RESEARCH DOSSIER
Marine and Aviation Insurance Exposure Data Quality
Abstract. Marine and aviation insurance exposure data describe insurable interests whose physical objects, journeys, legal responsibilities, locations, valuations, and applicable contracts change through time. Their reliability depends on correctly joining several different kinds of observation: a ship or aircraft registry identifies an asset; a tracking or flight record describes operational activity; a cargo document identifies a consignment; an occurrence database records a safety event; an insurance contract determines the insurer's obligations; and a claims record documents a potentially indemnifiable consequence. None of these records alone establishes the whole insured exposure. This monograph develops a source-critical international analysis of the quality, completeness, provenance, temporal validity, harmonization, and statistical use of these joins. It reviews IMO vessel identification and casualty systems, AIS limitations, EU maritime reporting, ICAO ADREP and aircraft registrations, IATA and EASA accident information, freight data exchange, Lloyd's transaction standards, and EU actuarial data-quality obligations. Its original contribution is an integrated analytical account of how measurement error propagates from entity matching and exposure denominators into accumulation, pricing, reserving, and reinsurance decisions. Reproducible synthetic calculations illustrate missing-location bounds, errors in earned exposure, and duplicate counting of cargo legs. The study finds that risk data quality must be defined against a specified insurance decision, at a specified historical time and for a specified unit of account. A defensible exposure register therefore preserves legal and physical identities, policy versions, evidence provenance, original observations, uncertainty intervals, and explicit reconciliation to premiums, claims, and reinsurance shares. The analysis makes no claim to a new theory, a proprietary empirical dataset, or observed market-wide error prevalence.
Keywords: Marine insurance; aviation insurance; exposure data quality; insured interests; vessel identification; aircraft registration; AIS; ADREP; cargo data; missing data; exposure denominators; event aggregation; data lineage; claims linkage; probabilistic sensitivity; solvency; reinsurance; insurance risk science.
Introduction
The financial meaning of an insurance exposure is inseparable from the terms under which a risk is transferred. A shipping register can list a vessel that has been sold, a tracking feed can show its last received location, and a cargo manifest can describe goods aboard; none proves that a particular insurer has an active share of a particular loss. Equally, an airline's flight schedule may show a planned sector, while airport movement records show a completed sector and an annual safety report counts only occurrences meeting a defined accident classification. An insurance analyst must reconstruct the connections among these records before estimating frequency, severity, catastrophe accumulation, or claims-paying requirements.
Marine and aviation classes share mobile assets, cross-border operations, time-dependent hazards, valuable concentrations at ports and airports, and complex layers of insurance and reinsurance. They diverge in the underlying business units: hull and machinery, cargo, liability and protection and indemnity in marine markets; aircraft hull, passenger and third-party liability, aviation products and ground risks in aviation markets. The field therefore needs an account of data quality that treats both common computational requirements and line-specific contract economics.
A general foundation is offered by Wang and Strong's consumer-centered data quality study. Their 1996 work demonstrated why accuracy by itself is an insufficient account of information usefulness. The later ISO/IEC data quality model, published in 2008 and confirmed as current in the ISO record in 2025, supplies a structured framework for assessing information held in computer systems. For insurance purposes, these concepts acquire operational meaning when an analyst states the decision to be made, the unit of exposure, the acceptable latency and error, and the mechanism by which an error can alter an estimate.
Consider a vessel that changes ownership during a voyage, a cargo item that travels through several terminals, or an aircraft that moves from flight to maintenance to storage. These are ordinary changes in the observed world. If an underwriting database overwrites a previous owner or location with the current value, a historical claim can be joined to the wrong risk state. If a risk system instead retains every snapshot without a unique logical entity key, it can inflate a catastrophe exposure through repeated records. Both errors can occur in the same portfolio.
The original research question is therefore precise: which documented, reproducible links are needed to establish the exposure actually assumed by an insurer at a specified time, and how do failures in those links affect risk measurement? The method is source-critical comparative synthesis of official primary materials, a review of established missing-data scholarship, and mathematical thought experiments using declared hypothetical data. The evidence covers selected international and EU/UK institutional frameworks and globally relevant transport data; it is not a census of insurers' proprietary databases.
This study treats an exposure register as a time-indexed evidentiary record of insured obligations. Its proper object is the conjunction of an insurable interest, a risk-bearing entity and share, specified perils and coverage terms, and the operational conditions that make loss possible. The physical risk unit remains visible but distinct. In that sense a loss model, a regulator, a claims handler and a reinsurance analyst may interrogate the same underlying record while using different projections from it.
Foundations of Exposure and Data Quality
The exposure unit and the unit of account
An insurance exposure is an economically meaningful risk-bearing position under a specified contract and time window. It must identify who stands to lose, what interest can be affected, what event can produce an insured obligation, how much of the loss is covered, and when the responsibility attaches or ends. The unit can be a vessel-year for hull frequency analysis, cargo value in a specified storage interval for port accumulation, aircraft flight hours for operational risk, passenger movements for a particular liability model, or an insured contractual limit for certain liability scenarios. A unit appropriate for one calculation can be invalid for another.
A reporting system must therefore record separately the policy unit, insured interest, operating asset, journey or flight segment, insured location, event, claim, layer and insurer participation. A ship identifier is not a substitute for a cargo shipment identifier; a passenger count is not a proxy for the hull value; and a policy occurrence is not necessarily an accident under an aviation safety classification. The distinction follows directly from the differences between the marine economic interests in section 3 of the Marine Insurance Act 1906 and the operational information collected by transport authorities.
For a particular risk contract, an exposure record can be represented analytically as a tuple E = (i, a, j, c, t, v, s, q), where i is the insurable interest; a is the asset or operation; j is the journey or location history; c is the contract and version; t is its effective time interval; v is the economic value or relevant liability measure; s is the insurer's participation; and q is the provenance and quality description. This tuple is an explanatory representation used in the present analysis, not a proposed industry schema or newly authored theoretical model. A separate claim may reference one or more such tuples when several interests or policies respond to one occurrence.
The analytical denominator must be specified before any aggregation. The annualized number of insured vessels, vessel operating days, flight sectors, flight hours, cargo shipments, cargo-value-days, declared insured value, and total aviation passengers all have different dimensions. A loss ratio has still another denominator, earned premium, that reflects commercial pricing and accounting rather than a direct physical hazard duration. Whenever an estimate silently changes its unit, even perfectly transcribed data can give a substantively wrong answer.
Quality as fitness for a governed use
Accuracy tests whether values conform to a trusted observation; completeness asks which necessary values are available; timeliness tests whether they exist soon enough for the decision; consistency asks whether sources use compatible values and definitions; traceability records their origin; representativeness concerns which risks and events are included; and accessibility determines whether a user can legally and technically reproduce the decision. These dimensions apply to individual fields, relationships between fields, and the population represented. ISO/IEC 25012 defines a general structured-data framework whose intended uses include the acquisition, integration, evaluation and improvement of data quality; the present study selects insurance-specific tests from that broad framework.
Intrinsic accuracy and decision-specific fitness may conflict. A vessel's latest location can be accurate at 18:00 UTC and irrelevant to a loss occurring at 09:00 UTC. A correctly stored flight number can correspond to multiple operational dates, while a correctly formatted cargo code can identify a shipment that has been divided into separate insured consignments. Quality is accordingly assessed over keys, times, cardinalities, source versions and join assumptions rather than over isolated non-null fields.
A crucial distinction is between unobserved, invalid, inapplicable and withheld information. An unknown cargo value requires estimation or supplementary documents; a blank field for an inapplicable passenger liability exposure may be exactly correct; an AIS gap may reflect communications limits or a legitimate period without transmission; and a confidential accident report may be known to an authority yet unavailable to the insurer. Treating every blank as zero systematically changes insured totals. Treating every blank as a data defect obscures the legally justified absence of information.
