RESEARCH DOSSIER
Liability Claims-Made Reporting Lag and IBNR Estimation
Abstract. Claims-made liability insurance determines the contractual significance of claim timing through the dates on which a claim is first made, notified, and accepted within a reporting window. The actuarial significance of timing is broader: events, demands, notices, administrative registrations, case-reserve revisions, and payments form distinct observable processes. This monograph examines their interactions in estimating incurred but not reported claims (IBNR), incurred but not enough reported development (IBNER), and the total unpaid cost of liability obligations. The research synthesizes primary U.S. statutory-accounting guidance, U.S. actuarial practice standards, New York regulatory interpretation, European solvency rules, IFRS insurance-accounting principles, and original actuarial literature on chain-ladder, Bornhuetter–Ferguson, reporting delays, claim-level counts, and claims-processing backlogs. The analysis establishes a contract-conditioned, time-indexed interpretation of reported versus unreported exposure; derives the observed-data selection problem; examines reporting-year, occurrence-year and policy-year reserving; develops reproducible illustrative calculations; and proposes an evidence-based validation program. An essential finding is that the frequency distribution of a reporting delay has no stable reserve interpretation until the claim event, coverage trigger, reporting recipient, evaluation cutoff, eligibility rule, and accounting objective are identified. Policy changes may affect coverage and visibility simultaneously. The empirical literature supports multiple estimation approaches but does not provide a universally transferable liability-claims-made delay curve. The findings are analytical interpretations of established sources, not a declaration of a new proprietary theory.
Keywords: Claims-made liability insurance; claims-made-and-reported; reporting lag; IBNR; IBNER; RBNS; retroactive date; extended reporting period; claim development; notification hazard; exposure; delayed reporting; reserve uncertainty; actuarial data quality; insurance contract boundary.
Introduction
A liability claim acquires several dates during its life. A harmful act or omission may precede an injured person's awareness, a formal demand, the insured's notice to an intermediary, the insurer's recorded notification, an adjuster's first reserve, and final settlement by years. These dates are neither interchangeable statistical timestamps nor interchangeable legal triggers. The contractual definition of a claim decides which demands matter, and the wording of the policy decides whether and when reporting is required. Actuarial analysis then uses the information that has become observable by a specified valuation date to estimate future payments and losses not yet included in the file. The central question is the joint behavior of coverage eligibility, claim emergence, administrative observation, and unpaid economic severity.
The 2026 NAIC statutory-accounting manual, particularly Statement of Statutory Accounting Principles (SSAP) No. 65, distinguishes occurrence, claims-made, and extended-reporting contracts. It associates the first with loss recognition at occurrence, the second with reporting as defined in the contract, and extended reporting with specific additional treatment. That accounting classification must be understood in its U.S. statutory context. A different measurement frame appears in IFRS 17 Insurance Contracts, which uses current estimates of fulfillment cash flows, and in Solvency II Article 77, which defines a best estimate as the probability-weighted present value of future cash flows. These regimes cannot be substituted for one another by relabeling a triangle.
The actuarial object is an estimate, not an assertion that an unreported allegation has already established liability. The U.S. unpaid-claim estimation standard, ASOP No. 43, explicitly requires the purpose and scope of an unpaid-claim estimate, including coverage, claim adjustment expense, gross-versus-net treatment, and relevant uncertainties. The legal object is the claim defined by contract and applicable law; the statistical object is the process by which relevant information enters the observable population. A reporting-lag distribution estimated without the legal definition can be precise numerically yet wrong for the insurance obligation.
A concrete reason for separating processes comes from Verrall and Wüthrich's study of reporting delays. Their analysis uses two real individual-claim data sets, one property and one casualty, and addresses trends and seasonality in claim arrival and delay. It does not establish one delay curve for all liability policies. Verrall, Nielsen, and Jessen's joint RBNS and IBNR study further distinguishes the delay until reporting from the delay after reporting until payment. A recent study by Lindskog and Wüthrich of claims-processing backlogs shows why the date on which a case enters an administration system may itself be sensitive to operational capacity. These are different estimands and must be recorded separately.
The geographic approach is international and comparative. The principal legal/accounting evidence examined here originates in U.S. and European sources because their instruments are openly identifiable and directly relevant; the findings are stated at the geographic level their source permits. The monograph does not extrapolate a New York coverage interpretation to all U.S. states or a European solvency calculation into a global definition of an insured event. All numerical portfolios below are deliberately synthetic teaching examples; they are not observed claims experience, fitted insurer models, or regulatory reserve opinions. The study uses formal identities, source comparisons, and sensitivity analysis to establish a replicable framework for evaluating real insured portfolios once their underlying policy forms and claim records are available.
Definitions, History, Legal Sources, and Subject Boundaries
The dates that govern the liability claim
For a claim record indexed by i, distinguish six dates: the underlying act or event date E_i, the first third-party demand or claim-made date M_i, the date notice reaches the contractually relevant recipient N_i, the date of administrative registration or handling H_i, the time of the first substantive case estimate C_i, and the dates of settlement and payment S_i and P_i. If the underlying harm develops gradually, E_i is an event-history descriptor rather than a necessarily unique legal trigger. A policy may define a wrongful act, an occurrence, a claim, related claims, or a deemed date differently. When a coverage determination uses M_i, the exposure unit should preserve its definition and supporting document rather than derive the answer from N_i. When the form defines a claim as a demand, an administrative complaint, or a proceeding, the earliest valid M_i may have to be established by legal review.
Useful lag measures include L_EM = M_i - E_i (event-to-first-demand), L_MN = N_i - M_i (demand-to-notice), L_NH = H_i - N_i (notice-to-registration), L_HP = P_i - H_i (processing-to-payment), and L_EP = P_i - E_i (total event-to-payment). They have different conditioning populations and sometimes different censoring mechanisms. If the event date is uncertain, a negative or impossible computed lag is a data issue to investigate, not a value to discard automatically. A multi-year latent illness can make L_EM long while an insured may still notify promptly after the first demand, making L_MN short.
The contractual eligibility indicator for a particular claims-made policy can be written abstractly as
[ I_i = 1{M_i \in W_M,\ E_i \text{ or the covered wrongful act is within the retroactive scope},\ N_i \in W_N \text{ if required},\ \text{other operative terms satisfied}}. ]
Here W_M is the first-made window under the form and W_N is the permitted notice window. The precise fields are derived from policy wording, applicable regulation, and the circumstances of the claim. The expression is a classification identity, not a legal determination of any actual case. Related-claims wording can deem multiple claims to have been made on the date of an earlier related matter; this can change the apparent cohort as well as eligibility. The same event may generate damages, defense obligations, exclusions, multiple insureds and potentially competing policy years. A claims file may be opened without establishing indemnity liability; defense expenditure can arise before liability is adjudicated.
