Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Property Insurance Claims Development by Peril and Repair Stage

Abstract. Property insurance loss development is an evolving relationship among physical damage, coverage, discovery of damage, repair production, insurer adjudication, and settlement cash flows. Its statistical representation is frequently compressed into paid and incurred development triangles. That compression remains useful for aggregate reserve estimates, but it can conceal material heterogeneity. Fire, wind, flood, and other property hazards generate different damage mechanisms, access constraints, inspection requirements, contractor dependencies, supplemental estimates, and settlement arrangements. Simultaneously, a property's repair stage and a claim's administrative stage may diverge: an insurer can make substantial advance payments before reconstruction begins, a contractor can complete repairs before a final invoice is accepted, and an insured can reopen a claim after an initial closure. This monograph examines the interaction of peril, reporting and inspection delay, repair-stage progression, cost revision, cash-payment development, and insurance liability valuation. It derives accounting identities linking ultimate covered loss, cumulative paid loss, case outstanding amounts, and remaining development; formulates observable stage-specific transition and payment representations without presenting them as a new proprietary theory; and constructs transparent hypothetical examples. The source base comprises actuarial practice standards, insurance accounting and prudential rules, original flood claims procedures, regulator-observed wildfire settlements, historical and stochastic reserving research, inflation measurements, and market-conduct rules. A comparative reading of the United States, United Kingdom, European Union, and Australia establishes which obligations are jurisdiction-specific. The empirical evidence shows why advance-payment frequency cannot be interpreted as completed reconstruction or final reserve adequacy. The scientific conclusion is that credible estimation requires consistent definitions, peril-sensitive stratification where credible, time-stamped repair evidence, explicit reporting of missing data and uncertainty, and reconciliation between operational milestones and financial liabilities. Original contributions are documentary synthesis, formal clarification, and worked analytical interpretation. No new proprietary model, doctrine, or empirically validated predictive advantage is asserted.

Keywords: Property insurance; claims development; loss reserving; incurred but not enough reported; IBNR; IBNER; paid loss; incurred loss; claims adjustment; peril classification; fire; wildfire; windstorm; flood; repair stages; reconstruction inflation; supplementary claims; catastrophe claims; reserve validation; survival analysis; actuarial uncertainty.

Introduction

An insured property can be damaged in a moment and restored through a sequence of inspections, estimates, mitigation measures, contractor bids, permits, partial payments, supplementary discoveries, and final settlements that extends across months or years. The accident date, first notification, insurer acknowledgment, first inspection, scope agreement, commencement of repairs, physical completion, disbursement of funds, and ultimate claim closure therefore refer to distinguishable events. The settlement trajectory involves interactions among the hazard, the property, the insurance contract, the policyholder, the adjuster, the repair market, and occasionally a mortgage lender, court, reinsurer, or public authority.

Actuarial claims development aggregates these paths for estimation. A paid-loss triangle measures cumulative disbursement by origin cohort and development age. An incurred triangle combines paid amounts with recorded estimates of unpaid obligations, according to an identified data convention. Reserve decisions require projections beyond observed development, and the significance of those projections depends on payment timing, case estimate revisions, unreported claims, claim-adjustment expenses, and the legal measure of the insurer's liability. The property/casualty unpaid-claims standard explicitly calls for consideration of the claims adjustment process, development characteristics, recoverables, and estimation uncertainty (Actuarial Standards Board, 2011).

The question pursued here is how these familiar quantities change when the analyst conditions on peril and repair stage. A flood claim can require drying and inspection of concealed moisture before a reliable reconstruction estimate exists. A wildfire claim may involve total rebuilding, debris removal, temporary living costs, and rebuilding permits. Hail or wind claims may be resolved through component-specific repairs, but contractor availability and concealed damage can still revise scope. Every such statement is a mechanism to test against claim-level evidence rather than an assumed universal time-to-settlement ranking.

The research proceeds in six substantive chapters. The first fixes definitions and sources. The second examines how reporting and inspection determine the information available for reserving. The third analyses repair milestones, cost change, and multiple payment calendars. The fourth treats reserve estimation and data architecture. The fifth establishes empirical and statistical tests, including a precisely delimited regulatory case study. The sixth integrates these results across jurisdictions and identifies the conditions under which additional stratification is informative.

The primary scientific object is settlement runout and cost development for insured property damage. Related objects—policy coverage litigation, catastrophe vulnerability modeling, reinsurance pricing, consumer law, and full property replacement economics—are engaged only where they explain that development. Claims under heterogeneous insurance forms must be kept distinct. Government catastrophe-loss estimates cannot be substituted for insurer claim payments, and a building reconstruction bill cannot be treated automatically as covered loss.

Definitions, Historical Context, and Primary Source Foundations

The unit of observation and the sequence of insured loss

A peril is a cause or class of causes of physical loss, such as fire, flood, or wind. Its operational definition must match the insurer's coverage coding and the underlying policy wording; a weather-event label is not itself a legal coverage determination. A catastrophe event can comprise many claims, coverages, and insured locations. A single claim can also contain several damage components—building, contents, additional living expenses, business interruption, and covered professional or mitigation costs—with different estimates and payment schedules.

The observable sequence can be expressed as hazard occurrence → physical damage → discovery and documentation → coverage and valuation investigation → adjustment → compensation. Physical reconstruction interacts with this sequence but is not identical to it. The insurer may pay policy proceeds before the work is performed, as illustrated by California's wildfire advance-payment requirements (California Department of Insurance, 2025). Conversely, a claimant can complete emergency mitigation before submitting invoices. Hence “repair stage” and “claim stage” should be recorded separately and time-stamped, rather than combined in an ambiguous field labeled status.

Five amounts require distinct accounting treatment:

  1. Physical damage cost is the expense of restoring or replacing damaged property under an engineering and price basis.
  2. Economic loss includes consequential losses, delays, and costs borne by the affected owner or community.
  3. Covered loss results from applying the relevant policy's insured interests, perils, conditions, exclusions, deductibles, limits, valuation, and other clauses.
  4. Insurer liability is the amount the insurer is obliged or expected to settle under the applicable accounting or prudential measurement basis.
  5. Paid loss is the cash transferred in discharge of identified insurance obligations as of the valuation date.