Measurement and decision error
An incorrect IMO number can join a claim to an unrelated hull. A missing port coordinate can remove insured cargo from a coastal accumulation. A stale hull valuation can understate a modeled replacement-loss limit; the reverse may happen where an agreed contractual value differs from market sale value. A duplicate air cargo leg can double the amount at risk. An incomplete flight-exposure denominator can overstate an observed accident rate if the numerator still includes the whole operational population. The specific direction of distortion depends on the joint behavior of numerator, denominator and record selection.
The correct unit for quality assurance is often a relationship: vessel–policy, voyage–cargo, aircraft–flight, event–claim, claim–reinsurance treaty, or port–time interval. These relationships have evidence requirements of their own. Their audit record should include the original raw identifier, normalization rules, confidence of match, effective timestamp and human adjudication where multiple possibilities remain. A quality dashboard limited to syntactic validity can pass every row and still hide these economically material failures.
Historical and institutional context
Longstanding marine underwriting dealt with named ships, routes, bills of lading, declared interests and the uncertain arrival of goods. Aviation underwriting added recurrent high-frequency movements, fleet declarations, liability passengers and significant ground concentrations. Today's electronic systems can merge these records at speed, but formal harmonization did not eliminate the diversity of underlying legal interests. The distinction is especially important for the present international comparison because the UK marine legislative definition, global maritime safety reporting, European occurrence reporting, IATA operational statistics and private London-market placement data serve different institutional functions.
Contemporary institutions recognize the need for consistent transaction data. Lloyd's Core Data Record captures critical data at the written line to support premium settlement, first-notice claims matching, tax and regulatory processes. The Lloyd's Market Association also expressly identifies exposure data quality and modeling among its management concerns. These initiatives strengthen market processing, while the insurer remains responsible for demonstrating that the recorded business corresponds to a real covered economic risk.
Marine Exposure Registers and Operational Evidence
Vessel identity and changing attributes
The IMO vessel number is an unusually useful reconciliation key because the IMO identification number scheme describes a seven-digit number that stays with the ship and is never reassigned to another vessel. The scheme has coverage criteria and exclusions. Databases also contain names, flag, call sign, management, owner, classification status, dimensions, gross tonnage, build information, machinery, operating area and prior incidents. Several of these descriptors change during the asset's life. The permanent identification number can therefore link historical records even when a ship is renamed, reflagged or sold.
A robust register retains both the persistent IMO identity and date-bounded descriptions. It records the source for each attribute and the difference between observation date and effective date. A class notation seen in a current registry cannot be assigned retrospectively to an earlier voyage without checking historical class records. Ownership and bareboat-charter relationships cannot be inferred from ship name or flag alone. Even when the ship's IMO identifier is reliable, the insured interest and contracting party may attach to a different legal person.
A single hull may support several concurrent positions: owner hull and machinery cover, mortgagee interest, disbursements, charterers' interest, protection and indemnity liabilities and cargo under separate contracts. A database that collapses all positions into one vessel insured value loses the distinction between physical damage, earnings, costs and liability. Conversely, separate records for those interests should reference one common asset identity when modeling a collision or port event. This is the first major tension in marine exposure quality: insured interests are many; the underlying casualty can be one.
Tracking data, voyages and missing positions
IMO's AIS carriage and functionality description identifies the categories of ships required under SOLAS to carry Automatic Identification Systems and the kinds of information the technology transmits, including vessel position, course, speed and navigational status. An AIS record is a dated transmission or derived observation, not an indemnity statement, not an authoritative cargo inventory and not a guarantee of global continuous visibility. Coverage exemptions, legitimate switching off, geographic reception, quality of inputs and gaps in transmission affect completeness.
The analyst should distinguish an observed point, an interpolated path, a reported destination, a planned voyage and an independently corroborated port call. For moving accumulations, an interpolation may be adequate to locate a broad oceanic zone over a short interval; it can be unacceptable for deciding which insured vessels were inside a particular harbor at an exact catastrophe time. An operational status field marked 'at anchor' should be corroborated with location and timing where the exact exposure concentration matters. The measurement purpose determines the acceptable error.
A voyage record should preserve departure and arrival times in UTC, local-time source and time-zone conversion, route legs, transshipment or diversion, and contractually significant attachment or termination events. If a ship sails across a date line, naive local-date joins can mistakenly create a negative duration or duplicate a day of coverage. If an AIS stream ends before the arrival, the exposure interval remains subject to evidence from the charterparty, terminal logs, marine policy or other documents. An absence of signal is not evidence that the insurance ended.
Port codes and names require similar care. The UNECE's UN/LOCODE location system facilitates internationally comparable identification of trade and transport locations. The insurance record still needs a location's operational boundary: port complex, terminal, berth, bonded warehouse, airside cargo facility or inland distribution center. A standardized code that points to an entire city does not establish a precise catastrophe footprint. Historical code revisions and local aliases must be retained.
Cargo movements and accumulation
Cargo insurance is organized around goods, financial interests, transit stages, valuations and terms of cover. A shipping container is a useful logistics unit, but it may contain multiple consignments with separate insured interests; one consignment can be split among containers; and an open cargo policy can insure many successive shipments. A bill of lading or booking reference helps identify movement but can differ from policy certificate, invoice, packing list and claim reference. Quality controls must document the cardinalities among these identifiers.
The critical exposure question at a port is which covered values were present in the relevant hazard footprint during the event window. A record of 100 shipments arriving during a month does not show 100 concurrent shipments. Cargo-value-days can summarize time at risk, while a peak-location sum identifies a scenario's simultaneous values. Terminal dwell-time omissions can drive underestimation of an extreme accumulation. Duplicate representations of the same container as inbound, stored and outbound can create a different, upward bias.
International freight-data modernization improves interoperability without determining the insured interest. The European Commission's eFTI regulatory framework concerns standardized electronic freight information for regulatory exchanges. A transport data field becomes insurance exposure evidence only after appropriate coverage and valuation joins are made. Multiple cargo modes demand particular attention: a multimodal consignment may travel sea, road and air legs, while liability and cargo policies attach over different windows and interests.
The original 2009 Institute Cargo Clauses (A) register provides a traceable edition for a commonly referenced contractual form. The existence of a market wording does not show that every cargo policy incorporates that edition unchanged. Each insurance record requires the actual clause incorporated, endorsements, declared valuation basis, insured transit limits, exclusions, deductibles and the legally operative document version. This avoids applying a current template retrospectively to a loss under an earlier wording.
Casualty evidence and selection boundaries
The IMO GISIS Maritime Casualties and Incidents database includes reported occurrences and investigations. IMO's harmonized reporting procedures connect reporting duties to international instruments, but some GISIS information is restricted to IMO Members. The public access path for investigation material can require an IMO web account. A research design that depends on a member-restricted field cannot assert independent public reproducibility on the strength of a public landing page.
The EU's marine reporting architecture introduces additional inclusion conditions. The European marine accident investigation directive sets out reporting to EMCIP and contains specific treatment of smaller fishing vessels in its consolidated text. EMSA's annual maritime casualty overview accordingly describes maritime incidents involving EU interests and reporting rules, rather than all insured ships worldwide. The EMCIP publications disclaimer specifically cautions about completeness, accuracy and currency of the displayed statistics.
The implications are methodological. A reported marine casualty may involve an uninsured vessel, an excluded peril, an insured cargo interest under a different company's policy, or several claims under one common occurrence. Conversely, a covered machinery claim may not correspond to a casualty reported to an international safety database. A count of incident reports therefore requires a matching exposure population and contractual filter before it can be interpreted as an insurance claims frequency.
