Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Disability Income Claim Reserves by Occupation and Duration

Abstract. Disability income claim reserves are conditional estimates of the cost of insured periods of incapacity that have already begun. Their measurement depends on time already spent disabled, the work and earnings to which eligibility refers, contract-specific benefit rules, recovery and death, administration, and the financial basis of valuation. Occupation enters this measurement through functional demands, income, claim selection, disability definitions, claims management, and the composition of each surviving claim cohort. Duration enters through the evolving distribution of underlying impairment, survival of claims, waiting periods, eligibility reviews, changes from own-occupation to other definitions of disability, and the remaining benefit term. This research dossier integrates original insurer experience studies, actuarial reserving standards, occupational classification frameworks, and comparative financial regulation. It distinguishes actual-to-expected termination ratios from transferable termination probabilities, demonstrates the effect of occupation-duration interactions on prospective benefit reserves, and derives a reproducible illustrative sensitivity analysis. A review of the September 2026 SOA/LIMRA group long-term disability termination report reveals pronounced differences across duration and broad occupational classes, while its restricted insured population prevents unqualified transfer to individual disability policies or other markets. The central finding is that occupational stratification is useful only when the underlying claim states, contract terms, exposure measures, and credibility are aligned. Reserve adequacy is ultimately established through coherent cash-flow estimation, empirical monitoring, and reconciliation to the applicable statutory, accounting, and supervisory measurement objective.

Keywords: Disability income insurance; disabled-life reserves; group long-term disability; individual disability income; occupation class; disability duration; recovery hazard; mortality hazard; competing risks; benefit termination; actual-to-expected ratio; own-occupation definition; conditional survival; actuarial credibility; technical provisions; incurred claims; IFRS 17; Solvency II; reserving validation.

Introduction

The central unit of analysis is a claim already incurred as of a defined valuation date. For that claim the insurer has information about an insured worker's occupation, age, date of disability, diagnosis, contract, earnings, benefit amount, paid and unpaid periods, and current claim status. A prospective disabled-life reserve estimates the future contractual benefits and appropriate settlement expenses arising from that incurred claim, subject to subsequent recovery, death, loss of eligibility, benefit expiry, and other legally effective transitions. This valuation is made for a specific accounting or regulatory purpose. It is distinct from the premium or contract liability for disability claims that may occur in the future.

The research question is how occupational classification and elapsed duration jointly alter the conditional expectation of unpaid disability income benefits. The answer requires more than multiplication of a monthly benefit by a generic expected duration. The future probability that a benefit remains due changes with past duration; benefit amounts can change with offsets, earnings and indexation; and contract definitions can change at specified anniversaries. Actuarial Standard of Practice No. 5, Incurred Health and Disability Claims, explicitly addresses tabular estimation of known long-term claims and emphasizes recovery, mortality, and governmental offsets.

The empirical anchor is the SOA Research Institute and LIMRA 2015–2024 Group LTD Termination Study, publicly released in September 2026. This source provides observed-to-expected comparisons by duration, diagnosis, elimination period, short-term disability accompaniment and broad collar classification. Its results are compared with the 2019 Group LTD study, individual disability evidence, and the established U.S. valuation-table history. The analysis also considers the international conceptual distinction between work limitations, insured benefit eligibility, and public disability status, drawing on the WHO International Classification of Functioning.

Methodologically, this work combines primary-document analysis, comparative actuarial interpretation, established competing-risk survival mathematics, deterministic cash-flow examples, and critical assessment of exposure and selection. The numerical example is explicitly hypothetical and reproducible. Neither raw individual claims nor the subscription-only SOA/LIMRA aggregated dashboards were supplied to this study; no portfolio-specific fitted hazards, empirical causal effects, or externally validated occupational coefficients are claimed. The contribution consists of documented analytical integration, identification of estimand and classification problems, and a transparent reserve-sensitivity demonstration.

Conceptual Foundations and Historical Development

The claim reserve as a conditional liability

An incurred disability income claim can create a stream of contingent monthly payments extending years beyond the date of disability. A reserve for a known disabled claimant depends on what has happened up to valuation: elapsed disablement, claim status, benefits already settled, satisfaction of the elimination period, and effective contract conditions. In U.S. actuarial practice the term disabled-life reserve frequently describes the case-level prospective reserve. The broader incurred-claims liability can additionally include incurred-but-not-reported claims, reported claims awaiting adjudication, processed-but-unpaid benefits, and relevant settlement costs. These components should be separately inventoried so that the total is both complete and free of duplication.