An estimate of physical loss may exceed a policy limit; a case reserve may incorporate only amounts considered payable; a supplementary estimate can be contested or later withdrawn. Thus a larger repair invoice does not mechanically increase insurer liability dollar for dollar.

Paid, case incurred, ultimate, and unpaid development

For claim i at observation date t, write P_i(t) for cumulative paid covered loss, C_i(t) for the recorded unpaid case estimate on the same loss-only basis, and I_i(t) = P_i(t) + C_i(t) for recorded incurred loss. Let U_i denote the eventual ultimate covered claim payment or an actuarial estimate thereof. On compatible, undiscounted and gross bases, the estimated unpaid claim amount is R_i(t) = E[U_i | information at t] − P_i(t). The amount beyond the current case outstanding estimate is D_i(t) = R_i(t) − C_i(t), equivalently E[U_i | information at t] − I_i(t). The sign of D_i can be positive or negative.

For a group of claims, these quantities are summed without changing the definitions. When the group includes future reporting of already incurred events, the additional amount includes true incurred-but-not-reported claims. For already reported claims, subsequent increases or decreases are often described as incurred-but-not-enough-reported or incurred-but-not-enough-reserved development, although terminology varies by insurer. Claim-adjustment expenses require their own consistent inclusion rule. Reinsurance, salvage, and subrogation recoveries must be reconciled separately to avoid mixing gross and net observations. These distinctions follow the scope and definitions of ASOP No. 43 (Actuarial Standards Board, 2011).

Historical actuarial literature developed methods for such projections well before repair-event timestamps were commonly available in unified digital claim files. Bornhuetter and Ferguson's original treatment of IBNR (1972) explains the use of prior expected ultimate losses alongside observed development. Mack's chain-ladder uncertainty derivation (1993) provides a distribution-free standard-error calculation subject to chain-ladder assumptions. England and Verrall's survey of stochastic reserving (2002) examines models reproducing traditional estimates and extensions for uncertainty, smoothing, tails, and predictive distributions. These sources support the existing statistical toolkit. They do not demonstrate that a particular repair-stage segmentation improves reserve accuracy in every portfolio.

Reporting and repair time as different clocks

At least four clocks must be identified:

  • Accident time begins at the physical insured event.
  • Administrative development time advances through first notice, inspection, scope agreement, interim decisions, closure, and reopening.
  • Repair time advances through stabilization, drying or demolition, engineering scope, permitting, procurement, physical work, and inspection of completion.
  • Payment time is measured by cash disbursements, including advances, progress payments, additional settlements, and recoveries.

The clocks interact through operational constraints and contractual terms. They cannot be aligned by treating a paid claim as a repaired building. Distinguishing clocks also resolves a common source of misleading duration statistics: time from notice to first payment can be short while time from the event to physical occupancy is long. In California's 2025 wildfire response, advance benefits for contents and living expenses were expressly directed at prompt relief, rather than proof that complete rebuilding had occurred.

The origin cohort must also be defined. Accident year, catastrophe-event cohort, report month, policy year, inspection month, and repair-start cohort describe different selections. Using report month as the origin removes some reporting latency from the measured runout; using completion month selects claims that have survived to repair completion. Claims observed before entry into the sample are left-truncated, while incomplete repair paths are right-censored. These selection mechanisms affect any reported development factors.

The legal and prudential frame

The United Kingdom's ICOBS 8 claims-handling rules require insurers within scope to handle claims promptly and fairly, provide reasonable guidance, avoid unreasonable rejection, and settle promptly after agreement (Financial Conduct Authority, 2026 consolidated view). The U.S. National Association of Insurance Commissioners lists an Unfair Claims Settlement Practices Act and a separate property/casualty model regulation (NAIC, undated). Model laws guide possible state enactments but do not become binding law in every state simply by appearing in the NAIC catalog. State-specific timing, notice, and loss-settlement provisions must therefore be checked when analyzing an actual insurer.

Prudential reserve measures also vary. The Australian Prudential Regulation Authority's GPS 340 liability valuation standard requires outstanding claims estimates, including handling expenses, and separates central estimates and risk margins (APRA, 2023). Under EU Solvency II, Article 77 on technical provisions provides for probability-weighted discounted cash flows and a risk margin, with reinsurance recoverables calculated separately (European Union, 2009, as consolidated). The IFRS 17 measurement framework uses current fulfillment cash flows and a risk adjustment for nonfinancial risk within its reporting architecture (IFRS Foundation, 2023). No single accounting term “reserve” can be moved between these frameworks without reconciling discounting, expenses, risk adjustment, contract boundaries, and presentation.

This chapter establishes the central object: development refers to changing information and cash obligations associated with the same insured losses, while repair progression supplies an additional, imperfectly observed explanatory dimension.

Reporting and Inspection Delays

What is delayed, and why peril matters

Reporting delay is the interval between insured occurrence and first notice to the insurer, defined against precise timestamps and reporting rules. Inspection delay may be measured from notification to first visit, from access becoming safe to inspection, or from notification to a sufficiently complete technical examination. These definitions yield different statistics. The delay from first inspection to an agreed scope should be separately measured because repeated inspections can materially change severity.

Peril matters through more than the physical form of damage. Floodwater can restrict safe access and require moisture assessment after water recedes. Wind can damage roofs, enclosures, and internal finishes while making sites temporarily unsafe. Wildfire may prevent access through evacuation, debris, hazardous substances, and utility failure. Fire and water damage can also coexist in one incident. A single “peril” code may consequently omit relevant interacting hazards or claim components.

The June 2025 NFIP Claims Manual provides an unusually detailed primary account of the flood adjustment journey, including notice, first contact, inspection, documentation, damage estimate, payment, and supplemental activity (FEMA, 2025). Its discussion of arranging inspections and informing insureds how to preserve documentation when inspection is delayed demonstrates an institutional recognition that access and evidence preservation affect claim development. This manual governs a specified U.S. federal flood-insurance program; its procedural time frames should not be generalized to private fire or wind policies or foreign insurance systems.