Aviation Exposure Registers and Safety Evidence
Aircraft, operators, flights and fleet state
ICAO's nationality and registration mark guidance reflects the aircraft registration framework established under Annex 7. Aircraft marks identify the State and registration of a civil aircraft at a given time. They should be treated as legal registration attributes with effective dates, supplemented by the manufacturer's serial number, aircraft type, operator, maintenance/airworthiness records, lease arrangements and actual fleet assignment. A policy may follow an identified fleet, specified aircraft or a declaration of additions and deletions, rather than automatically following every registration mark seen in an external feed.
For aviation hull insurance, an aircraft-year may represent duration insured, while utilization data capture flight hours or cycles. Aircraft stationed at an airport for maintenance have low airborne frequency exposure while retaining substantial ground catastrophe exposure. Idle, stored, leased, test-flight, training, charter and commercial passenger operations differ in both hazard and contract scope. The register should therefore include each aircraft's time-varying operating status alongside policy cover and the relevant aircraft identification keys.
The flight itself is another entity: operator, carrier, flight number, scheduled date, actual origin/destination, aircraft used, departure/arrival, and relevant operational category. Commercial flight identifiers recur, schedules change, and a delayed flight may cross a calendar boundary. A system that treats flight number as a unique key can merge two separate sectors; a system that treats the schedule as the flown operation may include canceled flights. The correct denominator for a particular model demands records of actual eligible movements, their covered risk intervals and the business segment.
Passenger count, available-seat kilometers, ton-kilometers, sectors, hours and hull insured values each answer different questions. Passenger and carrier liability require their own connection between person or consignment, operative carriage conditions and insured liability. The Montreal Convention liability framework applies to defined international air carriage under its treaty conditions; its periodic monetary limit adjustments do not automatically identify an airline's complete insurance limit. The reference notes the December 28, 2024 revision of certain convention limits, which underlines why liability records must carry the event date and relevant legal version.
Occurrence taxonomies and reporting populations
The ICAO ADREP reporting taxonomy organizes aircraft accident and incident attributes with common definitions for event types, operation phases, injury severity and related categories. Its current institutional description acknowledges ongoing taxonomy revision. The ICAO ADREP reporting mechanisms include national reporting and transfers or mapping through ECCAIRS 2. A database harmonization project must therefore preserve taxonomy version and original native values, particularly when an occurrence category changes between editions.
The EU's aviation occurrence reporting regime provides a specific regional regulatory foundation under Regulation (EU) No 376/2014 and amendments. That legal framework is different from a private insurer's claims-notification requirements. Reports may be gathered for safety learning, and the protection of information sources imposes restrictions and safeguards. Data quality in this context includes lawful access, purpose limitation, confidentiality and defensible linkage, as well as correctness of technical fields.
IATA's 2025 annual aviation safety results give an instructive denominator: 51 accidents among 38.7 million flights, reported as 1.32 accidents per million flights within the organization's classification and reporting methodology. These are industry safety statistics, not a hull premium loss ratio, nor a measure of claims per insured fleet-year. Applying this numerator to a smaller dataset of insured aircraft hours would create an undefined rate without first establishing which flights, aircraft and occurrences belong to the same population.
EASA's Annual Safety Review 2026, published in October 2026, analyzes 2025 safety performance across several aviation domains. The differentiation among commercial airplanes, helicopters, general aviation, aerodromes and other categories is analytically consequential: rates from the broadest category should not be assigned to a specialized insurer's fleet without adjustments for operations, geography, temporal coverage and reporting. It is also important to distinguish a 2026 publication date from the 2025 observation period.
Aircraft ground accumulation and catastrophic scenarios
Aviation risk concentration can materialize when aircraft are colocated at a hub, when a maintenance facility contains multiple high-value airframes, or when a hailstorm, flood, fire or other event affects a fleet during a short interval. A current asset schedule may be adequate for annual premium accounting yet poorly matched to an airport catastrophe at 03:00 UTC. The exposure analyst needs time-valid positions, hangar or stand detail when material, aircraft replacement or agreed hull values, and any applicable deductibles, sublimits and participation shares.
The relationship between observed position and covered loss is particularly demanding in aviation: an airframe may have taken off, parked for servicing, been placed under storage cover, or been transferred to another policy. Ground-handling liability may attach to an entity different from the aircraft owner. An event can produce hull losses, passenger or third-party liabilities, airport/operator claims and business interruption under separate contracts. An occurrence graph must connect one event to several valid contract claims without adding the same claim twice at insurer or reinsurer level.
Air cargo and transactional integration
IATA's ONE Record data-sharing standard defines a common model and secured API approach for air-cargo stakeholders. The promise of interoperability is substantial for reconciling consignment, waybill, transport-leg, warehouse and handover information. The schema nevertheless governs logistics exchange. Its presence cannot prove the insured goods' declared value, named interest, attached policy, or application of a particular exclusion. Insurance joins need separate legal and financial validation.
A cargo parcel moving through multiple carriers and customs regimes can have a single economic value and several records of custody. A system should preserve each carriage leg and custody handover while maintaining one logical consignment identity and an explicitly versioned insured interest. Summing the values of all leg records creates an inflated simultaneous accumulation even though it may be a legitimate summary of movement volume. The choice of aggregation must follow the question: volume moved over a period, duration in a specific location, or economic value present at the instant of an insured event.
Contractual Identity, Coverage, and Loss Attribution
Policy lineage and multiple insurance interests
The contractual register is the bridge between operational activity and insurable economic consequences. Its minimum history includes policy or binder identity, applicable insurer or syndicate participation, insured and named interests, coverage terms, inception and expiry, amendments, declarations, premium basis, valuation, limits, deductibles, conditions, sanctions provisions where relevant, and related reinsurance. A change to a location, a vessel or an aircraft may trigger a declaration or endorsement whose effective time differs from the time entered into a system. A claim made after expiry may arise from an occurrence during the policy period. These distinctions make bitemporal recording valuable: effective time describes when a fact was true for insurance; transaction time records when the system learned or recorded it.
A single placement can be shared among underwriting participants. Lloyd's describes its written-line Core Data Record as a transactional source for premium settlement and claims matching, alongside tax and reporting functions. Exposure systems must preserve the difference between 100 percent gross risk, the insurer's signed line, the written line if altered, the insurer's retained net risk, and reinsurer allocation. An aggregation that sums each underwriter's full gross value can overstate the market total. An aggregation that records only one lead underwriter may understate a member's or market's participation.
An asset's registration number cannot resolve these obligations on its own. A vessel may carry an insured interest under a mortgagee clause while another insurer covers hull damage; an aircraft may have different cover while flying and while on the ground; and a cargo owner may depend on a separate open policy whose declarations are reported after shipment. The relevant join is the historically valid contractual relationship to an entity and an event. The analyst should maintain the complete chain from original placement to claims adjudication instead of relying on the latest flat table export.
Liability and policy limits
Liability exposure is contingent on a legal relationship and a theory of responsibility. In marine markets, it may arise from cargo handling, collision, pollution, passenger injury, crew obligations or removal of wrecks. In aviation, it can arise from carriage of passengers or cargo, injury to third parties on the surface, products and airport operations. Each peril has its own attachment conditions, defenses, exclusions and loss allocation. The existence of a ship or airframe and its insured value tells an analyst little about the ultimate distribution of third-party losses.
