claims sample was audited, no commercial catastrophe model was run, and no empirical tail parameter was fitted. These boundaries delimit the reported research result. Independent peer review remains a separate governance decision.
Scientific Synthesis and Research Questions
The reconstructed evidence supports the following integrated analysis. Cargo catastrophe accumulation is a changing intersection of three observable systems: physical goods, transport activity, and insurance obligations. Event geography connects the first two, while policy terms and insurable interest connect the second to financial responsibility. The intersection must be evaluated at a defined time, for a defined peril and set of contracts. This explains why a port's maximum recorded stock value may be informative for stress testing yet cannot by itself be called its insurer's maximum probable cargo claim.
Physical and contractual accumulation should be controlled at distinct but connected resolutions. The physical unit is a uniquely identified interest in goods with a time-valid position and vulnerability. The insured unit is the covered portion of that interest under an identified contract. The financial event unit is a collection of such claims grouped under a particular policy or treaty definition. For a reinsurer, the financial event may join independent cedents whose claims arise from a shared hazard; the reinsurer must trace them without multiplying the underlying economic interests.
A material distinction also exists between observed concentration and potential concentration. Observed concentration is a dated snapshot with evidence. Potential concentration is a modeled distribution of future time-varying inventories, routing and catastrophe coincidence. Throughput statistics, annual premium volumes and registered container capacity inform the latter only when supplemented by dwell-time, goods-value and insurance participation data. The transport evidence in the UNCTAD global review and the IUMI marine market statistics is therefore complementary, not interchangeable.
The case material leads to a further finding about financial reporting. Swiss Re's multi-line Tianjin estimate, the World Bank's Beirut economic assessment and the WSC's observed container losses cannot be ranked as if they measured the same cargo-loss outcome. Each demonstrates a different facet of hazard, asset exposure or financial consequence. Their proper scientific use is to test categories and mechanisms; transferring one into a cargo-only insured tail number without a bridge would create false precision.
The practical benefit of transparent calculation is visible in the illustrative portfolio. One set of cargo values and damage ratios produced a central USD 11 million direct claim, yet alternative event grouping increased insurer retention from USD 6 million to USD 8.4 million. Additional sensitivity to physical location and severity generated distinct changes in gross and net loss. The controlling principle is simply the correct application of preexisting insurance mathematics and contract terms, with explicit assumptions. This is analytical synthesis, not a new named theory, doctrine or independently validated universal model.
Several research questions remain valuable for future empirical work. What proportion of major marine cargo claim variance is attributable to outdated terminal location records rather than peril intensity? How do port dwell-time shocks change extreme instantaneous cargo exposure relative to historical averages? Which combinations of open-cover declaration lag and multi-insurer participation produce the largest hidden accumulation? How sensitive are net catastrophe loss estimates to plausible occurrence interpretations in actual treaty portfolios? Can anonymized shipment-event-claim panels support commodity-specific vulnerability estimation without exposing commercially confidential trade data? These questions call for datasets and prospective validation rather than unsupported claims of settled results.
A robust research program would join port-level movement and inventory snapshots, normalized cargo interests, event hazard reconstructions and anonymized direct claims, then reproduce treaty outcomes with separately documented event grouping. It would pre-register source eligibility, timestamp handling, missingness treatment, valuation units, contract editions and validation design. This permits future comparison across regions while preserving the legal and operational differences that make marine cargo catastrophe exposure distinctive.
Conclusion
Marine cargo catastrophe loss aggregation is the financial consequence of an event acting upon distinct insured goods whose positions, physical states and contractual protections change throughout a transport chain. Its measurement requires five successive reconciliations: a causally defined physical footprint; unique cargo interests and time-valid locations; credible estimates or observations of damage; legally and contractually covered direct claims; and the occurrence and treaty rules that determine the insurer's retained loss.
Institutional maritime evidence confirms that transport concentration is economically significant and operationally dynamic. Public catastrophe records demonstrate that severe events can destroy goods in transportation hubs while their published economic and insured loss totals have different boundaries. Original marine clauses, statutory provisions and carriage treaty statuses show why geographic proximity does not automatically establish coverage or aggregation. The hypothetical calculation proves that an unchanged set of claims can produce materially different net results when occurrence definitions change, and that uncertain shipment location and severity must be treated as separate sources of model risk.
The resulting scientific conclusion is a reproducibility requirement: every material catastrophe cargo estimate should be traceable from event coordinates and timestamps through shipment identity and valuation to exact policy response and treaty recovery. Reliable estimates depend on the quality of those links, the visibility of unresolved data and contractual uncertainty, and independent challenge of the resulting loss distribution. This conclusion integrates established marine insurance law, catastrophe modeling, shipping operations, and actuarial governance without claiming a new theory. The research edition is complete as a source-based manuscript and remains unpublished until a separate release authorization and production verification.