The reporting-to-inspection information process

A claim can pass through notification, first inspection, specialist investigation, and estimate revision. Each step changes the information set available to the case adjuster and reserving actuary. A time-stamped observation may be represented by X_i(t), including hazard characteristics, property construction, photographs, estimated damage extent, coverage indicators, accessibility, and outstanding technical questions.

The expected ultimate loss conditional on the information set at valuation date t is E[U_i | X_i(t), prior history]. As information accumulates, this conditional expectation can increase or decrease. An increase after opening a wall may reflect newly discovered hidden damage, not inflation. A decrease after engineering examination may reflect a corrected quantity or a coverage decision. A large upward incurred development factor is therefore not, by itself, proof that initial claims personnel systematically under-reserved.

The NFIP claims checklist advises the insured to supply photographs, review estimates, and seek additional payment for discovered or misidentified damage (National Flood Insurance Program, undated). This source demonstrates the existence of supplementary claim documentation in a defined program. It supplies neither a portfolio-wide probability of supplemental payment nor a universal claim-reopening rate.

To test the information mechanism, the dataset must separate the date of initial inspection from the date an estimate was authorized; the number and type of specialist visits from the number of adjuster contacts; and physical-damage discovery from administrative processing. Hazard, building condition, geography, access restrictions, and selected contractor can influence both the delay and the ultimate cost, creating confounding. The analysis should not attribute all extra cost observed after a slow inspection to the inspection delay itself.

Censoring, reporting cohorts, and selection

A reporting-lag distribution estimated only from claims already reported at a valuation date omits claims that are still unseen. The tail is therefore dependent on reporting completeness. For a claim observed on valuation date v, let T_N = first-notice date minus occurrence date and T_I = first-inspection date minus notice date. A complete historical dataset records these intervals and the exposure period defining their sampling frame. Unreported claims are latent; reported but uninspected claims contribute censored inspection intervals.

Repair-stage datasets create a related selection bias. A reported median time from repair start to completion, calculated only for completed projects, preferentially includes faster projects. Projects still awaiting permits, contractors, payments, or completion are right-censored. Duration analysis must retain these open paths and define whether a claim that switches to a cash settlement counts as completed restoration, a competing outcome, or a separate endpoint.

Cumulative-incidence methods can be used when there are multiple endpoints, such as repair completion, replacement, cash settlement without repair, abandonment, or litigation. Standard survival methods can support inspection and repair-time estimation, provided their assumptions on censoring and competing events are examined. They are methodological applications, not claims of a new hazard model.

Delays in catastrophe accumulations

Catastrophes generate correlated reporting and handling pressures. Many properties become inaccessible at the same time; adjusters, engineers, remediation crews, roofers, and permitted contractors face demand surges. Telephone interruptions, evacuation orders, limited labor supply, and debris removal affect reporting and inspection. A single event creates within-event dependence across insureds, so treating all claims as independent observations understates uncertainty.

An event-clustered design should identify catastrophe event, hazard intensity, location, property vulnerability, coverage, contractor network, and date of notification. The National Weather Service storm records demonstrate that external event reports describe weather phenomena and sometimes estimated physical damage, with variable observation and reporting practices (National Weather Service, undated). They provide useful external event covariates but are not an insurer's paid-loss ledger. Linking these records to claims requires geographic and temporal alignment and explicit uncertainty in event matching.

Reporting curves under catastrophe conditions should be compared with matched noncatastrophe cohorts only after adjustment for property type, event severity, location, contract mix, and calendar period. The observed difference is descriptive unless a defensible identification design isolates a causal change.

Research interpretation

A short notice-to-inspection interval is an operational indicator, but it is not sufficient to establish accurate reserves. A fast preliminary estimate can be followed by extensive supplemental scope and late payments. Conversely, an initially longer inspection interval can produce a more comprehensive estimate if safe access and full technical assessment become possible. The relevant scientific question is whether a given delay changes the amount and timing of information entering the loss estimate, conditional on physical damage and coverage. This question can be evaluated with longitudinal claim files and explicitly stated comparison designs.

Repair Milestones and Cost Updates

The physical restoration sequence

Repair stages should be recorded as observable milestones with dates, documentary evidence, and explicit transitions. For example, a damaged building may move through emergency stabilization; drying, testing or demolition; detailed scope and contractor bids; permit or code review; materials procurement; structural and systems reconstruction; final inspection; and occupancy. These stages can overlap or recur. A property may contain several repair packages in different stages at once, and an uninhabitable building can generate separately developing living-expense payments.

The building's physical completion and the insurer's administrative claim closure are therefore different variables. A claim may be closed after a negotiated cash settlement even if the insured elects not to rebuild, subject to the contract and law. A contractor may finish work while an agreed sum remains unpaid due to document review, mortgage escrow, or a dispute. The October 2026 California agency investigation of insurer-proceeds disbursement addresses delays associated with financial institutions holding post-loss proceeds, illustrating the importance of distinguishing insurer disbursement, lender release, and contractor receipt (California Business and Consumer Services Agency, 2026).

A repair-stage variable has analytical value only if its meaning is stable. “Repair initiated” might denote a signed contractor contract, first invoice, first demolition, or first physical reconstruction. The dataset should specify which event it measures. An adjuster-entered textual status and a timestamp supported by a completion certificate provide different levels of evidence.

Scope changes and inflation

The estimated ultimate cost evolves through identifiable mechanisms: corrected quantities, newly discovered hidden damage, building code or safety requirements, approved changes in repair methodology, contractor repricing, market supply constraints, and coverage determinations. These mechanisms should appear as separate reason codes rather than being pooled into a generic adjustment.