The International Group of P&I Clubs' pooling agreements and governance demonstrate how large marine liability claims can be shared among clubs according to pooled categories and rules. Claims passing through a club's retention and pooled layers cannot be treated as independent insured losses at every level. For a reinsurance model, the event, individual claim, policy recovery, club pool and external treaty recovery should each carry a unique identity and provenance. The 2026/27 market arrangements may change over time, so a historical claim must use the agreement applying to its own policy year rather than an assumed current structure.
International carriage rights also depend on the legal route and governing treaty. The IATA summary of the Montreal Convention 1999 explains a legal framework for passenger injury, baggage and cargo in qualifying international air carriage. A liability calculation still needs the exact cause of action, forum, event date, carriage facts, any applicable treaty status and the contract's aggregate insurance limits. Treaty limits and policy limits are distinct financial variables and must remain distinct in a well-designed exposure record.
Event, damage, claim and recovery
A transport event is a factual occurrence; a loss is an economic effect; a claim is a demand or notification under a specified arrangement; an indemnity payment is a financial result after coverage analysis; and a recovery is a further allocation of loss. These entities have many-to-many relationships. One maritime casualty can produce several cargo claims and a hull claim; one aircraft accident can produce passenger, ground and property claims; one insured may submit multiple notices relating to a single physical event. Deduplication by similar date, position and vessel name can merge valid separate claims, while failure to group common events can understate correlated tail loss.
The analyst should keep both an occurrence ID and distinct claim IDs. Each financial movement should carry claim-version date, currency, reserve or paid category, allocated expense classification, subrogation or salvage, insurer share, treaty layer, and exchange-rate methodology. A claims database frequently changes as payments and estimates develop. A snapshot used for reserving must be reconstructible as of the valuation date, including late notifications and revisions. Replacing historical reserve values with current estimates destroys the possibility of reproducing a prior reported loss ratio.
A high-quality accident register is useful evidence for event occurrence and safety classification. It is not by itself a complete claims register. Marine casualty information in IMO GISIS and aviation occurrence information reported through ICAO ADREP were developed for safety and investigation objectives. The insurance analyst must document each additional step connecting the underlying event to coverage, loss and payment. This distinction prevents a common methodological error: treating all detected incidents as insurance claims and all claims as officially classified accidents.
Valuation, currency and insured share
For hull risks, the agreed value and the physical asset's market or replacement value may differ. Cargo values can be derived from invoice, freight or specified uplift, with coverage terms determining the operative amount. For liability, the sum insured is a contractual constraint on potential insurer payment and may not measure expected compensation. The exposure schedule should therefore preserve the valuation concept and date rather than placing every economic amount in an unqualified 'value' field.
A uniform reporting currency is necessary for portfolio aggregation, but currency conversion must follow a defined date and purpose. Current-market conversion of a historically agreed fixed-currency hull amount may be useful for today's purchasing-power analysis; it is not a substitute for reconstructing the amount contractually payable at the loss date. The register should separately record original contract currency, original numeric value, conversion source, effective conversion date and presentation currency. Otherwise currency fluctuations may appear falsely as changes in the underlying fleet or cargo risk.
Gross insured sum, net retained share, reinsured share, occurrence limit and aggregate exhaustion should be reconciled with unique placements and claims. Deductibles operate at specified levels and may depend on a per-vessel, per-incident, per-shipment, per-aircraft or policy-wide definition. Applying a deductible to each data row, when several rows form one insured occurrence, changes modeled recovery even when the total underlying damage remains the same. An exposure table is only actuarially meaningful when its fields contain sufficient contractual detail to reproduce the correct level of financial aggregation.
Harmonization, Provenance, and Temporal Reconstruction
Data interfaces and semantic matching
Technical interoperability requires a common representation of transport entities, insurance interests, time and geography, together with a stable mapping to the original evidence. It can reduce manual reconciliation and improve processing speed. A common data model also creates new opportunities for systematic error when a source field is mapped to an apparently similar but economically different destination field. A 'flight' may be a scheduled flight number in one system and a completed departure sector in another. A 'shipment' can mean the commercial sale, consignment, container unit or carriage document. A 'loss date' may mean accident occurrence, discovery or notification.
A useful mapping therefore has five documented layers: original field and definition; normalized field; transformation rule and version; legal or business meaning; and validation evidence. This method accommodates transport standards without converting them into insurance law. The UN/LOCODE reference system, IATA ONE Record model, and Lloyd's Core Data Record address different portions of the information chain. No one of them establishes a universally complete combined marine-and-aviation insurance exposure ledger.
Version discipline matters. A code may have been reassigned, an organizational name changed, or a port subdivided. The latest master dictionary is unsuitable as the sole reference for all historical transactions if the same label represented a different object at an earlier date. An event-time join should use the dictionary version in effect at the event or a documented historical crosswalk. Effective time for a transport movement, contract amendment and source update can be different, and the system should represent each separately.
Identity confidence and uncertainty
Matching can be deterministic when a stable identifier and time-consistent attribute provide sufficient proof. A vessel identified by IMO number can have substantially more reliable longitudinal linkage than a vessel identified only by a normalized name; matching an aircraft by registration marks alone still requires relevant historical validity checks. For cargo, links through invoice, waybill, container, purchase order and insured declaration often require explicit many-to-many handling. For names, addresses and manually entered locations, fuzzy similarity alone does not establish legal or physical identity.
When deterministic matching fails, probabilistic or rule-based entity resolution can rank candidate links, but the resulting confidence should stay visible. A model can assign a record to 'unknown vessel' or 'unknown terminal' rather than forcing a high-confidence classification with insufficient evidence. Human review can adjudicate material high-value conflicts, with the decision preserved as an audit event. A threshold optimized for routine administrative convenience may be unacceptable for catastrophe concentrations, where one false negative can remove a large risk and one false positive can duplicate it.
The key scientific distinction is between unknown identity and unknown exposure. A vessel's position may be missing even though its insured value and policy are certain. An airline flight's precise aircraft may be unknown while the airline's contract and operational movement are known. An unidentified cargo consignment could be known to be in a warehouse from inventory totals. Separate uncertainty fields permit analysts to use partial evidence and produce bounded estimates instead of discarding entire records.
Chronology and auditability
A defensible system preserves historical snapshots and source lineage. Raw records should be retained, where lawful, with source URI or document reference, extraction time, validation results, version, access restrictions and a checksum or other integrity marker. Normalized records should point to the originals and record transformations. Corrections should be reversible and logged. An analyst reconstructing a 2024 accident in 2026 must be able to recover what was known at each decision date and which later corrections became available.
This requirement is especially salient for liability and claims data. A case's legal classification, financial reserve and policy allocation may change following investigation or judgment. Using the latest outcome in a historical pricing exercise can introduce look-ahead bias. It is therefore useful to distinguish 'as-was' portfolio snapshots from 'as-known' claims development at successive valuation dates. Neither series should be silently rewritten when new records arrive.
Temporal validity should include precision as well as a timestamp. A port call may have an exact arrival moment, a date-only manifest timestamp, or an estimated time interval based on incomplete tracking. A safety report can publish a time in local civil time that requires conversion to UTC. If the uncertainty spans an insured event, aggregation should account for both plausible states. A categorical answer extracted from a single interpolated point would assert more precision than the evidence supports.
International interoperability and lawful access
Cross-border marine and aviation data incorporate distinct regimes of safety reporting, commercial confidentiality, insurer contracts and personal information. ICAO defines common ADREP terminology; IMO promotes marine reporting; EASA and EMSA operate regional legal and statistical systems; IATA and private industry bodies develop exchange infrastructure. The existence of standard fields does not confer permission to combine all sensitive records. EASA's information-source protections demonstrate that the legal right of access belongs within the quality assessment.