Occurrence, claims-made, and claims-made-and-reported
An occurrence policy generally attaches through a defined covered event in its coverage period, even when a demand is first reported later. A claims-made form usually links attachment to a claim first made during the policy period, subject to the stated retroactive date. A claims-made-and-reported form adds an operative notice condition and deadline, sometimes modified by a post-expiration reporting grace period or an endorsement. The New York Insurance Department's 2007 interpretation quotes New York's Regulation 121 definitions of claims-made, retroactive dates, and prior-acts coverage. It also explains that prior unknown and unreported exposure may be insured in particular circumstances. This is a dated state regulatory opinion, not a source of uniform international case law.
The NAIC SSAP No. 65 distinction is unusually important for reserve construction because it explicitly ties its claims-made accounting category to reported insured events. An actuary examining a stock of historical harmful acts under a current claims-made contract must first identify whether each potential future demand is within the contract's reporting protection. An expected demand that would first be made after the expiration of an unextended claims-made reporting window may be a real societal loss while falling outside that contract's prospective liability. Conversely, an extended reporting endorsement may preserve the opportunity for a previously unreported demand to become covered. Reserving the two situations identically would conflate a latent injury with an insured obligation.
Foundational actuarial development
Bornhuetter and Ferguson's 1972 article The Actuary and IBNR developed a practice-oriented response to immature experience and uncertain reporting. Their work remains a central historical basis for combining prior expected loss with observed emergence. The classic aggregate chain-ladder estimates future development using observed age-to-age ratios, but its uncertainty and sensitivity became a separate formal subject in Mack's 1993 distribution-free chain-ladder analysis. Schnieper's 1991 true IBNR and IBNER distinction addresses the substantive separation of unreported claims from development on known claims. England and Verrall's stochastic reserving synthesis places such approaches within an explicit discussion of predictive error, tail extrapolation, generalized linear models, Bayesian models and simulation.
The intellectual progression is relevant because claims-made reporting-lag analysis often requires more than the classic accident-year paid triangle. If the contractual reporting date defines the attachment cohort, a traditional event-year triangle can mix covered claims, uninsurable unreported demands, and cases transferred between reporting windows. Individual-event information and contract metadata can make cohort definitions consistent. Simultaneously, mature historical claim-count development can be valuable where event dates are reliable and eligibility rules stable. Method selection follows the information-generating process rather than an abstract preference for claim-level or aggregate models.
The scope of recognized obligations
At valuation time t, the universe of economically adverse events is larger than the population of covered claims and the population of reported claims. A conceptual partition of ultimate cost is (a) covered and reported open claims, (b) covered but unreported claims meeting the applicable policy and reporting conditions, (c) future development on claims already recognized, and (d) defense and adjustment expense according to the contract and accounting purpose. Uncovered demands, excluded incidents, and purely speculative future acts do not enter an incurred-claims estimate merely because they are plausible. This does not prejudge other liabilities or premium provisions that an accounting standard may require.
The EIOPA non-life obligations rule, Article 36 of Delegated Regulation (EU) 2015/35, describes claims provisions for claims events already occurred whether claims have been reported or not, separately from premium provisions for future claim events. Applying that language to a claims-made contract entails identifying the coverage and contract boundary used by that regime; it cannot be taken as an instruction to treat every historically occurred negligent act as the insurer's incurred claim. The IFRS 17 framework requires measurement of insurance contract groups and fulfillment cash flows with current information; its accounting measurement and risk adjustment should be presented separately from a simple nominal unpaid-claim count.
These definitional distinctions govern every empirical calculation in the remainder of the monograph.
Reporting Lag Distributions
Population, observation, and right truncation
The first question in a reporting-lag study is what generated the records. A claim file collected at calendar valuation t normally contains claims reported at or before t. Events that occurred but have not yet been reported are absent from the observed file. For an event with date E and delay L, membership in that extract often requires E + L <= t. Thus, among events occurring near the valuation cutoff, long delays are systematically excluded. The empirical distribution of L in the extracted cases estimates a conditional distribution of delays among the claims observed by t; it does not automatically estimate the unconditional reporting-delay distribution of the ultimate insured event cohort. The more recent the event cohort, the stronger the observation filter.
Let N_e denote the ultimate number of relevant and contract-eligible claims originating in an event cohort e, and let F_e(a) = Pr(L <= a | e, eligibility, recorded definitions) be the cohort's reporting-completion probability at development age a. Under a simplified homogeneous-count assumption, the expected reported number is N_e F_e(a), while the expected outstanding unreported count is N_e [1 - F_e(a)]. For estimation from observed R_e(a), the elementary completion-factor estimate Nhat_e = R_e(a)/Fhat_e(a) gives unreported IBNR count R_e(a)[1/Fhat_e(a) - 1]. This identity is conditional on a reliably fitted completion factor and an appropriate population definition. Where Fhat is close to zero, the estimator is unstable; where eligibility changes with time or first-made requirements, N_e itself must be modeled as contract-conditioned rather than inferred from a generic event count.
A more granular representation uses event arrival intensity lambda(e, x) for exposures with characteristics x, a probability q(e, x) that an event ultimately generates a claim within the relevant insurance coverage, and a reporting-delay distribution F(a | e, x, contract). For a fixed cutoff t, the expected unreported number for eligible historical events takes the illustrative form
[ E[N_{\mathrm{unreported}}(t)] = \int_{e\leq t}\int \lambda(e,x)q(e,x){1-F(t-e\mid e,x,contract)},dx,de. ]
This formula assumes that q and F refer to the same eventual contract-eligible population and that eligibility can be treated consistently over the projection horizon. In a claims-made policy that expires before some future first-made dates, q must embody the remaining coverage window and any extended reporting rights; a conventional stationary unconditional delay curve is insufficient. The integral is a conceptual decomposition, not an imported published estimator.
Verrall and Wüthrich (2016) analyze a nonstationary reporting-delay model with observed individual-claim data, showing why calendar trends and seasonality deserve direct treatment. Their analysis contains both casualty and property examples; it supports the modeling mechanism rather than proving a long delay distribution for every liability line. Abdelrahman, Badescu, Craiu, and Lin's 2025 online publication, appearing in the January 2026 issue of ASTIN Bulletin, gives a marked Cox IBNR-count model that jointly models delay and claim intensity, with an application to automobile claims. The model demonstrates an available method for dependences and changing risk attributes, while the auto application cannot be used as an empirical prior for long-latency professional or environmental liability without separate validation.
Delay clocks and coverage cohorts
For occurrence coverage, event or injury year is often a natural organizing axis when historical occurrences correspond to the insured event and the claim exposure continues after policy expiry. For claims-made coverage, report year and first-made year may be the dominant underwriting and reserve axes because the portfolio accepted during a calendar reporting interval can include acts from multiple earlier years. Policy year and first-made year coincide only under explicitly stated renewal dates, reporting clauses and eligibility rules. Claims-made-and-reported coverage further requires the notice receipt date and valid recipient. An annual triangle that combines renewals with changed retroactive dates will not preserve consistent exposure even if its reported amounts are perfectly reconciled.