The California wildfire claims guidance advises insureds to obtain contractor estimates and explicitly describes post-disaster demand surge as an increase in rebuilding costs when labor and materials are scarce (California Department of Insurance, undated). The BLS Producer Price Index program provides official series measuring changes in prices received by domestic producers (U.S. Bureau of Labor Statistics, 2026). A broad national PPI is not a local, claims-specific reconstruction-price index. Appropriate cost trending requires attention to material mix, labor, geography, trade, index revision, contractor margins, procurement timing, and scope composition.

For repair component k, a useful decomposition is revised projected cost = revised quantity × revised unit price + separately documented permissible soft costs. The change from an earlier estimate can be attributed to quantity or scope, unit price, approved compliance expense, and contract coverage adjustment. Interaction terms may arise when quantities and prices change simultaneously; assigning the entire interaction to inflation would misstate the drivers.

Inflation at a common valuation date should be distinguished from additional inflation expected between that date and actual payment. An amount already spent is historical; an unpaid future repair may require forecast price escalation. Under GPS 340's claims escalation provisions, Australian insurers must address relevant future escalation in central liability estimates (APRA, 2023). The resulting actuarial estimate remains sensitive to project timing and price assumptions.

A transparent hypothetical scope-revision illustration

Consider a single hypothetical insured property with an initial documented reconstruction estimate of $100,000. Subsequent inspection identifies $12,000 of concealed covered damage, a building-code upgrade adds $10,000 that is assumed covered under the illustration's specific endorsement, material and labor repricing adds $8,000, and a reconciled scope review removes a $5,000 duplicate item. The final estimated covered reconstruction amount is then $125,000, before any separate deductible, applicable limit, adjustment expense, or other settlement clause.

Revision Amount Revised estimated covered cost
Initial physical scope $100,000 $100,000
Concealed damage discovered +$12,000 $112,000
Assumed covered code work +$10,000 $122,000
Procurement and labor repricing +$8,000 $130,000
Duplicate estimate removed −$5,000 $125,000

These values are invented for an accounting demonstration and do not represent insurer experience. The total upward development of $25,000 is driven by four offsetting changes, not by a homogeneous 25% price inflation rate. Moreover, if a hypothetical applicable policy limit were $110,000 for that component, the amount payable could be capped subject to all relevant policy terms. The case-incurred development must therefore be measured after the relevant coverage transformation, not inferred directly from the contractor estimate.

A supplementary claim may arrive after earlier payments. Its timing can reflect scope discovery or delayed invoicing. Distinguishing these mechanisms matters because an actuarial model based solely on calendar-month payment factors can mistake a change in insurer payment procedures for a change in physical-loss severity.

Peril comparison

Fire and wildfire commonly create combinations of structural rebuilding, smoke remediation, contents replacement, debris removal, utility restoration, and temporary accommodation, with duration dependent on destruction, access, permits, and coverage. Flood claims may require contamination removal, drying, material replacement, and careful separation of covered direct physical loss from uncovered consequences. Wind and hail claims may involve roofing, siding, openings, interior water intrusion, and localized or widespread contractor demand. Accidental water damage may require hidden-moisture investigation and staged mitigation.

These are plausible operational pathways, not immutable actuarial classifications. A severe windstorm can destroy an entire structure; a contained fire can be repaired rapidly. A peril category alone does not identify severity, construction type, regional contractor supply, or the share of claims at each stage. Separate stage effects from case mix with severity proxies and credible exposure information.

Government and contractual rules can reshape the path. The FEMA flood policy wording documents proof-of-loss requirements in the Standard Flood Insurance Policy (FEMA, 2025). The NFIP Claims Manual addresses supplementary proofs and adjustment procedures. A procedural requirement or event-specific extension changes claim administration and can shift the observed timing of payments independently of the time required to restore a building. Such program-specific rules are not universal private-property-insurance conditions.

Payment schedules and reverse development

Advance payments, progress payments, recoveries, supplemental payments, and reversed or corrected transactions make cumulative paid amounts operationally complex. Gross cumulative paid loss ordinarily increases with completed disbursements, but an administrative dataset can record negative corrections or recoveries; the treatment must be specified. Incurred loss can move downward when an estimate is reduced, a duplicate is removed, a coverage assessment changes, or the insurer receives reliable evidence of a lower ultimate obligation.

The cash-flow analyst should preserve each transaction date and payment type, distinguish a payment authorized from one cleared by a bank, and reconcile payment ledgers against financial statements. Payment to a mortgagee or escrow agent can be an insurer disbursement even though the household or contractor has not received the proceeds. The 2026 California lender-disbursement investigation is an institutional example of why this distinction is observable and material.

Research interpretation

Repair-stage information has its greatest analytical value where it explains the residual cost or payment timing beyond what the development age and peril already predict. A stage label should be judged on predictive value, data quality, coverage coherence, and stability across claim administrators. It may otherwise introduce noise, leakage, or selection bias. A financially open claim may be physically complete; a physically incomplete property may have received an advance settlement. These states must remain visible to reserve estimation and claims-conduct assessment.

Reserve Calibration and Data

Reserve measures and accounting reconciliation

Reserve calibration begins with a clearly specified target: undiscounted gross ultimate payment, net ultimate loss, expected future claims cash flows, liability for incurred claims, or a prudential technical provision. They cannot be substituted for one another. For a claims cohort with cumulative paid P, known-case outstanding C, and estimated ultimate U, the undiscounted remaining claims amount is U − P. The case-incurred amount is P + C. Any estimated development beyond currently recorded case incurred equals U − (P + C), assuming comparable definitions and a common valuation date.

Consider three entirely hypothetical peril groups, each containing 100 already reported claims. All dollar amounts below are in millions and exclude claim-adjustment expenses, recoveries, discounting, unreported claims, and changes in coverage basis.

Reported peril group Cumulative paid Case outstanding Estimated ultimate Estimated unpaid Development beyond case incurred
Fire 3.2 4.3 8.9 5.7 1.4
Wind 4.8 2.1 7.3 2.5 0.4
Water damage 2.0 3.0 6.1 4.1 1.1
Total 10.0 9.4 22.3 12.3 2.9

The total verifies the identities: 22.3 − 10.0 = 12.3; 10.0 + 9.4 = 19.4; 22.3 − 19.4 = 2.9. Since the illustration expressly assumes all claims are known, the final column represents further estimated development on reported claims, not true unreported-claim emergence. This illustration does not establish an empirical ranking of perils or a prudent booking level. If the case estimate includes expenses while ultimate excludes them, none of the identities can be applied until definitions are reconciled.