An insurer can meet certain needs with restricted aggregated statistics, pseudonymized identifiers or authorized record linkages rather than unrestricted personal or security-related data. Where an upstream dataset permits only statistical use, underwriting deployment or claims matching may require another legal basis or different data source. 'Accessible' must mean accessible for the lawful analytic purpose, not merely downloadable at a technical endpoint. Auditors should document the source license, permitted uses, retention, disclosure limits and audit trail for each input.
Missing Data, Bias, and Statistical Inference
Missingness mechanisms
Incomplete exposure data arise from operational reporting gaps, delayed bordereaux, nonresponse, confidential documents, source coverage limits, changes in insurer systems, and errors during data translation. Rubin's 1976 analysis of missing data provides a foundation for understanding why missingness may or may not be ignorable in statistical inference. The later Little and Rubin treatment places missing-data methods in a coherent likelihood and applied-statistical framework. These sources support a distinction between missing completely at random, missing at random conditional on observed data, and nonignorable missingness, subject to precise model assumptions.
Marine and aviation systems offer plausible mechanisms for nonrandom missingness. Smaller or older ships may be less consistently represented in some operational feeds; particular trading routes may suffer observation gaps; records for complex losses may enter claims systems late; and sensitive aviation incidents may be subject to restrictive access. Consequently a missing-location or missing-value indicator can itself correlate with loss severity, operating environment, commercial line or regulatory region. Generic mean imputation risks suppressing tail exposure if high-risk records are disproportionately missing.
A missing-data test must specify the source population. If the insurer has a complete policy schedule but incomplete position data, positions may be imputed conditional on known policy and movement characteristics. If the schedule itself excludes a set of vessels or aircraft, imputing fields in the observed rows cannot restore the missing insured population. Reconciliation to independently derived policy totals, premium registers, aircraft or fleet declarations, and bordereaux is essential.
Numerator and denominator selection
The elementary frequency estimator is observed qualifying events divided by the exposure at risk. This expression only estimates a meaningful rate if the event definition, covered population and time interval are aligned. A marine casualty population may cover EU-relevant reportable events while the vessel denominator comes from a global insured hull book. An aviation numerator may count IATA-classified commercial accidents while the denominator is all aircraft hours from a different sector. No algebraic correction can repair this mismatch without evidence about the populations.
IATA's 2025 classified accident statistics illustrate this issue: the reported 1.32 accidents per million flights follows from 51 selected accidents in 38.7 million flights. Those counts are informative for their defined commercial aviation population. Using the figure as an actuarial claim frequency for a general aviation hull portfolio would require unsupported assumptions about comparable operations, coverage, event severity and insured exposure. EASA's multi-domain safety review reinforces the reason for partitioning domains rather than pooling unlike operations.
A consistent frequency analysis uses policy-effective exposure in a common unit, the corresponding claims exposure, and a controlled set of eligible events. Earned exposure adjustments should account for policy additions, cancellations, periods of laid-up operation and changes of covered use. Deduplication must occur before event counting; late reporting must be accounted for when determining the observation window. Segment heterogeneity can be addressed by explicit stratification or statistical modeling, with limitations stated where sparse experience makes calibration uncertain.
Coverage error versus measurement error
Missing exposure can bias estimates even when the number of loss events is known exactly. If 20 percent of eligible exposure is omitted from the denominator while events across the full population remain counted, the naive observed rate is 25 percent higher than the rate using complete exposure. Conversely, if missing exposure and claims are omitted together, the reported rate can appear stable while the absolute gross or net concentration is materially understated. The direction of error is thus conditional on which records have been lost and how missingness correlates with hazards.
Measurement error also affects explanatory variables. An aircraft reported as stored when it was flying can be assigned the wrong hazard basis. A port coded only to a city center may fall outside a flood footprint even if the insured warehouse was directly exposed. An assumed ship speed derived from a sparse AIS track can be wrong at a point of collision. These errors can be differential: high-risk routes and ports may be observed differently from low-risk ones, changing regression results or predictive classifications.
Event attribution errors are another source of bias. A single catastrophe can be represented as multiple safety incidents and several insurance claims. For pricing, counting its distinct claims may be appropriate for individual claim frequency; for aggregate catastrophe analysis, the common event should be preserved. Merging claims with distinct policy coverage because they share a port and date can also conceal recoveries. Sound inference requires a declared unit and controlled many-to-many joins.
Sensitivity and defensible bounds
When an uncertain field is consequential, a single imputed value should not be treated as a certainty. Multiple plausible scenarios can represent alternative locations, utilization histories, classifications, values and reporting mechanisms. Bounds can be calculated from known insured totals and plausible allocations of the unknown records. A decision can then be described in relation to a range of potential modeled loss rather than a falsely precise point estimate. The range itself depends on explicit physical and contractual assumptions.
For sparse lines, external operational data can provide useful context without being substituted directly for insurer claim rates. Public safety statistics can support hazard hypotheses and identify candidate event types; insured loss data must determine policy-level recovery, development and deductibles. Where the underlying proprietary claims data cannot be accessed, the correct scholarly output is a documented analytic structure and an illustrative calculation rather than a market-wide empirical coefficient.
The EU prudential framework expresses a relevant governance obligation. Under Solvency II Article 82, insurers and reinsurers must have internal processes to ensure appropriate, complete and accurate data for technical provisions, and specified circumstances permit approximations where adequate data are unavailable. The actuarial function requirements require examination of sufficiency and data quality. These provisions apply in their legal jurisdiction and purpose; they do not prescribe one universal global scheme for every marine or aviation catastrophe input. Their significance for the present analysis is the principle that uncertainty is to be evaluated, documented and governed.
Reproducible Analytical Demonstrations
The following examples are synthetic and constructed solely to test consequences of data definitions. They are not observations from a particular insurer, port or fleet. All amounts are hypothetical U.S. dollars; no market-wide frequency or severity estimate is inferred. Each calculation separates physical exposure, contractual share and reporting unit.
Vessel-years and a missing denominator
Suppose a hull book includes five separately insured vessels. Three are covered continuously for 365 days in a conventional 365-day analytical year, a fourth for 180 days and a fifth for 90 days. Gross insured vessel-days equal 3 × 365 + 180 + 90 = 1,365 days. Earned exposure is 1,365 / 365 = 3.739726 vessel-years. If a data extraction omits the fifth vessel and its 90 insured days, the derived denominator becomes 1,275 / 365 = 3.493151 vessel-years, an omission of 6.5934 percent of actual vessel-days.
If the complete book has four qualifying hull claims, the complete-book rate is 4 / 3.739726 = 1.06960 claims per vessel-year. Holding the numerator at four while using the incomplete denominator gives 4 / 3.493151 = 1.14510, an upward rate distortion of 7.0588 percent. Both numbers are too unstable to extrapolate from this toy portfolio: the example demonstrates the denominator effect only. If the omitted vessel's claims were also omitted, the relative distortion would depend on its actual claim count and could differ in sign.
More generally, if complete exposure is E and the observed extracted denominator contains (1 − p)E while the numerator N still represents all exposure, then the observed rate N/[(1 − p)E] exceeds the correct N/E by the factor 1/(1 − p). At p = 0.20 the factor is 1.25. This identity says nothing about real missingness prevalence: p is a hypothetical assumption whose empirical value must be established separately.
Port accumulation with unknown locations
Consider 100 insured cargo consignments, each hypothetically valued at $2 million for a single catastrophe aggregation, with no overlapping consignment identities. Location evidence places 12 consignments inside a defined port surge footprint, 68 outside it and leaves 20 without an event-time location. With no additional constraints, the verified lower sum insured in the footprint is 12 × $2 million = $24 million. An extreme upper sum, if all 20 unknown consignments were inside, is (12 + 20) × $2 million = $64 million. These bounds concern gross insured consignment values, not expected losses or immediately payable claims.