A useful claims register therefore retains, for each case, at least: (1) insured and claimant identifiers; (2) policy and endorsement identifiers; (3) policy inception, expiry and retroactive dates; (4) event, discovery and first-made dates, each with precision status; (5) date of notice to insurer and channel; (6) late-notice/extended-reporting facts; (7) related-claim group; (8) administrative entry and reopening timestamps; (9) coverage status and decision history; (10) paid and case estimates by indemnity, defense and allocated expense. Aggregating only by one reported date loses information that may determine whether a case belongs to the insured population at all.
Three practical delay curves should be estimated independently if data allow: event-to-first-made; first-made-to-valid-notice; valid-notice-to-registration. A fourth, reported-to-final-payment, measures run-off after claim notification. Their convolutions can describe the temporal path from act to settlement under stated dependence assumptions, but may fail when a claims-made endorsement changes the probability that an event reaches the reported dataset. Conditional reporting delay among accepted claims is not, by itself, evidence about rejection frequencies, claims withdrawn before notification, or legal eligibility of future demands.
Hazard rates, competing events and covariates
A discrete reporting hazard h_k(x) is the conditional probability that a claim is first reported at lag k, given no prior valid report, cohort covariates x, and ultimate eligibility. A discrete completion curve follows from F(a|x) = 1 - product over k <= a of [1-h_k(x)]. Reporting may depend on claim severity, the nature of injury, regulatory reporting duties, changes in attorney representation, insured insolvency, and claims-handling protocols. Severity-driven reporting creates informative selection: the first 20% of reported cases need not represent the same expected cost as the last 20%. A count-completion factor cannot be applied to aggregate cost without testing count–severity dependence.
Competing outcomes include a demand made in time but reported after a contractual deadline, a claim discovered after a retroactive-date restriction, a qualifying demand reported within an extended reporting period, a duplicate demand aggregated with a related earlier claim, and an event whose facts never produce a legally compensable allegation. A model may distinguish these as labeled outcomes rather than pretend that every eventual claim is merely waiting for one reporting timestamp. The temporal distinction is essential for exposure denominators and affects the interpretation of reporting hazard estimates.
In claim-level estimation, right-truncation corrections can be obtained through appropriately specified event-arrival and delay models, likelihood construction or retrospective completion factors. Calendar cohort validation should simulate the original observation cutoff: to assess a forecast made at t0, remove from training any data not yet available as of t0. Resampling full modern records while allowing later information to reveal earlier latent claims would constitute look-ahead leakage. Model design should also distinguish a first report from a post-report administrative reopening.
Synthetic reporting-lag calculation
Consider four independent event-year cohorts with exactly 100 ultimately eligible claims each. This is a wholly hypothetical population with identical coverage protections, fixed reporting definitions, stable case mix and a common reporting pattern. Suppose 60% report in the event year, 25% in the next year, 10% in the second year, and 5% in the third year. At the close of 2025, events of 2022, 2023, 2024 and 2025 have reporting ages 3, 2, 1 and 0, respectively. The completed proportions are 100%, 95%, 85% and 60%; therefore reported counts are 100, 95, 85 and 60. Their unreported counts are 0, 5, 15 and 40. The total of 340 reported out of 400 ultimate eligible claims implies 60 true unreported claims, subject to the hypothetical assumptions.
| Event cohort | Reporting age at 2025 year-end | Reported | Assumed ultimate eligible | Unreported |
|---|---|---|---|---|
| 2022 | 3 | 100 | 100 | 0 |
| 2023 | 2 | 95 | 100 | 5 |
| 2024 | 1 | 85 | 100 | 15 |
| 2025 | 0 | 60 | 100 | 40 |
| Total | — | 340 | 400 | 60 |
This example displays arithmetic rather than estimates from an insurer. If a claims-made policy written for 2025 had no later first-made or extended-reporting protection, a portion of the 2025 underlying incidents scheduled to give rise to first demands in 2026–2028 might be excluded from that policy. In that case the assumed 100 ultimately eligible claims would have to be reduced according to the policy's actual first-made and notice rules. The calculation does not imply that every event has a reserve attached to that claims-made contract. Under occurrence wording, more of those later demands could relate to the historical policy period, subject to all other conditions. Comparisons require the same ultimate-eligible denominator.
Data-generating changes and administrative artifacts
Observed claim emergence can accelerate because claimants learn to notify earlier, an insurer improves first-notice capture, claims are registered from a legacy backlog, reporting laws change, or case mix shifts toward easily discovered injuries. A steeper observed development curve does not necessarily mean that underlying insured liability became more frequent or that the ultimate severity increased. Lindskog and Wüthrich's analysis of hidden backlogs treats the distinction between incoming reporting and downstream processing as a substantive modeling problem. A claims-made reserving study should therefore reconcile legal notice, system creation and claims handling capacity before interpreting triangle development as the reporting process.
The reported delay distribution is the result of both claimant behavior and the contract's selection mechanism. A change in retroactive-date scope can remove older events from the numerator; a grace period can add late claims; claim-intake automation can move H earlier while leaving N unchanged. The research conclusion is a measurement result: reporting lag must be estimated from an explicitly dated, contract-defined, and operationally audited sampling frame.
Incurred-but-Not-Reported Losses
True IBNR, IBNER, RBNS and the reserve bridge
At a valuation date, 'IBNR' is used with different widths. In a narrow claim-count sense, true IBNR concerns eligible claims arising from past covered experience that have not been validly reported. IBNER concerns later revisions to estimates for already reported claims: new medical evidence, legal strategy, social inflation, defense costs, coverage disputes and settlement outcomes. RBNS (reported but not settled) names a reported-claim population, usually containing both case reserves and future development. Some reserving reports use 'IBNR reserve' broadly to include case reserve inadequacy, even when it does not correspond to unreported claim count. Schnieper's 1991 separation of true IBNR and IBNER illustrates why these concepts deserve explicit accounting.
For a portfolio whose total ultimate insured nominal cost U and cumulative nominal paid cost P_t are defined consistently, total nominal unpaid cost at t is U - P_t. If case estimates for reported open claims sum to C_t and closed claims may reopen, a practical identity is
[ \text{Unpaid estimate}(t) = C_t + A_t, \qquad A_t = Uhat_t - P_t - C_t. ]
The balancing term A_t is sometimes labeled bulk IBNR or additional actuarial provision; it may include true unreported claims, adverse or favorable development on reported claims, unallocated expense, and reopening assumptions according to the system's definition. It cannot be interpreted as a count of unknown claims without a further decomposition. A_t may be negative if case estimates exceed the selected ultimate estimate; a negative amount is not, by itself, evidence that reporting delay is negative or that all claim events have emerged. Expense classification and recoverables must remain consistent on both sides. The ASOP No. 43 reserve guidance requires attention to the intended measure, gross-versus-net status and claim-adjustment expenses.