The American Academy of Actuaries' Group LTD Valuation Standard report (2013) identifies reserves for known ongoing disabled lives as a major liability for group LTD carriers and describes seriatim valuation using expected remaining payments and termination assumptions. The NAIC Health Insurance Reserves Model Regulation provides model minimum standards for claim, premium, and contract reserves; its prospective gross-premium adequacy test is conceptually broader than the mere selection of one disabled-life table. A model regulation is a regulatory model: actual binding valuation requirements depend on jurisdiction, adopted rules, applicable dates, and insurer-specific provisions.

A claim reserve must be distinguished from four adjacent quantities. First, an outstanding payment that is unconditionally due is a payable, although reporting conventions may include it within claim liabilities. Second, an incurred-but-not-reported estimate covers events already occurring but absent from reported claim inventory. Third, the liability for remaining coverage concerns future insurance service and insured events under a recognized group of contracts. Fourth, solvency capital or a regulatory risk margin addresses uncertainty beyond an underlying best estimate under the applicable regime. These distinctions determine which probabilities, expenses, margins and discount rates belong in each reported figure.

Historical tables, population boundaries, and experience revision

Early disability valuation relied on pooled industry tables with limited segmentation relative to modern insurer data. The Academy's 2013 report traces a shift from the Commissioner's Group Disability Table 1987 to experience based on group LTD claims from 1997–2006; its discussion of the 2008 experience table separates death and recovery and recognizes diagnosis, gross monthly benefit and definition-of-disability effects. The proposed 2012 GLTD valuation basis adds explicit attention to experience credibility and graduation. A valuation table embeds a regulatory and actuarial measurement purpose, whereas an experience table summarizes observed terminations under defined study protocols. Direct replacement of one by the other without margins, exposure adjustment, or legal checks can be wrong.

The SOA 2013 Individual Disability Income Valuation Table documents a separate statutory reserve morbidity basis approved by the NAIC in August 2016 for individual disability income policies. Individual insurance typically differs from group LTD in underwriting, insured benefit levels, occupational selection, contractual riders and lapse behavior. The SOA Individual Disability Experience Committee's 2006–2014 termination analysis explicitly compares later observed experience against the 2013 IDI valuation-table termination assumptions. These source histories show why an occupational effect inferred from one insurance market cannot simply be imported into another.

The 2015–2022 Group LTD Incidence Study examined about 294 million insured life-years and roughly 1.2 million claims from 19 companies, with specific restrictions including fully insured group business and exclusions for short elimination periods and certain other product types. Incidence denominators are insured lives at risk of a new claim, not disabled lives already in payment. Incidence is critical to prospective premiums and future coverage liabilities, but a disabled-life reserve for a known claim principally uses conditional persistence, payments and remaining eligibility. Incidence study results can inform the composition of a claim cohort, but their denominator cannot be substituted for a termination exposure denominator.

International limits of the insurance construct

The WHO's functioning classification treats disability in relation to health conditions and environmental factors. That broad measurement domain is useful in evaluating occupational demands, but a private policy may impose a narrower threshold, waiting period and benefit formula. Disability status under a public social-insurance program can employ a separate definition. For example, the U.S. Social Security Act § 223 contains statutory disability entitlement rules materially different from every private contract's possible wording. Government benefits may enter private reserving as offsets, but an award of public benefits does not automatically establish the same contractual entitlement under every private policy.

Under IFRS 17 Insurance Contracts, incurred-claims obligations are measured through fulfillment cash flows, with current estimates, discounting and a risk adjustment for nonfinancial risk. IFRS' key-terms guidance separately defines liability for incurred claims and liability for remaining coverage. Solvency II Article 77 instead frames technical provisions as best estimate plus risk margin and requires expected present-value cash-flow measurement. Both frameworks require purpose-specific application: an IFRS 17 risk adjustment, a Solvency II risk margin and a U.S. statutory valuation margin are not interchangeable objects.

The estimand and its conditioning information

Let the valuation date be zero. Let a claim have occupational classification o, elapsed disability duration d, current attained age x, policy and claim state z, and future calendar month m. Define the insured event of interest as remaining in a payable state at payment date m, rather than merely being alive. The desired prospective expected benefit value is the conditional expectation of contractually payable benefits and eligible costs, discounted according to the governing basis. The conditioning variables include claim start, current status, benefit period, age at onset, benefit definition and offsets. Occupation is an information-bearing covariate, not itself an independent legal cause of benefit termination.