Chain-ladder and related established approaches

A paid development triangle groups cumulative payments by origin cohort a and development age d, often written P_(a,d). Under a conventional aggregate chain-ladder construction, an age-to-age factor f_d can be estimated as a weighted ratio of total observed P_(a,d+1) to total observed P_(a,d), using cohorts for which both ages are available. Projected cumulative ultimate amounts multiply the latest cumulative observation by appropriate future factors, including any tail factor. The core assumption is that the selected development pattern is sufficiently stable and transferable to immature cohorts.

The Mack (1993) analysis studies uncertainty under distribution-free chain-ladder assumptions. Its standard errors do not automatically encompass all future catastrophe regimes, structural repair-market changes, policy-form revisions, or errors in peril coding. The England and Verrall (2002) review situates chain-ladder-like estimates within stochastic reserving and emphasizes predictive distributions. The Bornhuetter–Ferguson paper provides an established reason to combine exposure-based prior expectations with observed development, particularly when immature experience is unstable (Bornhuetter and Ferguson, 1972).

Peril-specific triangles are informative when each cell has credible volume and historical comparability. A sparse wind-hail subtype may require grouping or hierarchical credibility rather than independent, volatile factors. Catastrophe years should be identified so a portfolio's development factor does not silently average structurally different event and non-event conditions. Changes in claims handling, coverage forms, payment authorization, and inflation can invalidate historical comparability even where the actuarial arithmetic remains correct.

Two-dimensional conditioning and the limits of aggregation

The research object requires distinction between development age and repair stage. Let p denote peril class, s(t) the observed repair state at time t, z property and contractual covariates, and a the origin cohort. An analyst may estimate future payment increments conditional on (a, t, p, s(t), z) using an established regression, survival, multi-state, reserving, or simulation framework. The task is to estimate a conditional distribution of future payments while preserving the total ultimate and accounting identities.

Such conditioning presents three difficulties. First, repair status is observed only after a claim enters the system, so it cannot by itself describe the complete population of future reporting claims. Second, repair stage can be an outcome of insurer payment timing, making it endogenous to the variable predicted. Third, sufficiently detailed combinations of peril, geographic region, event, severity, coverage, and stage may leave too few observations to estimate stable probabilities.

The remedy is methodological discipline: document the target quantity, use past information available at the valuation date, consider partial pooling or credible grouping, and evaluate whether the additional field improves independent out-of-time predictions. There is no presumption that a more detailed classification produces a better reserve.

Cash-flow valuation and risk margins

Under the IFRS 17 contract measurement principles, estimated future payments may enter fulfillment cash flows alongside their time value and risk adjustment in the applicable measurement context (IFRS Foundation, 2023). Under Solvency II Article 77, technical provisions comprise a best estimate and risk margin; the best estimate is the probability-weighted present value of future cash flows (European Union, 2009, as consolidated). APRA GPS 340 specifies central estimates, risk margins, and separate gross and net valuation in the Australian prudential context (APRA, 2023).

For valuation date v, future covered payment amounts X_j due at times t_j may be represented for an expected-present-value calculation as sum_j E[X_j | information at v] × DF(v,t_j), where DF denotes a consistently chosen discount factor. The precise regulatory framework determines what is included, the discount curve, treatment of expenses, recoveries, risk adjustment or margin, and recognition. A statistical “unpaid reserve” in nominal claim dollars is therefore not automatically an IFRS 17 liability or a Solvency II technical provision.

A separate issue arises once settlement is agreed. The European supervisor's Q&A on agreed but unpaid claims explains that under the referenced Solvency II treatment, classification between technical provisions and insurance/intermediary payables depends on due date, rather than simply whether the settlement was agreed (EIOPA, 2023). This is a precise example of how administrative settlement stage, due date, and prudential balance-sheet category must be distinguished.

The EIOPA valuation guidelines applied from January 2023 provide further best-estimate guidance; EIOPA's July 2026 revision report identifies subsequent changes applicable from January 30, 2027 (EIOPA, 2022; 2026). An as-of-November-2026 assessment must keep the two applicability dates distinct.

Data architecture and necessary controls

A reliable study requires a linked but auditable set of data tables. The exposure table identifies policy and property, sums insured, coverage form, deductibles, limits, occupancy, construction, and valid policy dates. A peril-event table records event identifier, occurrence date, location, externally sourced intensity measures, and attribution uncertainty. The claim table contains claim and component identifiers, first notice, reported peril, coverage decision dates, current administrative status, and closure or reopening dates. An inspection table records inspection type, date, findings, and relevant scope revisions. A repair-stage table preserves the dates and evidence establishing each milestone. A transaction table preserves payment date, payee, component, gross/net basis, and reversals. A case-estimate history table records estimate revisions as immutable observations rather than overwriting the last number.

The actuarial data-quality standard requires professional attention to the suitability, review, use, limitations, and disclosure of actuarial data (Actuarial Standards Board, 2016). The actuarial modeling standard addresses model selection, intended purpose, validation, and communication (Actuarial Standards Board, 2019). Applying these practices requires controls for duplicated event IDs, missing coverage codes, payment reversals, reopened claims, timezone inconsistencies, claim-component mergers, imputed milestones, and history overwritten by current statuses.

A claim may have multiple coverage components at different repair and settlement stages. The natural relational unit for payment and milestone analysis may be the claim-component observation, whereas an origin-cohort paid triangle may aggregate at claim or event level. The bridge between these units must be explicit to prevent multiplying a single claim into several apparent claims when joining to multiple repair records.