The reserve measure also depends on whether amounts are nominal or discounted, whether risk adjustment is included, and whether unpaid cost includes future defense expense. On a gross basis, estimated recoveries from reinsurance, deductibles, salvage, or subrogation can be considered separately; on a net basis they are governed by their own recoverability and timing assumptions. There is no general identity in which all recovery uncertainty disappears merely because gross ultimate cost is correctly estimated. The distinction matters acutely when a liability event traverses multiple excess layers or when reinsurer notification requirements differ from the primary form.
Development triangles and their conditions
Let C_{i,j} be cumulative nominal paid or incurred cost for cohort i at age j and let f_j be a selected age-to-age development factor. The deterministic chain-ladder ultimate estimate at observed age a_i is Uhat_i = C_{i,a_i} times the product of remaining factors, including any tail factor. The cohort's additional cost estimate equals Uhat_i - C_{i,a_i}. The method is transparent and can be effective with stable development, but reported-claim counts, contract changes, severity shifts, calendar shocks and policy renewals can change its meaning. A paid triangle measures payments and settlement behavior as well as emergence; an incurred triangle reflects case-reserving conventions. A reported-count triangle addresses a different observable process and should not be silently substituted for a paid-cost triangle.
Mack's 1993 standard-error derivation gives an established analytic measure of uncertainty under chain-ladder assumptions; the absence of a specified full claims distribution in that framework does not remove its conditional assumptions on link ratios and variability. England and Verrall's 2002 comparative review explores stochastic extensions, including process and parameter uncertainty. Applying these estimators to a claims-made portfolio requires demonstrating that the origin cohorts have comparable coverage and reporting rights and that structural breaks have been handled.
A report-year triangle may be more homogeneous when claims are attached by first-made year; a policy-year triangle may better link pricing and exposure; an event-year triangle supports lag analysis where E is credible. The right choice is empirically and legally dependent. For a claims-made contract, a report year may contain a mix of old incidents and recent incidents, each with different remaining payment tails. A single report-year paid development factor can obscure compositional shifts toward latent long-tail cases. Stratification by coverage class, years since act, claim severity or litigation stage, rather than simply by calendar year, may be necessary.
Prior-expected methods and sparse portfolios
The Bornhuetter–Ferguson approach balances the observed immature position with an independently justified prior expected ultimate. At a given age with expected unreported proportion 1-F, the simplified ultimate-loss form is reported (or otherwise emerged) loss to date plus prior expected ultimate cost times the expected outstanding proportion. In an event-count form, a prior ultimate eligible count K may be multiplied by 1-F(a) to estimate the number not yet reported. Such prior assumptions can come from credible older experience, exposure-based expected losses, or external benchmarks adjusted for line and contract differences. Historical Bornhuetter–Ferguson IBNR practice is particularly relevant to immature or sparse segments.
For professional liability, a prior based on occurrence-based exposures may be inconsistent with a short claims-made reporting window. A mature occurrence-based industry triangle is not automatically an appropriate expected ultimate for a claims-made-and-reported renewal whose retroactive date has been advanced or whose terms prohibit notice after expiration. The prior should refer to the same policy eligibility set as the observed losses. Conversely, a rapidly growing claims-made portfolio can generate stable report-year counts for some coverages while its historical act-year distribution shifts; reserving by report year alone then requires explicit severity-mix adjustment.
Combining claim counts and severity
A claim-count-first approach projects the eligible unreported counts by cohort and covariates, then multiplies each projected count by its conditional future cost distribution. In notation, under a definition of coverage consistent with the insurer's obligation,
[ E[\text{true IBNR cost}\mid\mathcal F_t] = \sum_{g} E[N^{\mathrm{unreported}}_g\mid\mathcal F_t] E[X_g\mid\text{unreported},g,\mathcal F_t] ]
only when conditional independence or the required factorization holds. In general the correct expression is the conditional expected sum of random costs, allowing dependence between the number of missing claims and severity. If slower reports systematically correspond to more serious litigation, a simple average severity for reported cases understates the reserve. If early reports are more severe, it can overstate it. Reported severity patterns must be tested in mature cohorts with transparent selection corrections.
Verrall, Nielsen, and Jessen (2010) demonstrate how paid amounts and reported counts can be used together to model delay before report and delay after report. Their paper supports the methodological separation, not a generic calibration for any specific liability insurer. An insurer's own case-level observations, reconciled to paid and incurred triangles, are required before a predictive claim-level model becomes a reserve estimate for financial reporting.
Synthetic severity-and-window example
Suppose, solely for illustration, that an audited model expects 80 ultimately eligible liability claims for a defined protected cohort, of which 48 have already been reported by t. Let the eligible reporting-completion proportion be 0.60, consistent with that count. The expected true IBNR count is 80(1-0.60) = 32. If the conditional mean ultimate nominal cost per future-reported claim, including the defined expense components, is $50,000, then the expected true IBNR cost is $1.6 million. This figure requires that $50,000 be appropriate specifically for later-reported claims and not merely for the 48 claims already reported.
Suppose further that the 48 reported claims have projected development beyond their current case provisions of $0.45 million, and other separately identified allocated expenses of $0.15 million have not been included in either projected cost. Under these non-overlapping definitions, the illustrative additional provision is $2.20 million. If the $50,000 figure already contains those $0.15 million expenses, adding them separately would double-count. If the protected population is reduced from 80 to 65 after a policy-wording review while the remaining assumptions are unchanged, the true-IBNR count and cost must be recalculated from that new eligibility set; it is incorrect simply to hold the 32 missing claims constant. This example demonstrates the sensitivity of a reserve to coverage classification independently of statistical fit.
Stochastic reserve distributions and calendar shocks
A point estimate can summarize a conditional expectation, but management and capital analysis frequently require the range and shape of potential outcomes. Process volatility arises from random numbers and claim outcomes; parameter risk arises because severity and lag parameters are learned from finite data; model risk arises from incorrect functional assumptions and structure. Correlated judicial decisions, court backlog, inflation, class actions and changes in claims defense can create calendar effects shared across cohorts. Ignoring those dependences understates aggregation risk in some settings. ASOP No. 43 expressly calls for consideration of model, parameter and process uncertainty as relevant to the intended use.
The related discounting question is time-sensitive. The current ASOP No. 20 cash-flow standard was revised in December 2025 and applies to covered work issued on or after June 1, 2026; it expands the prior narrower discounting scope to property/casualty cash-flow analysis. A nominal count-based example cannot be presented as an IFRS present-value measurement or as a regulatory best estimate without further timing, discounting and uncertainty assumptions.