This framing resolves a persistent analytical ambiguity: an observed termination can represent death, functional recovery, earnings recovery, exhaustion, a revised disability definition, an administrative closure, an adjudicated denial or some other coded exit. Only the first two are universally biological events. When estimating probabilities to support future payments, every exit definition must be reconciled with whether entitlement truly ceases, whether appeal or recurrence restores payments, and whether the observation is censored rather than terminated.

Occupational Segmentation

Occupation as work exposure and as policy classification

An occupation influences disability income risk through physical and cognitive task demands, exposure to injury and illness, ability to accommodate limitations, wage stability, vocational alternatives, employer accommodation, and return-to-work opportunities. Occupational descriptions change across employers and over time. A job title can conceal material differences between an office-based supervisor, a field technician and a self-employed professional. An occupational rating class may combine heterogeneous jobs to satisfy pricing, underwriting and administrative objectives; it is not automatically a granular clinical or labor-market classification.

The U.S. Bureau of Labor Statistics' 2018 Standard Occupational Classification and the International Labour Organization's ISCO-08 occupational classification provide independent statistical taxonomies. The European Commission's ESCO occupations framework links occupational concepts and skills to ISCO-08. These systems are useful for harmonizing job records, but a crosswalk from an insurer's internal occupation class must preserve uncertainty and information loss. “Blue collar,” “gray collar” and “white collar” are coarse study descriptors, not globally invariant occupational definitions.

The 2026 SOA/LIMRA termination report, Table 8, reports recovery actual-to-expected ratios of 159% for blue-collar claims, 161% for gray-collar claims and 135% for white-collar claims, against 150% in aggregate. Corresponding mortality A/E ratios are 80%, 81% and 83%. These are ratios to expected events under the study's 2008 reference table, not unconditional recovery or death probabilities. The result supports occupationally differentiated experience review; it does not prove that occupational type causally determines recovery or that the same coefficients apply to all claim durations.

Confounding, selection, and composition

Occupation-specific observations are conditional on being insured, experiencing an eligible disability, surviving and remaining open until the observed duration, and being included under the study's data protocol. This produces several selection mechanisms. Underwriting changes the distribution of insured workers. A longer elimination period changes which episodes reach payable status. Employer-sponsored coverage may overrepresent larger, stable employers. Claimant age, diagnosis, earnings, geographic region, employer benefits, and adjudication practices differ by occupation. These variables may explain part of an apparent occupational gradient.

A comparison of recovery rates for occupations with different mixtures of back injuries, cancer, neurological conditions and mental-health diagnoses is a mixture comparison, not a clean occupational effect. Similarly, an own-occupation-to-any-occupation definition change can produce a spike in administrative or vocationally defined terminations at a contract anniversary. A hazard model that includes the change as a time-dependent policy variable may attribute less unexplained effect to occupation. The study's broad collar A/E ratios cannot identify these conditional effects without joint exposure tables or claim-level information.

A robust occupational model therefore distinguishes the classification used at underwriting, the actual job at disablement, the occupation applicable under the benefit definition, the occupation or earnings at subsequent review, and the labor-market opportunities available for return to work. These fields can differ within the same claim. The estimated reserve should reflect the actual contract and claim record rather than a single occupational label applied indiscriminately to future states.

Credibility and rare cells

Fine occupational splitting creates sparse cells, particularly at long duration. Limited exposures, delayed data and rare competing events can produce volatile maximum-likelihood rate estimates. The Actuarial Standards Board's ASOP No. 25, Credibility Procedures, describes assessing subject experience and blending it with relevant evidence. A practical calibration may partially pool occupation-by-duration cells toward a common baseline while retaining documented, credible differences. Such pooling is an established statistical practice, not a new proprietary actuarial theory.

A model can use stratification, generalized linear or additive models, frailty structures, hierarchical Bayes, or credibility-weighted standardized rates. Each approach must make the target risk set explicit, allow for the claimant's attained age and calendar period, and avoid extrapolating a thin long-duration cell with unwarranted precision. A gradient should be treated as a current empirical estimate with uncertainty bounds and reviewed whenever underwriting, product wording or claim management changes.