Reserve governance and external consistency

The IAIS Insurance Core Principles emphasize reliable and transparent prudential valuation and sound insurance supervision (IAIS, 2024). Governance requires controls on model and process change, source-data provenance, booking adjustments, event-specific assumption review, and communication of uncertainty. A reserve number should be accompanied by its valuation basis, currency, gross/net status, inclusion of expenses, claims origin, peril taxonomy, inflation assumption, discounting convention, tail assumption, and validation period.

Where a regulator uses policyholder-protection, claims-conduct, or financial-reporting data, the dataset's purpose must be respected. A table of regulatory complaints is not a probability sample of all claims. Aggregate catastrophe disbursement reports may omit coverage-level development necessary for actuarial parameter fitting. These limitations do not make such sources uninformative; they determine which questions those sources can answer.

Empirical Verification, Uncertainty, and Competing Explanations

A real regulatory observation and its denominator

The California Department of Insurance published a Los Angeles wildfire claims tracker, last visibly updated July 6, 2026, showing 41,727 claims filed, 40,482 claims partially paid, and approximately $25.2 billion in claims payments to date for the Eaton and Palisades fires (California Department of Insurance, 2026). The regulator states that the reporting covers multiple loss classes—home, business, living expenses, automobiles, and other catastrophe-related claims—and includes advances, rebuilding or replacement payments, and debris-removal-related payments. It further warns that disbursements will grow as rebuilding progresses.

This is an instructive longitudinal snapshot precisely because it resists a simplistic conclusion. “Claims partially paid” describes a payment milestone, not physical reconstruction completion. “Claims paid to date” is an aggregate cash amount, not the ultimate insured cost or insurer booked reserve. The number of claim records is not identical to the number of destroyed structures, policies, households, or repair projects. The report's mixture of coverage components and insurer types prevents using it directly as a uniform home-repair completion curve.

A crude ratio of partially paid to filed claims based on the displayed figures is 40,482 / 41,727 ≈ 0.970. It is a descriptive ratio of displayed categories as of the report date, not an estimate of the share of claims finally settled, fully indemnified, or repaired. To interpret it actuarially, one would require the tracker metadata, claim-level definitions, cohort composition, handling of multiple coverages, and historical snapshots. The documented statutory advances after the fires provide a concrete reason why first payment may be early relative to reconstruction (California Department of Insurance, 2025).

California's living-expense guidance describes conditional coverage and rebuilding-delay extensions in the relevant state-law context (California Department of Insurance, 2025). It illustrates that occupancy-related loss components can continue after initial building payouts. The 2026 mortgage proceeds review independently shows that disbursement from an insurer may precede funds becoming available to the insured for construction (California Business and Consumer Services Agency, 2026). Together these primary records demonstrate separate clocks without proving a single causal magnitude for how those clocks influence reserves.

Empirical designs that can test stage effects

A robust observational design begins by fixing the valuation date and reconstructing exactly what was known then. Historical future developments are used only for validation, never as contemporaneous features. The dataset is divided by event and calendar period so that related catastrophe claims do not leak into both training and evaluation subsets. Where possible, an entire event is held out; an independent future calendar window tests performance under later repair costs and institutional conditions.

The dependent target can be remaining nominal gross payment from valuation date to eventual settlement, an appropriately discounted expected cash flow, or future change in recorded incurred loss. The target must be identified before selecting features. Candidate predictors include peril, event, property class, coverage form, incurred/paid amount at valuation, claim age, inspection lag, repair stage, specialist inspections, known permit status, local contractor market, and seasonality. Some features may represent decisions the insurer can influence and must be interpreted cautiously in causal analyses.

Three useful comparisons are: an age-only triangle benchmark; a peril-conditioned benchmark; and a benchmark supplemented by repair-stage information observed at the valuation date. All are established analytical types. The incremental question is whether repair-stage information improves calibrated reserve predictions after its capture cost and risk of missingness are taken into account.

Performance should be compared at claim and portfolio levels. For a point estimate, use development bias, mean absolute error, and exposure-normalized error, with care where a cohort has near-zero ultimate claims. For a predictive distribution, examine interval coverage, proper scoring rules, and tail underprediction. Assess performance by peril, event, severity decile, report-lag group, repair duration, contract type, and jurisdiction. A single pooled error value can hide systematic deficiency in catastrophic fire or large commercial water claims.

Measurement error and alternative explanations

A correlation between slow repair and high ultimate loss has several plausible explanations. Severe damage causes both expensive reconstruction and long repairs. A catastrophe restricts contractors, creating delay and repricing simultaneously. Documentation difficulty raises the probability of scope supplementation and slows inspection. A payment advance can accelerate cash development without changing physical reconstruction. A complex coverage dispute may lengthen handling but result in a lower payable amount. Different adjusters or suppliers may apply different stage codes.

These pathways motivate causal diagrams or other explicit assumptions before claims of delay-induced inflation. If local contractor congestion affects both completion time and prices, conditioning only on observed repair stage can produce biased interpretation. If a reserve analyst uses the eventual repair completion date to predict a prior valuation's unpaid amount, the result involves future-information leakage and cannot be presented as out-of-sample performance.

Inflation measurement has its own ambiguity. The BLS PPI framework measures producer selling prices and provides useful economic context but is not a universal building-cost deflator for individual catastrophe claims (U.S. Bureau of Labor Statistics, 2026). Reported rising average paid severity can arise from more expensive repairs, larger insured properties, different deductible structures, changed claim mix, supplemental coverage activation, or higher costs. A decomposition must distinguish the exposure denominator and policy transformation.

Model uncertainty and correlation

Claims development is subject to process variability, parameter uncertainty, and structural or model uncertainty. The Mack chain-ladder formulation addresses uncertainty for specified chain-ladder assumptions (Mack, 1993); stochastic reserving methods can generate a broader predictive distribution, depending on their specifications (England and Verrall, 2002). Neither automatically accounts for a newly introduced statutory advance-payment rule, a new building-code requirement, or an extraordinary contractor shortage.

An empirical confidence interval derived from claims assumed independent may be too narrow for catastrophe data because event-level factors are shared. Validation should employ clustered sampling or hierarchical event effects, event-level stress tests, and scenarios for price escalation, payment delay, reopening, and claim development beyond available experience. Extrapolation beyond observed mature claims should be disclosed explicitly.