Retroactive Dates and Data Adequacy
The retroactive date as a population filter
A retroactive date determines the boundary before which certain acts or incidents fall outside a claims-made contract, subject to the particular form. It changes the eligible event-history population and therefore the implicit denominator of claims development. A portfolio that moves from an unlimited prior-acts grant to a fixed recent retroactive date can experience an abrupt reduction of old-act reporting, even if underlying negligence, claimant behavior, and court procedures remain unchanged. The observed shorter development tail is partly the contract's selection effect. If the reported history is spliced across the change without an exposure correction, an actuarial model can misread the structural break as improved notification speed.
A U.S. example is the New York regulatory prior-acts discussion. The April 27, 2007 New York Department opinion explains prior-acts coverage and a retroactive date in terms of New York Regulation 121 and the unknown-and-unreported status of certain older acts. It provides useful legal definitions for studying the interaction of policy inception and historical acts. It does not resolve the enforceability of every late-notice or related-claims provision in other jurisdictions or later periods.
An explicit contract extract should record the retroactive date at initial writing and each renewal, replacement, or portfolio transfer. The insured's understanding of continuous coverage may diverge from actual form continuity: a lapse, change of insurer, excluded predecessor entity, altered definition of wrongful act, narrowed related-claims language, or amended notice channel can materially change what is eligible. These are underwriting and legal source facts that an actuarial team must obtain rather than simulate as ordinary development noise.
Consider a hypothetical liability coverage lasting from January 1 through December 31, 2025 with retroactive date January 1, 2021. Suppose an act occurred in 2022, the first written demand was made in October 2025, and notice arrived in November 2025. Under an assumed claims-made-and-reported form covering those facts, the claim may satisfy the timing fields, subject to other conditions. If the first demand instead arrived in February 2026 and there was no appropriate reporting extension or successor policy, it would not be eligible under the assumed 2025 first-made condition. A demand first made in December 2025 but noticed in February 2026 poses a different problem, whose outcome depends on the form's reporting deadline, any extended reporting provision, and governing law. The model must distinguish all three, rather than assign a single event-to-notice lag and infer the same eligibility.
Extended reporting endorsements and runoff
An extended reporting period (ERP) can extend the opportunity to report claims following the termination of a claims-made policy. A policy may offer a short automatic period, an optional purchased period, an unlimited tail for specified insureds, or different conditions for retirement, death or disability. The NAIC SSAP No. 65 ERP provisions distinguish fixed from indefinite extended reporting endorsements for statutory revenue and liability recognition, and specify a separate policy-reserve treatment for certain free extended-reporting benefits. These U.S. statutory details demonstrate that 'tail coverage' is not a single homogeneous reserve quantity.
An ERP can change ultimate eligible claim count through its reporting window while leaving the time of the underlying act unchanged. It can also generate an earlier recognized insurance obligation under a relevant accounting basis because coverage for as-yet unreported claims is provided during the extension. Therefore an actuarial study needs at least two scenarios: expiry with no ERP and the contractually applicable ERP. A single reporting-delay curve applied to both without its eligibility filter would typically overcount one set or undercount the other. Coverage disputes and uncertainty about the completeness of the reporting extension are separate model and data concerns.
A claims-made contract can also be transferred in runoff or commuted. A transaction's insured obligations, reinsurance recoveries and reporting rules must be documented as of the transaction date. A book transferred with historic claims-made policies may contain claims reported but unsettled, claims first made but unnotified during allowed periods, ERP obligations and potentially liabilities under ancillary covers. The exposure definition follows the contracts transferred and any legally effective transfer instrument, not the mere presence of a premium record in the new system.
Cross-jurisdiction accounting and regulatory comparisons
The NAIC 2026 statutory manual is U.S. statutory-accounting guidance, whose SSAP No. 65 treats claims-made events as reported when defined in the policy. It also cross-refers to SSAP No. 55 for unpaid claims and loss adjustment expenses. This classification is analytically distinct from international financial reporting and EU solvency measurement.
The IFRS 17 accounting standard applies to annual reporting periods beginning on or after January 1, 2023. The IFRS Foundation describes groups of contracts measured using current estimates of fulfillment cash flows, discounting, risk adjustment and, where applicable, the contractual service margin. The precise treatment of a claims-made policy requires identifying contractual service, contract boundaries, claims incurred, and the applicable measurement approach. The EU consolidated IFRS adoption regulation incorporates IFRS 17 in the applicable EU-adopted standards; the March 8, 2026 consolidation is the current version identified by EUR-Lex at the research cutoff. Its accounting provisions do not themselves establish a claims-made contractual coverage trigger, which must be determined from the relevant policy and governing law.
In EU prudential valuation, Solvency II Article 77 requires a probability-weighted discounted best estimate plus a risk margin, and the non-life claims provision rule requires separate claims and premium provisions. The EU provision for historical claim events incorporates reported and unreported cases within the applicable insurance obligation. A statistical projection of every unreported harmful event would therefore be overbroad unless the legal and contract-boundary filter is applied. Conversely, a reporting-based classification alone can conceal future cash-flow obligations attached to validly reported claims or ERP coverage. A useful international comparison maps the obligation first and only then selects the reserve metric.
The IAIS Insurance Core Principles provide a worldwide supervisory framework, with relevant revisions last reported in December 2024. They establish the international supervisory setting and cannot be treated as directly enforceable law in every IAIS jurisdiction. National implementation, the particular valuation regime and the policy wording remain separate objects of verification. The comparative result is an ordering of analytical steps: define contract protection, identify reporting status, determine accounting obligation, forecast cash flows and disclose valuation uncertainty.
Data lineage, reporting-channel changes, and completeness
Claim-reporting data are created by a chain of organizations and systems. A claimant may contact the insured, broker, claims administrator, group captive, insurer, or reinsurer in sequence; policy wording can make receipt by one person legally different from receipt by another. A date in a claims database may be the earliest notice, a converted legacy timestamp, a date automatically populated after import, or the date an adjuster confirms a complete claim. Without a source-system map, a purported notification lag can partly measure migration history.
The ASOP No. 23 data-quality standard requires attention to selecting, reviewing and using data and to reliance on data supplied by others. Its application to a claims-made study begins with a data dictionary identifying the provenance, original form and transformation of each decisive coverage field. The insurer should reconcile counts and amounts at policy, insured, claim and transaction levels, particularly at renewal, year-end cutoffs and reporting-system transitions. The existence of a retroactive date in a spreadsheet is insufficient unless it can be linked to the operative endorsement or schedule and the relevant insured entity.