Equity, classification governance, and data minimization

Occupational stratification also poses practical governance questions. It may improve reserve accuracy for contractually relevant work-capacity differences, yet classifications used for pricing, claim acceptance and financial estimation have distinct purposes. A reserve model should not silently determine eligibility or claims decisions. Occupational data must be current, auditable and appropriate to the applicable jurisdiction's privacy and anti-discrimination law. Where a classifier serves as a proxy for age, health, sex, employment insecurity or other protected attributes, interpretation of the classifier requires legal and statistical scrutiny. The reserve remains an expected contractual payment, not a judgment about an individual claimant's worth or capacity.

Duration and Termination Assumptions

Multiple clocks in one claim

Disability income reserving involves at least five possible clocks: time from medical onset, time from first inability to work, time since contractual incurral, time since the beginning of payable benefits, and calendar time. They may differ by days or months. The elimination period determines when benefits may commence; a valuation at 18 months since disablement does not necessarily imply 18 monthly payments already made. At the same time, duration to a contractual definition switch or maximum benefit date can be measured from whichever trigger the contract specifies.

The 2026 SOA/LIMRA public report explicitly uses duration since disability for its duration tables. A model calibrated to that axis should preserve this origin or perform a documented conversion when using payment-duration data. An analysis that joins two studies with inconsistent origins can generate artificial changes in early recovery hazards. Calendar time is independently material because economic conditions, claims practice, judicial interpretation, diagnostic coding and employment opportunities evolve.

Conditional survival and competing exits

Let R denote a recovery-related exit, D death, and E other loss of payable entitlement. Define cause-specific monthly hazards q_R(o,d,z), q_D(o,d,z) and q_E(o,d,z) conditional on being in the payment-eligible risk set at the start of a month. Under a discrete mutually exclusive exit convention, the conditional probability of still being payable next month is p(o,d,z) = 1 - q_R - q_D - q_E, provided the probabilities are measured for the same risk set and period. Future survival to payment month m is the product of the month-by-month transition probabilities, allowing covariates to evolve. If hazards are continuous, the cumulative incidence for each cause is obtained by integrating its cause-specific hazard against survival of all competing causes; separate Kaplan–Meier curves that censor other exits are not equivalent to cause-specific cumulative incidence.

An observed recovery/termination label should be mapped to future cash-flow entitlement. Recovery may be full work resumption, a sufficient earnings recovery, a deemed vocational recovery after a definition change, or a recorded administrative determination. In a standard simplified model these events may be collapsed into one payment-ending hazard. A more detailed contract with partial disability, recurrent claims or contested determinations benefits from additional states, for example full-pay disability, partial-pay disability, suspended benefit, returned to work, death, exhausted benefits, and appealed closure. Those states can generate future payments through recurrence or appeal; absorbing recovery is then a modeling assumption to test.

The 2013 Academy GLTD valuation analysis emphasizes separated death and recovery termination behavior. This distinction remains important because cause-specific trends can diverge and because a group of policies with the same total termination count may have materially different risk implications if death and recovery evolve differently. An increasing recovery hazard tends to shorten payments; changes in mortality affect both payments and the composition of survivors. The direction of the combined reserve change depends on the timing and size of the rates and on benefits following each transition.

Direct evidence of duration dependence

The 2026 SOA/LIMRA duration-by-duration results, Table 4, report mortality A/E of 54% and recovery A/E of 119% in the first quarter, versus 87% and 236% in quarter 10; recovery A/E reaches 208% in quarter 9. The all-duration totals are 81% for mortality and 150% for recovery. The comparison basis is the 2008 LTD experience table. A/E ratios change across duration partly because the baseline expected rates themselves change; a rising recovery A/E does not imply that the absolute recovery probability is rising with elapsed disability duration. The correct calculation requires expected event counts or baseline rates for each duration cell and an appropriate treatment of claim exposure.

The study reports that the ending of own-occupation periods has a more pronounced effect on recoveries than estimated by the 2008 table. This provides a concrete warning against smoothing away contract anniversary effects. It also finds different recovery A/E ratios by elimination period: 119% for one month, 139% for three months, and 172% for six months, although these should not be interpreted as causal effects of waiting-period design. The insured cohorts differ, and multiple covariates can vary together.

The public report also presents 2015–2024 recovery A/E by calendar year, increasing from 122% in 2015 to 178% in 2024, and mortality A/E declining from 85% to 75%. These are observed comparisons with a fixed reference table, not a forecast of future improvement. An actuarial assumption should consider period, claim settlement processes, population composition, emerging diagnoses and the limited time horizon of the study. The SOA's earlier 2009–2017 termination study provides a comparison series and methodological background rather than an automatically exchangeable exposure base.