If the insurer's consistent participation is 60 percent on each covered consignment, the corresponding insurer-attributed gross-share values are $14.4 million and $38.4 million. This 2.6667 ratio arises entirely from unresolved event-time positions. Applying a hypothetical 50 percent physical damage ratio before policy deductibles would produce $7.2 million and $19.2 million at the 60 percent share. The calculation is a sensitivity demonstration; no empirical damage probability, contract deductible or reinsurance recovery is assumed.
The range can be narrowed by using historical berth, warehouse, customs and transport handover records. An unreliable point location can be replaced by a time interval and plausible set of storage areas, then intersected with the event footprint. Each additional constraint should be supported by evidence rather than inferred from the same missing feed. The analyst should also recognize that the 20 unknown consignments may be systematically concentrated at a particular terminal, making a random-location assumption inappropriate.
Multimodal leg duplication
Suppose one $10 million insured industrial consignment moves through four transport segments: warehouse-to-port road movement, sea carriage, port storage and an inland rail leg. Four logistics records legitimately describe the movement, each containing the same declared cargo value. Adding the values of the four records gives $40 million, a correct sum of values across segment records only for a deliberately defined cumulative segment-volume statistic. It is incorrect as a measure of four independent goods interests or as simultaneous value in one location.
If a model asserts that the goods were present in the marine port during a certain six-hour storm window, it should count $10 million at that location if the evidence and coverage support that state. If the consignment is insured under several policies, coverage overlap and each insurer's contractual share must be examined before estimating each insurer's response. A normalization rule that simply deduplicates all repeated cargo identifiers globally can also be wrong when the original consignment was legitimately split or sold during transit. Proper aggregation uses insured economic interest, time and location rather than a mechanical global unique-row count.
Safety rate and the wrong population
The IATA 2025 figures report 51 classified accidents and approximately 38.7 million flights. Arithmetic produces 51 / 38.7 = 1.317829 accidents per million flights, which rounds to the officially reported 1.32. Replacing the denominator with a hypothetical 40 million while retaining the same event count produces 1.275 per million, illustrating sensitivity to the denominator. It does not establish a correction to the IATA report, because 40 million is not an observed alternative count for the same population. The defensible analytical conclusion is that denominator scope and method must accompany every published accident rate.
An insurer could translate operational frequency into claim cost only after demonstrating comparability of the covered population, inclusion criteria, exposure unit, severity and policy obligations. The transport safety numerator and the insurer's claim numerator may differ dramatically. The demonstration therefore supports an audit requirement rather than an empirical insurance pricing coefficient.
Event-level aggregation and insurance shares
Suppose a hypothetical event produces two separate hull losses of $6 million each and a cargo loss of $4 million. Total underlying physical damage is $16 million. If the same insurer covers 40 percent of each hull loss but only 25 percent of the cargo loss, its pre-deductible gross-share amount is 0.40 × $12 million + 0.25 × $4 million = $5.8 million. A report that attributes all $16 million to the insurer ignores participation. A report that adds the $16 million market loss to the insurer's $5.8 million internal share double-counts the same underlying event at two accounting levels.
Further contractual adjustment requires deductible and limit information at the correct occurrence or policy level. If an event spans several claims, assigning a single deductible repeatedly to every data row can produce the wrong recovery; combining different insured interests under one deductible can likewise produce an error. Only the explicit policy terms can establish the financial result. This example demonstrates the necessity of linking occurrence, claimant, policy, participation and financial calculation while keeping their identities separate.
Reproducibility of the examples
All scenarios use ordinary arithmetic. In a research audit the source table, units, assumptions, formula and rounded result should be retained with the manuscript; a change to an assumption must produce a versioned new result. No Monte Carlo calibration or empirical loss distribution is claimed. The exact values can be reproduced independently from the displayed counts and percentages, and the assumptions are sufficiently explicit to permit disagreement about interpretation without losing the calculation's traceability.
Governance, Validation, and International Comparison
Stewardship throughout the exposure lifecycle
A reliable exposure register requires named responsibility at every stage of its production and use. The insured or transport operator originates factual descriptions of assets, activities and movements; brokers and coverholders transmit declarations; underwriters attach terms, shares and conditions; claims teams record reported events, liabilities and recoveries; actuaries construct analytical denominators and groupings. Each function may operate a legitimate system with a different purpose. A data-quality program succeeds when the organization reconciles these purposes through traceable shared identifiers and explicit ownership of conflicting attributes.
For a vessel, operational stewardship may start with the permanent IMO number and registry evidence. The underwriting steward must separately establish the legal assured, insured interest, vessel specifications relevant to contract conditions, and the time for which exposure is carried. For cargo, the logistics steward may own the bill of lading and consignment movement, while the insurance steward owns the declaration of value, insured transit scope and contribution of each subscriber. For an aircraft, a registration or serial identifier is joined to the relevant operator, fleet schedule, coverage endorsement and effective interval. A claim handler must preserve the original incident evidence and the successive adjustment decisions without retroactively changing the underlying event record.
The distinction between data producers and controllers is relevant to both completeness and confidentiality. A carrier can supply precise traffic counts while lacking authority to provide individual claim reserves. An accident authority may have richly documented causal findings while being subject to restrictions on the use of occurrence reports. A private insurer might observe a much narrower but contractually specific claim population. Agreements governing permitted uses and transformations of source data are therefore part of the technical design, rather than a separate afterthought. European aviation protections for occurrence information make the legal dimension of reuse especially visible.
Validation controls tied to substantive error mechanisms
Validation should proceed from simple, high-frequency failures toward harder, risk-specific reconciliation. First, schema checks confirm types, mandatory fields, permitted categorical values, units, currency codes, coordinate validity and date formats. Second, entity checks compare potentially duplicated identifiers with stable object attributes and document provenance. Third, temporal checks reject impossible or unexamined overlaps, such as one aircraft described as operating mutually incompatible flights at the same time, while allowing legitimately simultaneous overlapping insurance policies. Fourth, contract checks match coverage and participation schedules to effective dates, endorsements and insured interests. Fifth, event checks require coherent loss dates, locations, claim references and causal links before a catastrophe aggregation is accepted.
These checks require tolerances appropriate to the observation. A vessel may have a valid reported port-call estimate that differs by hours from a tracking timestamp because of voyage disruption. An aircraft may temporarily disappear from a public tracking feed without any change in legal exposure. A missing position creates a problem for a particular geographical scenario but does not by itself prove that the vessel or aircraft was uninsured. When a rule fails, the remediation record should identify the field, the affected source and interval, the consequence for the particular decision, the correction authority and the version of the analytical output that used the uncorrected data.
Data validation must also compare aggregate amounts with trusted financial records. A reconciled marine premium total alone cannot confirm that all vessel-days or cargo values are correctly represented; one omitted risk could be offset by a duplicated risk. Equally, accurate vessel counts can conceal erroneous policy shares. Independent control totals should therefore compare at least insured objects, effective insured time, declared values, premium transactions, claims and recoveries wherever the underlying record systems support those totals. The method of reconciliation must state its coverage population, currency treatment and accounting cutoffs.
Surveillance of change and data decay
Exposure records change continually. Vessels are sold, renamed, reflagged, withdrawn from class or altered; consignments move among terminals; aircraft change registration, operator, maintenance status or permitted use; policies are amended or canceled; claims are reopened. A quality score calculated only at original policy inception will decay as these facts change. The appropriate audit frequency depends on the rate of change, financial materiality and catastrophe time scale. High-value transit cargo may need event-time declarations; stable long-term hull schedules may be reconciled through a different periodic cadence, supplemented by change-event alerts.