Data adequacy should be evaluated by identifiable strata: policy form, cover type, insured occupation or industry, accident or wrongful-act vintage, claims-made year, notification year, coverage decision, severity range, jurisdiction, claim type, and ERP type. A sparse sample with a long unobserved tail can still support a scenario or prior-based estimate, but the confidence interval will depend strongly on the prior. An aggregate paid triangle can be computationally complete while its individual notice fields are missing, leading to inconsistent conclusions if the two are presented as equivalent evidence about reporting behavior.
Selection, comparability, and contractual counterfactuals
A change of policy wording creates a selection problem: reported case histories from the old form are drawn from a different eligibility population than the new form. To compare them scientifically, define a target coverage population and, where possible, reclassify historical events under a common hypothetical wording using documented case attributes. This procedure is a counterfactual sensitivity exercise, not a retrospective alteration of the historical contract. If necessary claim-made dates, event dates or related-claim identifiers are missing, some reclassification is unidentified. The uncertainty should be represented as a bounded scenario or missing-data range rather than a point estimate that silently assumes coverage.
Suppose the observed cohort has 200 event histories, of which 140 generate eventual demands, but only 110 satisfy the stated retroactive and first-made rules and 95 satisfy the additional reporting window. For that hypothetical contract the eligible count is 95, not 200 and not 140. A model calibrated on all 140 demands without the 45 disqualified cases being identified would infer a spurious higher ultimate. If an ERP adds protection for 15 otherwise late notifications, the eligible count becomes 110 under the new assumed contract. These numbers illustrate selection mechanics; they establish no empirical frequency or legal result in actual insured business.
Evidence, Uncertainty, and Model Validation
Source hierarchy and evidentiary limits
This research gives the strongest normative weight to the text of operative policy forms and applicable laws in the particular jurisdiction, then to the accounting and actuarial standards used for the specified valuation purpose. Published actuarial studies support estimation methods and limited empirical propositions tied to their studied portfolios. A regulatory interpretive opinion supports its identified jurisdiction, date and quoted legal framework. A contemporary journal article can test the plausibility of a model mechanism without establishing that the same parameters hold elsewhere. The source-to-claim evidence ledger records this difference explicitly.
The full ASOP No. 43 standard states that the actuary should consider the purpose, appropriateness of methods, assumptions, material changing conditions and uncertainty. The ASOP No. 56 modeling standard was adopted in December 2019 and applies from October 1, 2020 to its defined scope; it emphasizes model purpose, data, assumptions, model risk and output validation. These U.S. professional practice standards set process-level requirements for covered work, rather than prescribe a universal reporting-delay distribution or select the reserve for a particular insurer.
A validation report must identify four cutoffs: the contract period, the accounting date separating paid and unpaid amounts, the data valuation date through which transactions have been loaded, and the date to which materially relevant information is reviewed. This follows ASOP No. 43 disclosure guidance, which distinguishes these dates. A claim made before the accounting date but entered later may have been unknown to the reported database while having contract significance; it needs documented treatment under the chosen accounting regime. Backdated imported transactions should be reconciled rather than assumed to have been known by the original valuation.
A validation program for reserve estimates
First, verify source-document and coverage classification. Review samples from each major policy form and renewal vintage, with targeted examination of retroactive dates, related-claims clauses, notice requirements, automatic reporting periods and purchased ERPs. Compare the legal notice timestamps and their recipients with the database field used in the model. Assign an evidence-quality status to each decisive field and quantify missing or ambiguous records. In regulated or litigated cases, a qualified legal review establishes the correct coding for the applicable jurisdiction; a probability model does not adjudicate coverage.
Second, reconcile population and money. Counts by first-made date, notice date and event date must reconcile, with recorded explanations for duplicates, merged claims, reopened matters, reinsured cases and coverage denials. Paid, case-incurred and recoverable transactions should reconcile to financial-system totals. Allocation of defense costs inside or outside limits, deductibles, and reinsurance should be tested at claim and portfolio levels. Failure of the reconciliation indicates that model predictions could be correct relative to an incomplete extract yet wrong for the booked obligation.
Third, construct an as-of historical backtest. For each retrospective valuation cutoff t0, refit the eligible population and estimate future unreported counts using only policy, transaction and external information available by that date. Evaluate against emergence later observed and adjudicated under those policies. Compare results across accident, first-made, notice and processing-age cohorts. Measure count calibration, delay quantile coverage, cumulative paid error, unpaid reserve bias, predictive interval coverage and sensitivity to major-case exclusions. Use a rolling calendar sequence of cutoffs rather than selecting only one favorable period.
Fourth, compare at least three conceptually different approaches where credible data permit: an aggregate development triangle; an exposure- or prior-expected frequency method; and a claim-level delay/occurrence model. The 2016 real-data delay study offers a methodological example of individual-claim modeling and comparison with a Poisson chain-ladder approach. The 2010 RBNS/IBNR study demonstrates the value of separating pre-report and post-report delay. Requiring that all models produce identical numbers is inappropriate; documenting why they diverge is essential to understanding misspecification and the relevant economic or contractual changes.
Fifth, challenge structural changes explicitly. Re-estimate after excluding system-migration periods, large cases, delayed batch uploads, emergency notification programs and years when policy wording changed. Stratify by legal venue, business class, report source and ERP. Evaluate severity conditioning on late reporting, including uncertainty in the upper tail. Simulate different lapse and renewal patterns to determine whether policy-year exposure controls reproduce future first-made eligible claims. If the risk of an unobservable old-act claim is extreme, external qualitative legal evidence can support bounds but cannot make the missing distribution empirically identifiable.
Dependence, changing conditions and tail uncertainty
An insurer can experience simultaneous changes in frequency, reporting delay, ultimate severity and settlement timing. An active litigation environment may increase early notification, delay court disposal and increase legal costs at once. A claims-processing backlog can shift apparent reported counts and payments in different directions. Lindskog and Wüthrich (2025) provide an operational backlog mechanism explaining how processing capacity distorts observed development. The implication for causal interpretation is strong: even perfect knowledge of a paid triangle does not uniquely separate claim incidence, legal-reporting delay, administrative throughput and severity escalation.
The outer tail creates an identification problem. Very old acts with future demands may have no historical observation within the insurer's data horizon, particularly under newly introduced prior-acts coverage. Tail assumptions can use relevant external studies, expert priors and stress testing, with the transferability limitations documented. England and Verrall's 2002 reserve variability review explains why predictive distributions and tail extrapolation require explicit assumptions. Mack's chain-ladder reserve-error method yields precision information conditional on its stated data-generating assumptions; the numerical standard error is not a guarantee that an omitted contract boundary has no effect.
Claim reporting may be modeled by cause-specific hazards for legal demand, valid notice, termination and other competing endpoints. That is an analytical option, not a requirement imposed by any cited standard. Where a large portfolio has reliable individual timestamps and enough late-reporting observations, it can reveal nonlinear relationships absent from aggregate triangles. In a small specialty book, an elaborate model may overfit and produce an untestable tail. Selection should follow data maturity, decision purpose, explanatory fit and reproducibility.