Duration, diagnosis and work definition as interacting variables

At short durations a high share of recoveries may follow acute illness and injury; prolonged claims can contain more severe or persistent conditions. Because the pool is dynamically selected, the diagnosis distribution among open claims changes with elapsed time even without changing diagnoses for individual claimants. The 2026 study's diagnosis-level results show substantial differences in A/E: recovery A/E of 121% for cancer, 157% for back conditions and 183% for mental and nervous conditions (with mortality patterns also differing). These are broad study groups and cannot support clinical predictions for individual claimants.

Contractual transition points add another layer: a policy that begins with inability to perform the worker's own occupation and later applies a different criterion can change the probability of payable status without a change in medical impairment. A reserve model may implement a deterministic transition in eligibility criteria combined with probabilistic claim adjudication and employment outcomes. Treating that transition as an ordinary smooth recovery hazard hides the reason payments may cease and can corrupt comparisons among occupational groups with different prevalence of these clauses.

Relapse, rehabilitation, and censoring

A claimant can recover, reenter employment, relapse and file a continuation or new claim. Whether the insurer resumes payments under the previous claim is determined by the applicable recurrence clause and time limits. The distinction between a new incidence event and reopened old claim must be made consistently to avoid double counting future liabilities. Rehabilitation services, partial earnings and workplace accommodation may alter both health-related recovery and contractual benefit amounts. Administrative closures can subsequently be overturned by appeal; if material, expected reinstatements and arrears belong in the relevant cash-flow scenarios.

Right censoring occurs when a claimant remains on claim at the data-extract end; an observed-data cut must not treat that claim as terminated. Left truncation occurs when a study includes claims entering observation after onset; delayed entry must be represented in survival estimation. A recent-claims study may have better early-duration exposure than late-duration exposure, and an old claims extract can overrepresent long survivors. The data origin, study window and life-month counting convention must be disclosed before comparing fitted hazards across occupations.

Reserve Estimation and Validation

Prospective cash-flow formulation

Consider an already incurred claim i at valuation date t. Let I_{i,m} indicate that a benefit will be contractually payable at the end of the mth future month, and let B_{i,m} be the benefit amount, including permitted benefit escalation, partial-disability formulas and offsets. Let X_{i,m} represent claim-specific expenses associated with that future state, and let v_{i,m} denote the applicable cumulative discount factor. A simple claim reserve is

[ V_i(t) = \sum_{m=1}^{M_i} v_{i,m},\mathbb{E}t\left[I{i,m} B_{i,m}+X_{i,m}\right], ]

where M_i is the last possible payable month under the contract and the expectation is conditional on the complete known claim state at t. More complex cash-flow models incorporate conditional benefit amounts, reopened claims, expense timing, lump-sum settlements and recoveries under reinsurance contracts. The gross direct-insurer obligation and reinsurance recoverable must be separately reconciled when the accounting or solvency rules require separate presentation. Under Solvency II Article 77, best-estimate provisions are calculated gross of reinsurance and recoverables are calculated separately.

In an absorbing-payable-state approximation with monthly hazard q_{i,k} and level benefits, the probability of a payment at month m is S_{i,m} = \prod_{k=1}^{m}(1-q_{i,k}), assuming payment is contingent on surviving the entire month and exits occur before the payment. Hence

[ V_i^{\mathrm{benefit}}=\sum_{m=1}^{M_i}v_{i,m} B_{i,m} \prod_{k=1}^{m}(1-q_{i,k}). ]

The chosen timing convention matters. A policy paying at the beginning of each month requires a different survival factor. Recoveries within a payment month may generate prorated benefits and different cash-flow timing. A model of a policy with a guaranteed initial payment period must implement that promise explicitly even if the claimant returns to work earlier. A benefit ending at a contractual age or duration must stop at the specified boundary. Cash-flow models must be checked against the original policy terms as well as against event tables.