An append-only change log is useful when a reviewer must reconstruct what was known at the time of a pricing decision, an endorsement or a catastrophe. Two dates should be retained where relevant: the effective date of the underlying fact and the date on which the insurer first recorded or corrected it. A late-arriving report must not be silently shifted backward in time in a way that makes a past underwriting decision appear better informed than it was. Reproducible historical evaluation needs knowledge-time as well as valid-time history.
International comparability and its limits
The maritime and aviation regimes supply important common reference points, but the legal and operational meaning of fields depends on the regime and user. The IMO number is a stable vessel identifier in its applicable scheme; it is not a substitute for a complete contractual interest schedule. UN/LOCODE improves location coding, but does not identify a particular warehouse, berth or cargo stack. ICAO aircraft registration and classification data help with identity and occurrence analysis, but they do not identify all exclusions and insured shares. Harmonization must preserve these distinct functions.
A comparison between European accident datasets and private insurance portfolios should retain the respective reporting definitions, coverage populations and confidentiality restrictions. European maritime accident information is assembled through the regulatory reporting and investigation architecture reflected in Directive 2009/18/EC and EMSA's accident investigation publications. Aviation reporting and safety measures are organized under separate ICAO and European aviation frameworks. Neither framework fixes a universal insurance loss denominator. Claims-made and occurrence-based liability policies, vessel voyage and time covers, aircraft hull schedules and cargo open covers can require different exposure-time reconstructions.
Transport data-sharing initiatives can reduce repetitive entry and semantic mismatch. The IATA ONE Record initiative focuses on digital information exchange in air cargo; the European Commission's electronic freight transport information framework addresses electronic logistics information under its defined legal scope. Both may improve the upstream availability and standardization of certain operational observations. An insurer still needs a governed mapping from those observations to the relevant insured interest, declaration, endorsement and claim. Interoperability is a capability of an information system, while data validity for a specific actuarial or legal decision is established through the actual reconstruction and comparison.
The business case for quality investment
The value of remediation can be appraised through the decisions that depend on the affected field. Missing event-time port locations can materially change peak catastrophe accumulation even if the unlocated cargo represents a modest share of the total record count. By contrast, an outdated non-risk-bearing display name may have little effect when a permanent asset identifier and documentary chain remain sound. A decision-oriented quality program ranks errors by their likely effect on underwriting eligibility, pricing, claims reserves, risk concentration, reinsurance attachment and solvency reporting.
The distinction also provides a defensible way to allocate scarce validation resources. A manual audit can target records with unresolved identities and high insured values, especially where their possible locations overlap important catastrophe footprints. The selection must itself be documented; auditing only high-premium or already disputed cases can produce a misleading impression of quality across the whole portfolio. Independent sampling of lower-value and apparently complete cases provides a check on the detection process. The return on quality improvement is best measured through reductions in consequential error, avoidable reconciliation effort, repeated declarations and uncertainty of modeled decisions, rather than by the number of corrected fields alone.
Evidence Appraisal and Scientific Synthesis
Hierarchy of observations and support for claims
The primary sources reviewed for this research fall into materially different categories. Statutes and treaty-related materials can establish legal definitions and obligations within their applicable jurisdiction and time. Regulatory and supervisory documents specify data expectations and the limits of particular prudential responsibilities. Official transport identifiers and occurrence taxonomies describe how safety and operational authorities organize entities and events. Industry clause repositories document selected policy wordings but require contract-specific adoption before any particular coverage conclusion. Academic work on quality and missingness supplies general methods for evaluating uncertainty; those methods do not themselves furnish empirical frequencies for marine or aviation insurance.
Inferences across those categories are therefore conditional. The presence of an accident report is support for an observed or investigated event within the reporting framework, not direct proof of a covered claim. A recorded hull value is an underwriting or contractual amount, not necessarily the market cost of repair. A vessel position can support a geographic scenario, but a single coordinate from a tracking system may not establish cargo ownership, value or insurance inception. The monograph's conceptual and numerical arguments treat these distinctions as checkable relationships between evidence types.
The bibliography gives original institutional and academic records; its entries are dated or labeled as undated where appropriate. The research did not access confidential marine claims files, complete private airline schedules, proprietary catastrophe model inputs or all licensed insurer data feeds. Consequently, it makes no claim to estimating the actual global proportion of defective marine or aviation insurance records, the precise industry loss attributed to data errors or the relative quality of named insurers. The worked scenarios are explicitly synthetic and suitable only for checking mathematical and methodological implications.
Findings from combined maritime and aviation examination
The first finding is that exposure quality is fundamentally time dependent. Insurance risk is carried across intervals, whereas operational sources often report moments, legs or changes. This difference explains why an apparently complete register may still lack the information required to allocate values to a catastrophe footprint at the moment of impact. The correct analytical object depends on the question: an underwriter may require current vessel-years; a port loss analyst needs event-time presence; an aviation frequency study may use flight sectors; an aviation liability reconstruction may need passenger counts, jurisdiction and the terms governing the event.
The second finding is that identity resolution affects both the numerator and denominator of insurance statistics. One lost consignment reported on several transit legs can inflate modeled exposure if the rows are treated as independent goods. An aircraft with several historical registrations can disappear from a fleet sequence if names or tail numbers are used as permanent keys. Conversely, merging all events under a common commercial operator can conceal distinct aircraft, shipments, insured interests and liability policies. Explicit preservation of stable physical identity, changing commercial identity and contractual identity is essential for a defensible aggregation.
The third finding concerns the limits of external operational evidence. Transport authorities and industry bodies provide indispensable identifier standards, occurrence classifications and public statistics. Their populations and purposes differ from the populations and purposes of insurance portfolios. The authoritative status of a transport dataset establishes the basis of the observed transport fact; the insurer must independently demonstrate its suitability for a covered claim, actuarial assumption or capital calculation. This remains true even when both sides use apparently identical labels such as accident, voyage or exposure.
The fourth finding is that missingness matters through its mechanism and concentration. If records are missing at random, standard statistical adjustment may be possible under specified assumptions. If missing vessel locations cluster around disruption periods or missing aviation events correspond to particular kinds of operations, the observed sample can be systematically misleading. Rubin's missing-data analysis and the treatment by Little and Rubin support explicit examination of the missingness process. A portfolio-level completeness percentage alone cannot identify which missingness assumption is justified.
The fifth finding is that insurer-attributed losses require contractual layers. A $16 million physical event and a $5.8 million pre-deductible share in the synthetic example are different valid statistics. They answer different questions. This difference is amplified by overlapping hull, cargo, P&I, aviation liability and reinsurance records. A technically valid data harmonization exercise must preserve the distinction among occurrence, damage, claim, insured interest, policy share and ultimate net loss.
The sixth finding is that data provenance is part of the ability to review a scientific and financial result. A numerical estimate derived from valid inputs but without recorded transformations, assumptions and cutoff dates cannot be independently reconstructed. A reproducible record shows the original source, the version used, the legal or business meaning of each transformation, and the output that was affected. This requirement aligns with the broader quality dimensions identified by Wang and Strong (1996) while acquiring specialized importance in maritime and aviation insurance because exposure moves across jurisdictions, locations and contractual layers.
Outstanding research questions
Several research tasks follow from the evidence. The first is the empirical measurement of defect rates by exposure attribute and portfolio class using representative, appropriately licensed insurer data. A second is the estimation of how identity and location uncertainty propagate through catastrophe models and reinsurance structures, while retaining legal terms as case-specific inputs. A third is comparison of marine and aviation reporting latency and the resulting accuracy of historical exposure reconstructions. A fourth is evaluation of whether practical adoption of interoperable cargo data frameworks improves insurer data fitness after accounting for the quality of contractual joins. A fifth is development of transparent auditing studies that estimate data-error prevalence without selecting only records already flagged by automated systems. These are defined research questions; this monograph does not present them as established new theories or empirical discoveries.