Claims under investigation, appealed, litigated, or subject to supplemental proof have heterogeneous paths. A combined “closed” indicator that is overwritten when a claim reopens cannot recover the true history. Persistent event histories allow comparison of closure probability, reopening intensity, and supplementary payment amount. For adequate longitudinal evaluation, snapshots of the case estimate at each valuation are essential; a single current reserve field cannot reproduce prior actuarial information sets.

Falsification and sensitivity checks

A model using peril and repair stage should be subjected to practical falsification checks. First, test whether stage effects disappear when claim severity, event, and coverage are controlled. Second, repeat the analysis after excluding advance-payment periods or measuring payments by component. Third, evaluate results with separate insurer-disbursement and borrower-access dates when mortgage escrows are material. Fourth, test alternative repair-stage definitions and missing-stage imputations. Fifth, compare out-of-time event results with random within-event splits; performance that deteriorates sharply under event holdout signals limited transferability.

Further checks include negative incurred revisions, late reopened claims, changes in catastrophe reporting cutoff, indexed versus nominal amounts, and alternative tail assumptions. A benchmark with coarser groups can outperform a granular model when data are sparse. Such a finding would be scientifically useful because it identifies the boundary at which additional operational detail loses predictive credibility.

Evidence limits of this monograph

This research uses official documentary evidence and established actuarial literature. It has no access to a proprietary claim-level longitudinal dataset; therefore it reports no fitted development coefficients, validated hazard ratios, or measured incremental predictive accuracy for peril-and-stage conditioning. The numerical tables are expressly hypothetical and reproduce accounting logic only. The California tracker supplies a documented administrative snapshot, not an ultimate-loss development triangle. The research provides reproducible definitions, analytical comparisons, and testable questions; no fictional empirical results are introduced.

Integrative Comparative Scientific Synthesis

What the sources jointly establish

The literature and primary records support a three-part conclusion. First, claims development consists of financial amounts that can be defined and reconciled across successive valuation dates. This is the domain of existing claims-reserving methods and ASOP No. 43 (Actuarial Standards Board, 2011). Second, peril and repair process shape the timing and quality of information available for claim estimation, as concretely described in the FEMA claims journey (FEMA, 2025). Third, administrative payment and reconstruction are independent enough that an aggregate catastrophe payments tracker cannot identify restoration completion or ultimate settlement by itself (California Department of Insurance, 2026).

The synthesis has practical analytical consequences. Repair stage is valuable only insofar as it changes the conditional distribution of future covered claims payments or improves diagnosis of case-reserve revisions. Its informational content depends on capture quality, contractual component, peril, event, and observation date. It is often more reliable to begin with transparent aggregate development and evaluate incremental predictors than to assume an elaborate status classification is superior.

Comparative jurisdictional interpretation

The United States sources contain both program-specific and state-specific examples. The NFIP flood program has specialized procedures and policy forms; California's 2025 wildfire recovery shows state-related advance-payment and living-expense mechanisms; NAIC model provisions require state-law reconciliation. The United Kingdom's ICOBS 8 establishes conduct requirements within its regulatory scope. EU Solvency II technical-provision valuation and Australia's GPS 340 provide separate prudential measurement frameworks. IFRS 17 provides an accounting framework for reporting entities to which it applies. These regimes illuminate the same settlement process from different institutional perspectives and cannot be represented as interchangeable legal obligations.

International comparison should therefore be performed at the level of common observable variables—insured occurrence, notice, inspection, estimates, stage, payment, and ultimate obligation—while retaining local statutory, contractual, accounting, and reporting definitions. This permits cross-market research without falsely standardizing legal terms or denying differences in policyholder protection.

Source-to-observation hierarchy

A claims study should distinguish documentary standards, operational manuals, insurer transactions, construction evidence, regulator aggregates, and external hazard observations. A law or policy specifies an obligation; a claim file shows the adjustment history; a contractor document records scope and cost; a payment ledger records disbursement; an external hazard dataset records event characteristics. Each answers a different question. Combining them requires provenance, consistent identifiers, temporal alignment, and explicit reconciliation.

The IAIS supervisory framework emphasizes consistent, reliable valuation (IAIS, 2024), while ASOP No. 23 addresses actuarial data limitations and disclosure (Actuarial Standards Board, 2016). Together they justify treating source lineage as a substantive part of reserve analysis. An unsupported stage field may be less informative than a coarser but well-audited historical payment record.

Decision implications for reserving research

For a mature, high-volume peril segment with stable claims-handling practices, aggregate paid and incurred development may offer transparent and credible projections. For a catastrophe in which statutory advances change the early paid pattern, models based on historical cash factors require adjustment or separate payment-type treatment. For complex structural claims, documented repair milestones may add information on future cash-payment timing and scope uncertainty, although that incremental value must be validated. For sparse or newly emerging hazards, external expectations, expert judgment, partial pooling, and sensitivity ranges may be more defensible than volatile standalone factors.

A reserve governance program should preserve the financial reconciliation U = P + unpaid on compatible bases; separate claim handling and repair milestones; document how new repair information changes an ultimate estimate; isolate nonclaim inflation and payment-procedure effects; and evaluate both mean bias and tail behavior. It should also state clearly whether its result is nominal gross unpaid loss, net loss after recoveries, or a jurisdiction-specific discounted liability.

Research agenda derived from the evidence

Several empirical questions can be addressed using claim-level longitudinal data: whether inspection time predicts later scope adjustment after conditioning on hazard severity; whether repair-stage data improve out-of-event forecasts of payment runout; whether statutory advances alter paid triangles without comparable changes in ultimate severity; whether mortgage disbursement delays create economically important differences between insurer payment and contractor payment; and whether contractor-capacity shocks generate correlated development revisions after catastrophe events.

These questions propose tests of observed mechanisms. They do not constitute a new theoretical doctrine. Any claim of measurable improvement requires a defined dataset, publication of design choices and limitations, and independent predictive validation.