Occupation-duration interactions in valuation

A common pragmatic structure begins with baseline cause-specific hazards by elapsed duration and age, then adds occupational, diagnosis, contract and period effects. For example, a proportional cause-specific recovery hazard can be written lambda_R(u|o,x,z)=lambda_{R,0}(u,x) exp(beta_o + gamma_z + delta_{o,u}). Here beta_o captures an occupational main effect, while delta_{o,u} represents any supported occupation-duration interaction. This is a modeling specification drawn from familiar survival methods; the coefficients must be estimated from appropriate observations. Setting delta_{o,u}=0 is a testable simplification, not an observed empirical fact.

A claim-level implementation may be preferable to applying a single group factor, because future payment liabilities depend nonlinearly on hazards. Occupation-class portfolios with the same average termination rate can have different reserves if the distribution of termination rates differs. Mathematically, a portfolio average of individually calculated survival products generally differs from a survival product calculated using the portfolio-average monthly hazard. This is a consequence of nonlinear compounding and compositional heterogeneity, not a new theory.

Reproducible numerical illustration

The following illustration uses chosen probabilities solely to show the direction and approximate magnitude of reserving sensitivity. It is not fitted to SOA/LIMRA experience and does not represent a statutory valuation table. Assume a known claim is payable at the end of each surviving month for a maximum of 24 months, pays $2,000 per full month, carries a monthly effective discount rate of 0.4%, and has constant mutually exclusive monthly recovery and death exit probabilities of 2.0% and 0.1%. Ignore expenses, benefit offsets, appeal and partial payments. With q = 0.021, the expected present value is

[ V=2000\sum_{m=1}^{24}\left(\frac{1-0.021}{1+0.004}\right)^m\approx $35,558.97. ]

Reducing the monthly recovery probability from 2.0% to 1.6%, while retaining 0.1% death probability, increases the calculation to approximately $37,250.15, an increase of about $1,691.18 or 4.76%. This is a sensitivity to a specified absolute probability input, not to the 2026 report's recovery A/E. A 20% reduction of the assumed recovery probability produces a reserve increase smaller than 20% because payments are capped at 24 months, death still occurs, and future cash flows are discounted.

Now suppose two hypothetical occupational cells each pay the same benefit and face the same discount rate and remaining benefit limit. Class A has monthly recovery and death probabilities of 3.0% and 0.1%, producing a case value of approximately $31,742.32. Class B has 1.5% and 0.1%, producing approximately $37,688.86. If 60% of currently disabled claims are in A and 40% in B, with identical monthly benefits and other characteristics, the seriatim-weighted value per current claim is about $34,120.94. Applying the portfolio's average monthly exit probability of 2.5% to every claim instead gives approximately $33,964.83. The difference of $156.11 per claim appears even in this deliberately simple setting because the annuity factor is nonlinear in termination probability. This example isolates a mathematical aggregation effect. In real portfolios, current reserve weighting should also reflect heterogeneous benefits, remaining terms, ages, discounts and exposures.

Hypothetical calculation Monthly recovery Monthly death Present value per claim
Baseline, 24 months 2.0% 0.1% $35,558.97
Recovery sensitivity 1.6% 0.1% $37,250.15
Occupational cell A 3.0% 0.1% $31,742.32
Occupational cell B 1.5% 0.1% $37,688.86
Weighted 60% A / 40% B Class-specific 0.1% $34,120.94
Pooled-hazard approximation 2.4% 0.1% $33,964.83

All entries are calculated as 2000 × sum(((1−q)/(1.004))^m, m=1,...,24) using a fixed 24-month contractual limit. The stated 2.4% pooled recovery probability is the 60/40 weighted average of 3.0% and 1.5%. The illustration holds the starting mix fixed; in an actual portfolio, subsequent surviving-claim composition would move toward the slower-terminating cell. A model that uses an initial mix for every future duration implicitly ignores that selection and can diverge progressively from a seriatim projection.

From A/E ratios to assumptions

Actual-to-expected analysis is a comparison of observed event counts to expected counts under a specified reference table. If observed recoveries equal 150% of the expected count over a study portfolio, it follows that observed events / expected events = 1.50 under the study conventions. It does not follow that every claim's recovery probability is 1.50 times its previous estimate for every duration and occupation. The expected denominator may vary significantly by age, diagnosis, benefit amount, elimination period and disability definition. A/E ratios are also subject to changes in case composition, data quality and common exposure rules. An assumption-setting process should, wherever licensed data permit, reconstruct expected events by the precise risk cells and distinguish relative experience changes from differences in baseline composition.