References
Actuarial Standards Board. 2021. Actuarial Standard of Practice No. 38: Catastrophe Modeling (for All Practice Areas). Revised July 2021; effective December 1, 2021. ASOP 38 official standard.
Centre for Research on the Epidemiology of Disasters (CRED). n.d. EM-DAT: Known Issues and Limitations. EM-DAT Documentation, accessed November 17, 2026. EM-DAT limitations documentation.
Centre for Research on the Epidemiology of Disasters (CRED). n.d. EM-DAT: Economic Impact Variables. EM-DAT Documentation, accessed November 17, 2026. EM-DAT economic variables.
Grossi, Patricia, and Howard Kunreuther, eds. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer, Huebner International Series on Risk, Insurance and Economic Security, volume 25. DOI: 10.1007/b100669. Catastrophe Modeling publisher record.
International Association of Insurance Supervisors (IAIS). 2024. Insurance Core Principles and ComFrame: ICP 13 Reinsurance and Other Forms of Risk Transfer. Updated December 5, 2024, online consolidated principles, accessed November 17, 2026. ICP 13 official online text.
International Maritime Organization (IMO). 2026. International Maritime Dangerous Goods Code: 2024 Edition, Amendment 42-24. Mandatory from January 1, 2026. IMO IMDG edition description.
International Maritime Organization (IMO). n.d. Maritime Single Window. FAL Convention electronic ship-port reporting requirements effective January 1, 2024; accessed November 17, 2026. IMO maritime single window.
International Maritime Organization (IMO). n.d. Verification of the Gross Mass of a Packed Container. SOLAS regulation VI/2 amendments effective July 1, 2016; accessed November 17, 2026. IMO gross mass requirements.
International Union of Marine Insurance (IUMI). 2026. IUMI Stats Report 2026. Published October 23, 2026. IUMI report and highlights.
International Underwriting Association of London (IUA). 2009. Institute Cargo Clauses (A), CL382. Dated January 1, 2009, joint LMA/IUA clause. IUA original clause listing.
International Underwriting Association of London (IUA). 2009. Institute War Clauses (Cargo), CL385. Dated January 1, 2009, joint LMA/IUA clause. IUA war clause listing.
International Labour Organization (ILO), International Maritime Organization (IMO), and United Nations Economic Commission for Europe (UNECE). 2014. Code of Practice for Packing of Cargo Transport Units (CTU Code). Non-mandatory global code. UNECE CTU Code overview.
National Hurricane Center (NHC), National Oceanic and Atmospheric Administration. n.d. NHC Data Archive: Best Track Data (HURDAT2), Tropical Cyclone Reports and GIS. Accessed November 17, 2026. NHC archive and data.
Swiss Re. 2016. Global Insured Losses from Disasters Reach USD 37 Billion in 2015; Tianjin Explosions Cause Biggest Insured Loss. Zurich, March 30, 2016, press release accompanying sigma no. 1/2016. Swiss Re Tianjin findings.
United Kingdom. 1906. Marine Insurance Act 1906, section 55, included and excluded losses. Marine Insurance Act section 55.
United Kingdom. 2015. Insurance Act 2015, section 3, duty of fair presentation; non-consumer insurance statutory scope. Insurance Act section 3.
United Nations Commission on International Trade Law (UNCITRAL). n.d. United Nations Convention on the Carriage of Goods by Sea (Hamburg, 1978). Convention overview; effective November 1, 1992. Hamburg Rules convention overview.
United Nations Commission on International Trade Law (UNCITRAL). n.d. Status: United Nations Convention on the Carriage of Goods by Sea (Hamburg, 1978). Participation and entry-into-force dates, accessed November 17, 2026. Hamburg Rules status record.
United Nations Commission on International Trade Law (UNCITRAL). n.d. Status: United Nations Convention on Contracts for the International Carriage of Goods Wholly or Partly by Sea (Rotterdam Rules, 2008). Entry-into-force threshold and participation, accessed November 17, 2026. Rotterdam Rules status record.
United Nations Conference on Trade and Development (UNCTAD). 2025. Review of Maritime Transport 2025: Staying the Course in Turbulent Waters. UNCTAD/RMT/2025. Published September 24, 2025. UNCTAD maritime transport review.
World Bank, European Union, and United Nations. 2020. Beirut Rapid Damage and Needs Assessment: Q&A. August 30, 2020, public research summary. World Bank Beirut assessment.
World Shipping Council. 2026. Containers Lost at Sea Report 2026. Results for calendar year 2025, published June 2026. WSC containers lost report.
Kaczmarska, Jo, Stephen Jewson, and Enrica Bellone. 2018. Quantifying the Sources of Simulation Uncertainty in Natural Catastrophe Models. Stochastic Environmental Research and Risk Assessment 32 (3): 591–605. Published online February 18, 2017. DOI: 10.1007/s00477-017-1393-0. Springer study record.