Discounting, cash-flow timing and disclosure
A true IBNR count multiplied by ultimate severity is a nominal aggregate in the illustrative examples. A regulatory or financial-reporting reserve can require expected future payments by calendar period, discounted at a purpose-appropriate interest curve, plus or minus specified adjustments. Under the EU Solvency II best-estimate rule, the estimate takes the probability-weighted present value of future cash flows plus a separate risk margin at the technical-provisions level. Under IFRS 17, fulfillment cash flows include current estimates, discounting and risk adjustment subject to the standard's full measurement architecture. The professional 2026 ASOP No. 20 revision addresses U.S. actuarial cash-flow analysis, including discounting, under its own scope.
A transparent report should give the gross nominal estimate, allocated expense components, future cash-flow timing, recoverables, net position, discount-rate basis, risk or uncertainty adjustments, estimation date, valuation date and the applicable authoritative reporting regime. An expected value, a management-selected margin, a percentile and a booked liability have distinct mathematical or legal roles. A projection made for underwriting scenario analysis should be labeled accordingly rather than described as a certified statutory reserve.
Documentation and editorial audit
The ASOP No. 41 communications standard remains the adopted standard on actuarial communication; a publicly available proposal for its revision should not be confused with adopted text. The scientific manuscript and accompanying source ledger here distinguish source quotations, interpretations, mathematical identities and synthetic examples. All cited historical papers retain their original years rather than their later online posting dates. The 2025 online reporting-delay study by Abdelrahman and colleagues is identified consistently with its 2026 journal issue. Current standard editions and effective dates are stated for the date of research. This prevents retrospective importation of a later standard into earlier reserve valuations.
The study's empirical boundary is explicit. It has not independently examined a confidential insurer's claim register, matched policy schedules to individual litigated claims, or estimated a fitted delay curve for a real liability insurer. It does not claim an externally audited numerical reserve, a full all-jurisdiction coverage survey, or an empirical causal estimate of social inflation. Such work would require the underlying records, policy documents, jurisdiction-specific legal analysis and independent fit and validation data. The research supports reasoned model choice and a replicable validation design rather than fabricated observed results.
Integrative Scientific Synthesis and Research Conclusions
A reserve is a statement about a defined obligation at a defined information date
The comparative evidence leads to a specific interpretation. A reserve estimate for claims-made liability exposure is a conditional statement about future cost arising from a legally defined covered population, observed through an incomplete claims-information system at a specified date. The total future economic burden from wrongful acts can be larger than that population; the population of case records in the insurer's system can be smaller. The distinction is operational: a claims-made retroactive date changes possible coverage, the first-made and reported conditions change attachment, the claims administrator changes visibility, and the accounting regime changes measurement. An aggregate triangle preserves those distinctions only if its cohort and data history have been designed to preserve them.
A causal sequence can therefore be reconstructed without inventing a new theoretical doctrine. Exposure and underlying acts generate potential demands; policy wording and legal rules filter those demands into potentially covered claims; parties and reporting channels produce valid notices; information systems register cases; case handling and litigation generate paid and incurred development; accounting rules select the form and components of the reported obligation. Each stage has evidence requirements and may vary over calendar time. This ordering helps explain why superficially identical lag curves can imply different reserve outcomes across claims-made and occurrence portfolios.
Contract-conditioned completeness is the operative quantity
The usual empirical question, 'what fraction of claims are reported after k months?', is underspecified for reserving. A better specified question is, 'of all claims that will satisfy the contract's actual attachment and reporting requirements for this cohort, what fraction has validly reached the valuation information set at age k?' That completion proportion can be estimated where mature eligible cohorts and consistent registration histories are available. It cannot be inferred purely from a count of historical harmful events when future first-made dates fall outside the protected interval. The NAIC claims-made classification and New York prior-acts interpretation are independent primary-source support for the substantive importance of this filter.
Reporting-lag differences are therefore not just parameters; they can be evidence that the target population changed. For example, if a renewal shortens retroactive protection, fewer older-act claims may ever satisfy coverage and a fitted tail factor can decline. The resulting reserve change must be decomposed into reduced eligible exposure and changed delay among still-eligible claims. Without that decomposition, a historical development factor may be mechanically correct but economically misleading.
Count emergence and economic cost require separate tests
The reported-count and paid-amount model of Verrall, Nielsen and Jessen demonstrates the conceptual importance of a pre-report versus post-report distinction. The corresponding practical conclusion is to estimate late claim numbers, their conditional severity, and the future development of already-reported claims as separate components where data support separation. A count-based IBNR model with a stable tail does not prove stable ultimate indemnity or defense expense. Conversely, a paid triangle with reliable projected ultimate costs can mask changes in the timing of legally valid notice.
The source evidence also cautions against assigning an evidentiary superiority to complexity. A simple Bornhuetter–Ferguson estimate with explicit prior and coverage limitations can be more robust in a small newly written specialty program than an unvalidated machine-learning hazard calibrated to a short, censored sample. In a mature high-volume portfolio, individual-level covariates and explicit right-truncation correction may improve measurement. The ASOP No. 56 modeling guidance emphasizes appropriate design, validation, and intended use rather than a required algorithm.
International comparability follows contractual and accounting comparability
U.S. statutory accounting, IFRS reporting and EU solvency rules ask related but differently specified measurement questions. A common numerical data set can produce different formally appropriate measures because the recognition, discounting, risk-adjustment and contract-boundary definitions differ. The IFRS 17 measurement description and EIOPA Article 77 valuation rule support the distinction from a raw nominal loss projection; the NAIC SSAP No. 65 source supplies the contrasting U.S. statutory reporting basis. Numerical comparisons should state all three layers: covered population, timing and measurement regime.
A global scientific comparison should preserve original policy-language and reporting-law specificity. Claim made, claim reported, discovery period and retroactive date are not necessarily exact equivalents in every translation or national insurance product. Comparative research must record the governing language, date of form adoption, applicable mandatory legal modifications and the accounting standard used. This is the basis for credible international comparisons without implying unverified universal legal uniformity.
Research propositions supported by the evidence
The first conclusion is that claims-made reporting lag is a joint contractual and statistical phenomenon. The second is that true IBNR, later development on known claims and additional adjustment expense must be separated or their aggregation disclosed. The third is that cohort comparability depends on continuity of retroactive dates, claims-made definitions, notice windows and related-claims handling. The fourth is that administrative registration lag is a potential source of artificial development movement and must be audited independently of legal notice. The fifth is that reserve validation requires as-of backtesting, explicit coverage filters, alternative estimation methods and sensitivity to historical changes. Each conclusion follows from the identified contracts, standards and scholarly models; none is presented as a new named model or scientific law.