Using the 2026 SOA/LIMRA study report as a benchmark, an insurer can ask whether its overall or duration-specific actual-to-expected ratios exceed or fall below the published indicators. A defensible adjustment to reserving assumptions requires additional examination of the insurer's product mix, occupational distribution, eligibility definition, adjudication practices, claim size and segment credibility. A reported industry average is evidence about the industry sample. It is not a calibration guarantee for a particular insurer.

Statutory, financial-reporting and economic bases

Valuation bases differ in the treatment of discounting, adverse margins, risk adjustment, boundary, current assumptions and expense allocation. U.S. model reserving material considers health claim reserves and minimum morbidity standards, while particular states and product generations may have distinct operative requirements. An insurer should reconcile reserve calculations to adopted law and approved tables rather than applying a single U.S. source as if it were a universal mandate. The NAIC Health Insurance Reserves Model Regulation also makes clear that more than a table-based calculation can be required when reserve adequacy is in question.

For financial statements governed by IFRS 17, the liability for incurred claims belongs within the standard's fulfillment-cash-flow measurement, including nonfinancial risk adjustment and appropriately reflected discounting. The IFRS 17 standard overview explains current estimates and the recognition architecture. The liability for incurred claims must be differentiated from profit recognition through the contractual service margin for remaining coverage. Solvency II Article 77 technical provisions specifies a best estimate and risk margin; differences in the risk margin's future calibration must follow effective-date and jurisdictional rules. The European supervisory authority's 2026 revised guidelines report reports changes tied to a time-dependent lambda factor that will apply from January 30, 2027, so a valuation dated November 2026 must not casually treat those future requirements as already operative.

The Actuarial Standards Board's ASOP No. 7, Life or Health Cash Flow Analysis, effective June 1, 2026, frames cash-flow analysis for reserve adequacy and risk assessment. It directs attention to liability and asset cash-flow scope, economic conditions, product options, reinsurance, documentation and the purpose of the analysis. Consequently, duration-specific disability obligations that are long relative to the assets supporting them may require consideration of interest-rate and liquidity sensitivities in addition to claim termination stress.

Reinsurance, expenses, and offset dynamics

The projected benefit stream may be altered by coinsurance, excess-risk treaties, claim buyouts, reinstatement settlements or recoveries from public programs. For a gross insured liability, benefits should be projected before deducting amounts recoverable from a reinsurer; the recoverable is then analyzed under the applicable measurement regime, including counterparty risk where relevant. Net economics may differ from gross liability. Discount rate selection and the recognized treatment of ceded amounts are accounting questions as well as actuarial questions.

Monthly benefits can be gross of offsets or defined as a net formula. The onset of Social Security Disability Insurance or workers' compensation benefits may change cash payments; delays in awards may create retroactive recovery of prior overpayments. A monthly reserve projection can separately model the timing of awards, offsets, collections and claim settlement costs. Expected cash flows must reflect only legally and contractually supportable offsets, estimated with available evidence and without assuming universal enforcement of any one country's rules. A model that projects termination accurately but omits a material benefit offset can still overstate reserves; a model that reduces benefits before an offset is likely to become effective can understate them.

Validation targets and reserve attribution

Validation must cover cash-flow totals, not only the fit of termination rates. At each subsequent valuation date, the analyst can compare actual benefit paid, observed claim termination, remaining open claims, reopened claims and updated reserves against prior predictions. Roll-forward attribution should distinguish expected release through passage of time, paid benefits, changes in mortality and recovery assumptions, contract data corrections, discount changes, reserve additions for newly incurred claims, and experience differences. If external rates are adopted, an A/E study should be conducted on both event counts and projected claim cash flows, weighted to the purpose of valuation.

The ASOP No. 5 methodological guidance recommends considering multiple methods, sensitivity to reasonable assumptions, follow-up studies and documentation. A long-duration case-level model can be checked against aggregate development or projection methods for incurred claims, with the understanding that each addresses particular components of the total liability. A result that balances only at aggregate level may conceal offsetting occupational deficiencies. Conversely, fitting every fine cell exactly may capture noise. A useful validation hierarchy combines claims-level mathematical reconciliation, occupation-duration residual checks, segment-level adequacy, and total reserve reconciliation to financial statements.

Integrated Findings and Research Agenda

Findings established by primary evidence

The historical development of U.S. group LTD disabled-life reserve tables demonstrates the importance of cause-specific termination, benefit definition and claims experience. The Academy's 2013 valuation report places these within a formal valuation-table and credibility context. In 2026 the SOA/LIMRA termination study supplies updated publicly verifiable A/E evidence and shows material patterns by duration and broad collar category. These empirical observations are consistent with the need to review duration-conditioned assumptions and occupational heterogeneity rather than rely on a timeless undifferentiated rate.