Conclusions

Property insurance claims development by peril and repair stage is a well-defined research object at the intersection of actuarial reserving, claims administration, engineering restoration, contract valuation, and prudential measurement. Its core is the longitudinal change in information and cash obligations arising from insured property damage.

The analysis establishes that (i) peril categories describe relevant but heterogeneous damage and access mechanisms; (ii) report, inspection, repair, administrative settlement, and payment dates measure different processes; (iii) supplementary discoveries, scope correction, building-code work, and price escalation require distinct attribution; (iv) cumulative paid, case outstanding, estimated ultimate, and remaining development must reconcile on the same valuation basis; (v) granular peril and stage segmentation is justified by demonstrated data credibility and predictive performance; and (vi) legal and financial reporting regimes impose jurisdiction-specific measurement boundaries.

The hypothetical examples verify the accounting identities and reveal why a single upward development percentage cannot identify its underlying cause. The California regulator's published 2026 wildfire claims snapshot demonstrates that a high proportion of claims with partial payments is compatible with continuing reconstruction and growing future payments. FEMA's flood-claims procedure illustrates how peril-specific inspection and supplemental documentation can affect the timing and content of loss estimates. Established reserving literature supplies the statistical foundation, while ASOP, IFRS, Solvency II, APRA, and conduct rules define professional and institutional constraints.

The monograph's original analytical contribution is the documented linkage among distinct time clocks, operational repair evidence, coverage-adjusted cost changes, and actuarial liability measurement, with explicit tests for alternative explanations and uncertainty. It claims no empirical priority for a new model. Further research requires anonymized claim-component histories, verified repair timestamps, contract metadata, event-aware validation, and consistent payment and reserve definitions. The practical scientific standard is transparent evidence from occurrence through eventual cash settlement.

References

  • Actuarial Standards Board. 2011. Actuarial Standard of Practice No. 43: Property/Casualty Unpaid Claim Estimates. Adopted 2007; last revised May 2011. Official standard.

  • Actuarial Standards Board. 2016. Actuarial Standard of Practice No. 23: Data Quality. December 2016 revision. Official standard.

  • Actuarial Standards Board. 2019. Actuarial Standard of Practice No. 56: Modeling. December 2019 revision; effective October 1, 2020. Official standard.

  • Australian Prudential Regulation Authority. 2023. Prudential Standard GPS 340: Insurance Liability Valuation. In force July 1, 2023. Official standard.

  • Bornhuetter, Ronald L., and Ronald E. Ferguson. 1972. “The Actuary and IBNR.” Proceedings of the Casualty Actuarial Society 59: 181–195. CAS bibliographic record and original paper.

  • California Business and Consumer Services Agency. 2026. “California Orders Financial Institutions to Report Data on Insurance Payouts to LA Fires Survivors, Investigates Reports of Delayed and Denied Payments.” October 15, 2026. Official announcement.

  • California Department of Insurance. 2025. “Commissioner Lara Orders Insurance Companies to Provide Advance Payments on Claims to Speed Recovery for Wildfire Survivors.” January 23, 2025. Official press release.

  • California Department of Insurance. 2025. “Insurance Coverage for Additional Living Expenses if the Home Is Not Habitable Due to a Wildfire.” Consumer guidance.

  • California Department of Insurance. 2026. LA County Wildfire Claims Tracker. Figures shown as updated July 6, 2026. Official claims tracker.

  • California Department of Insurance. Undated. “Top Ten Tips for Wildfire Claimants.” Official claims guidance.

  • England, Peter D., and Richard J. Verrall. 2002. “Stochastic Claims Reserving in General Insurance.” British Actuarial Journal 8 (3): 443–518. Institute of Actuaries paper.

  • European Insurance and Occupational Pensions Authority. 2022. Revised Guidelines on Valuation of Technical Provisions. Published July 6, 2022; applicable from January 1, 2023. EIOPA publication.

  • European Insurance and Occupational Pensions Authority. 2023. “Q&A 2582: Classification of Agreed but Unpaid Insurance Claims.” Official supervisory Q&A.

  • European Insurance and Occupational Pensions Authority. 2026. Final Report on Revised Guidelines on Valuation of Technical Provisions — Solvency II Review. July 15, 2026; revised guidelines applicable January 30, 2027. EIOPA final report.

  • European Union. 2009, as consolidated. Directive 2009/138/EC (Solvency II), Article 77, “Calculation of Technical Provisions.” EIOPA rulebook transcription.

  • Federal Emergency Management Agency. 2025. National Flood Insurance Program Claims Manual. June 2025 edition. Primary claims manual.

  • Federal Emergency Management Agency. 2025. National Flood Insurance Program Flood Insurance Manual. June 2025 document; Standard Flood Insurance Policy forms and instructions. Official policy manual.

  • Financial Conduct Authority. 2026 consolidated view. Insurance: Conduct of Business Sourcebook, ICOBS 8: Claims Handling. Official handbook.

  • IFRS Foundation. 2023. IFRS 17 Insurance Contracts. Effective for annual reporting periods beginning January 1, 2023. Official issued-standard overview.

  • International Association of Insurance Supervisors. 2024. Insurance Core Principles and ComFrame. Revisions adopted December 5, 2024. Official ICP online tool.

  • Mack, Thomas. 1993. “Distribution-Free Calculation of the Standard Error of Chain Ladder Reserve Estimates.” ASTIN Bulletin 23 (2): 213–225. CAS publication and original paper.

  • National Association of Insurance Commissioners. Undated. Model Laws, including Unfair Claims Settlement Practices Act (MO-900) and Unfair Property/Casualty Claims Settlement Practices Model Regulation (MO-902). Official catalog.

  • National Flood Insurance Program. Undated. “NFIP Claims Checklist for Policyholders.” Official claims checklist.

  • National Weather Service. Undated. “Storm Report Records.” National Centers for Environmental Information storm-event reporting guide. Official storm-data description.

  • U.S. Bureau of Labor Statistics. 2026. “Producer Price Index Home.” Method and current index information; accessed November 15, 2026. Official PPI portal.