Further empirical questions
Empirical priorities include identifying how notification hazards vary with different notice formulations; determining whether severity and late reporting are positively related in particular professional-liability, product-liability and environmental books; measuring the actuarial effect of ERP purchases and policy replacement; quantifying the distortion produced by migrating legacy claims systems; and testing whether report-year paid development remains stable when the mix of historical wrongful-act vintages changes. These questions are testable only with correctly defined case-level and policy-level observations. Specific national-law findings must be added after examination of operative policy forms and enforceable local decisions; neither supervisory principles nor journal models substitute for that evidence.
Conclusion
Liability claims-made reporting lag connects a legal timing rule with a stochastic emergence process. It must be measured against a precisely identified coverage cohort, reporting definition and valuation information set. Retroactive dates and reporting extensions select the future demands that can become covered claims. Claimant behavior, notification channels and administrative backlogs shape the delay observed by the insurer. Settlement and case-reserve development then determine the future cost of reported cases, while claims not yet validly observed create additional potential cost within the protected population.
The reviewed authoritative materials establish complementary disciplines. NAIC statutory accounting distinguishes reporting-based coverage categories; New York's specific insurance opinion defines prior-acts protection in its jurisdiction; ASOP No. 43 requires purpose-specific unpaid-claim estimation; ASOP Nos. 23, 56, 41 and the revised 20 govern relevant professional practices and communications in their U.S. scopes; IFRS 17 and EU Solvency II establish differently constituted financial-reporting and prudential measures. The actuarial literature supplies the historical foundations and modern methods for aggregate and claim-level estimation of reporting delay, IBNR, IBNER and payment development.
The analytical synthesis has five operational consequences. First, a reserve cannot be inferred from historical injuries alone without proving the insurance eligibility set. Second, a lag distribution cannot be transferred across policy forms without verifying cohort comparability. Third, true unreported claims and subsequent development on reported claims should be estimated distinctly wherever reliable data permit. Fourth, validation must reproduce the information that was known at each historical valuation cutoff and expose administrative and contractual breaks. Fifth, any final figure must be reconciled to the measurement objective, including expenses, gross and net recoverables, cash-flow timing, discounting and uncertainty. These consequences form an auditable scientific foundation for claims-made liability reserving practice and future insurer-specific empirical research.
References
Abdelrahman, H., Badescu, A. L., Craiu, R. V., & Lin, X. S. (2025 online; 2026 issue). Marked Cox models for IBNR claims count: Continuous and discretized approaches with Dirichlet-driven reporting delays. ASTIN Bulletin: The Journal of the International Actuarial Association, 56(1), 60–88. https://doi.org/10.1017/asb.2025.10056.
Actuarial Standards Board. (2011 revision; effective September 1, 2007). Actuarial Standard of Practice No. 43: Property/Casualty Unpaid Claim Estimates. https://www.actuarialstandardsboard.org/asops/propertycasualty-unpaid-claim-estimates/.
Actuarial Standards Board. (2010 revision; effective May 1, 2011). Actuarial Standard of Practice No. 41: Actuarial Communications. https://www.actuarialstandardsboard.org/asops/actuarial-communications/.
Actuarial Standards Board. (2016 revision). Actuarial Standard of Practice No. 23: Data Quality. https://www.actuarialstandardsboard.org/asops/data-quality/.
Actuarial Standards Board. (2019 adoption; effective October 1, 2020). Actuarial Standard of Practice No. 56: Modeling. https://www.actuarialstandardsboard.org/asops/modeling-3/.
Actuarial Standards Board. (2025 revision; effective June 1, 2026). Actuarial Standard of Practice No. 20: Analysis of Property/Casualty Cash Flows, Including Discounting. https://www.actuarialstandardsboard.org/asops/asop-no-20-analysis-of-property-casualty-cash-flows-including-discounting/.
Bornhuetter, R. L., & Ferguson, R. E. (1972). The actuary and IBNR. Proceedings of the Casualty Actuarial Society, 59, 181–195. https://www.casact.org/abstract/actuary-and-ibnr.
England, P. D., & Verrall, R. J. (2002). Stochastic claims reserving in general insurance. British Actuarial Journal, 8(3), 443–518. https://doi.org/10.1017/S1357321700003809.
European Commission. (2015, as reflected in EIOPA's rulebook). Commission Delegated Regulation (EU) 2015/35, Article 36: Non-life insurance obligations. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2449_en.
European Parliament and Council. (2009, as reflected in EIOPA's rulebook). Directive 2009/138/EC (Solvency II), Article 77: Calculation of technical provisions. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2160_en.
European Commission. (2023, consolidated March 8, 2026). Commission Regulation (EU) 2023/1803 adopting certain international accounting standards in accordance with Regulation (EC) No. 1606/2002. Official Journal of the European Union, L 237, September 26, 2023, pp. 1–992 (original act). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A02023R1803-20260308.
IFRS Foundation. (2017; effective for annual periods beginning on or after January 1, 2023, with subsequent amendments). IFRS 17 Insurance Contracts. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/.
International Association of Insurance Supervisors. (2011; updated December 2024). Insurance Core Principles and ComFrame. https://www.iais.org/activities-topics/standard-setting/icps-and-comframe/.
Lindskog, F., & Wüthrich, M. V. (2025). Eliciting claims development patterns and costs hidden in backlogs. European Actuarial Journal, 15, 667–705. https://link.springer.com/article/10.1007/s13385-025-00431-w.
Mack, T. (1993). Distribution-free calculation of the standard error of chain ladder reserve estimates. ASTIN Bulletin, 23(2), 213–225. https://doi.org/10.2143/AST.23.2.2005092.
National Association of Insurance Commissioners. (2026, March edition). Accounting Practices and Procedures Manual, volume 1, SSAP No. 65: Property and Casualty Contracts; cross-reference to SSAP No. 55. https://content.naic.org/sites/default/files/publication-app-manual.pdf.
New York State Insurance Department, Office of General Counsel. (2007, April 27). Opinion No. 07-04-14: Retroactive Liability Insurance and Related Questions. https://www.dfs.ny.gov/insurance/ogco2007/rg070414.htm.
Schnieper, R. (1991). Separating true IBNR and IBNER claims. ASTIN Bulletin, 21(1), 111–127. https://doi.org/10.2143/AST.21.1.2005404.
Verrall, R. J., Nielsen, J. P., & Jessen, A. H. (2010). Prediction of RBNS and IBNR claims using claim amounts and claim counts. ASTIN Bulletin, 40(2), 871–887. https://openaccess.city.ac.uk/id/eprint/3807/.
Verrall, R. J., & Wüthrich, M. V. (2016). Understanding reporting delay in general insurance. Risks, 4(3), article 25. https://www.mdpi.com/2227-9091/4/3/25.