Conclusion
Marine and aviation insurance exposure data quality determines whether a reported vessel, voyage, cargo, aircraft, flight, location or loss can support the specific insurance decision for which it is used. The controlling scientific task is to reconstruct the insured economic interest across the correct entity, place, interval and contractual layer while preserving the provenance of the evidence. Completeness, accuracy, temporal coherence, consistency and representativeness must be tested in relation to that decision.
Marine registers benefit from stable vessel identification, port and transit coding, casualty reports and standardized cargo information. Aviation registers benefit from aircraft identity, flight-sector information, event classification and official safety reporting. Those data systems contribute to a defensible evidence chain; they become suitable for insurance modeling through contract-specific reconciliation, verified time alignment, correct denominators, insured-share allocation and explicit treatment of missingness. The same principle applies when a portfolio spans both sectors, or when an insured event implicates multiple carriers and reinsurance participants.
The comparative analysis and synthetic calculations establish how omissions, duplicates, uncertain positions and mismatched aggregation levels can change loss frequency estimates and insured-value accumulation. They also establish why a measure of transport safety cannot be silently converted into an insurance claim frequency. Internal scientific control should therefore include source-to-claim verification, legal and statistical comparability, error-sensitive scenario bounds, independently reproducible arithmetic and traceable versions of each exposure extraction.
The durable result of this monograph is an integrated, source-grounded assessment of the operational, contractual, statistical and governance conditions for valid marine and aviation exposure records. Its contribution is analytical synthesis and explicit demonstration of consequences of data defects. It claims no new authorial theory, no new general model and no original empirical observation of the global marine or aviation insurance market.
References
Wang, Richard Y., and Diane M. Strong. 1996. "Beyond Accuracy: What Data Quality Means to Data Consumers." Journal of Management Information Systems 12 (4): 5–33. https://doi.org/10.1080/07421222.1996.11518099.
International Organization for Standardization and International Electrotechnical Commission (ISO/IEC). 2008. ISO/IEC 25012:2008: Software Engineering—Software Product Quality Requirements and Evaluation (SQuaRE)—Data Quality Model. Geneva: ISO. https://www.iso.org/standard/35736.html.
Rubin, Donald B. 1976. "Inference and Missing Data." Biometrika 63 (3): 581–592. https://doi.org/10.1093/biomet/63.3.581.
Little, Roderick J. A., and Donald B. Rubin. 2019. Statistical Analysis with Missing Data. 3rd ed. Hoboken, NJ: Wiley. https://doi.org/10.1002/9781119482260.
International Maritime Organization (IMO). n.d. "IMO Identification Number Schemes." Accessed November 17, 2026. https://www.imo.org/en/ourwork/msas/pages/imo-identification-number-scheme.aspx.
International Maritime Organization (IMO). n.d. "AIS Transponders." Accessed November 17, 2026. https://www.imo.org/en/ourwork/safety/pages/ais.aspx.
International Maritime Organization (IMO). n.d. "Casualties." Accessed November 17, 2026. https://www.imo.org/en/ourwork/msas/pages/casualties.aspx.
International Maritime Organization (IMO). n.d. "Reporting." Accessed November 17, 2026. https://www.imo.org/en/ourwork/iiis/pages/reporting.aspx.
European Maritime Safety Agency (EMSA). n.d. "Accident Investigation Publications." Accessed November 17, 2026. https://emsa.europa.eu/accident-investigation-publications.html.
European Union. 2009, consolidated through December 26, 2024. Directive 2009/18/EC Establishing the Fundamental Principles Governing the Investigation of Accidents in the Maritime Transport Sector. EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A02009L0018-20241226.
International Civil Aviation Organization (ICAO). n.d. "Accident/Incident Data Reporting Taxonomy." Accessed November 17, 2026. https://www.icao.int/safety/AIG/taxonomy.
International Civil Aviation Organization (ICAO). n.d. "Accident/Incident Data Reporting (ADREP)." Accessed November 17, 2026. https://www.icao.int/safety/AIG/reporting-and-adrep-data.
International Civil Aviation Organization (ICAO). n.d. "Nationality Marks." Accessed November 17, 2026. https://www.icao.int/nationality-marks.
International Air Transport Association (IATA). 2026. "IATA Releases 2025 Safety Report." March 9. https://www.iata.org/en/pressroom/2026-releases/2026-03-09-01/.
European Union Aviation Safety Agency (EASA). 2026. Annual Safety Review 2026. Published October 22. https://www.easa.europa.eu/en/document-library/general-publications/annual-safety-review-2026.
European Union Aviation Safety Agency (EASA). n.d. "Occurrence Reporting." Regulation (EU) No 376/2014 and amendments index. Accessed November 17, 2026. https://www.easa.europa.eu/en/regulations/occurrence-reporting.
European Union Aviation Safety Agency (EASA). n.d. "Occurrence Reporting: Protection of Information Sources." Accessed November 17, 2026. https://www.easa.europa.eu/en/domains/safety-management/aviation-safety-reporting/occurrence-reporting-protection-information-sources.
International Air Transport Association (IATA). n.d. "ONE Record." Accessed November 17, 2026. https://www.iata.org/one-record.
United Nations Economic Commission for Europe (UNECE). n.d. "UN/LOCODE." Accessed November 17, 2026. https://unece.org/trade/uncefact/unlocode.
Lloyd's. n.d. "The Core Data Record." Accessed November 17, 2026. https://www.lloyds.com/market-resources/requirements-and-standards/core-data-record.
Lloyd's Market Association (LMA). n.d. "Exposure Management." Accessed November 17, 2026. https://lmalloyds.com/specialist-areas/exposure-management/.
European Union. 2009, as amended. Directive 2009/138/EC (Solvency II), Article 82: Data Quality and Approximations for Technical Provisions. EIOPA Single Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2170_en.
European Union. 2009, as amended. Directive 2009/138/EC (Solvency II), Article 48: Actuarial Function. EIOPA Single Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2131_en.
United Kingdom. 1906. Marine Insurance Act 1906, Section 3 (Marine Adventure and Maritime Perils Defined). Legislation.gov.uk. https://www.legislation.gov.uk/ukpga/Edw7/6/41/section/3?view=extent.
International Underwriting Association (IUA). 2009. CL382 Institute Cargo Clauses (A), 1 January 2009, clause registry record. https://www.iua.co.uk/IUA_Member/ClauseDownload?id=2&ordinal=898.
International Group of P&I Clubs. n.d. "Group Agreements." Accessed November 17, 2026. https://www.igpandi.org/article/group-agreements/.
International Air Transport Association (IATA). n.d. "The Montreal Convention 1999 (MC99)." Accessed November 17, 2026. https://www.iata.org/en/programs/passenger/mc99.
European Maritime Safety Agency (EMSA). 2024 (updated May 29, 2026). "Accident Investigation: Annual Overview." https://www.emsa.europa.eu/accident-investigation-publications/annual-overview.
European Commission. 2026. "The eFTI Regulation." Overview of Regulation (EU) 2020/1056 and subsequent implementing instruments. https://transport.ec.europa.eu/transport-themes/logistics-and-multimodal-transport/efti-regulation_en.
International Maritime Organization (IMO). n.d. "Marine Safety Investigation Reports." Accessed November 17, 2026. https://www.imo.org/en/ourwork/iiis/pages/marine-safety-investigation-reports.aspx.