The study's 150% recovery A/E is a comparison with historical expectations, not a model coefficient to apply universally. Its 81% mortality A/E similarly does not establish a corresponding 19% decrease in death probability for each claim; baseline mortality rates and risk-cell exposure determine that interpretation. The observed 159%, 161% and 135% recovery A/E values for blue-, gray- and white-collar groups establish a population-specific comparison, not a causal ordering of job types. The contract-defined own-occupation transition and elimination periods introduce additional duration-related structure that can coincide with occupational selection.

Original analytical synthesis

Three linked measurement choices determine whether occupational stratification materially improves a disability claim reserve. The first is risk-set comparability: occupation comparisons require a common definition of observed eligible exposure, disability duration and event coding. The second is state congruence: model transitions must correspond to whether benefits remain owed under the applicable contract, including partial, recurrent and appealed states. The third is valuation congruence: expected cash flows, discounting and uncertainty adjustment must correspond to the specific statutory, accounting or solvency reporting objective. These are criteria for coherent application of established actuarial methods, expressed as an analytical synthesis rather than as a new theory or proprietary doctrine.

The illustrative 60/40 occupation calculation proves within its stated assumptions that replacing heterogeneous hazards by a portfolio-mean hazard can understate the weighted seriatim reserve. The numerical result depends on the selected probabilities and benefit horizon; the mathematical nonlinearity is general. Further, as duration increases, conditional survival changes the occupation mix among remaining claims, so a portfolio mixture fixed at disability onset cannot remain an accurate description of surviving claims at every later month. These statements follow from ordinary probability algebra and do not rely on inventing empirical occupation effects.

The reserve sensitivity to recovery assumptions is especially important when the remaining benefit duration is long. For a benefit limited to only a few additional payments, the effect of a hazard change is bounded by that short horizon. For a longer-paying policy, the same proportional change in early exit may affect many more discounted payments. Age, benefit definition, exhaustion limits, time-varying expense and discount curves jointly determine the final magnitude. A defensible process therefore reports both aggregate reserves and decomposition by duration, occupation, policy generation and remaining benefit term.

Uncertainty requiring further empirical investigation

A genuinely portable estimate of the joint occupation-duration hazard surface would require suitably licensed claim-level or sufficiently disaggregated experience from insurers with documented consistent definitions. It would require sensitivity analyses for occupational code missingness, cohort and employer selection, policy transition points, partial disability, offsets, and revival of benefits after appeal or relapse. External validation in different insurance markets would also be necessary before claiming international transferability. The publicly accessible sources reviewed here do not provide those ingredients; the absence of them limits numerical empirical inference, but it does not prevent clear conceptual, mathematical and documentary conclusions.

Further research should quantify how claims selection changes occupational mix over long durations; examine whether apparent collar gradients remain after adjustment for diagnosis and own-occupation transition dates; compare seriatim and pooled-hazard cash-flow error under multiple policy designs; and test occupation-aware calibration on out-of-time insurer data. Additional work should investigate the link between rehabilitation programs, employer accommodations and payment status without assuming that observational association establishes causation. Comparative research must map private policy terms into local labor-market and benefit systems before generalizing the U.S. evidence base.

Implementation implications

For reserving practice, the findings justify a versioned claim-inventory schema, duration-sensitive cause-specific assumptions, documented occupation mappings, credible partial pooling, policy-specific benefit transitions and explicit validation of cash flows. They also justify distinguishing case reserves from IBNR, net cash flows from gross liabilities, and best-estimate from margins. Under the IFRS 17 standard or the Solvency II best-estimate provisions, these distinctions affect the reported liability. Professional standards such as ASOP No. 7 and ASOP No. 5 provide further procedural guidance in their actual jurisdictional scopes.

A research record should capture each source's publication date, accessible source content, evidence claim, denominator, applicable population and access restrictions. The most significant open scientific question is the empirical size and stability of the conditional joint occupation-duration effect after adjustment for diagnosis, selection and policy terms; the present dossier does not represent that parameter as estimated. A future empirical study can answer it through access to additional data, with clear governance and statistical design.

Conclusion

Disability income claim reserves are prospective conditional obligations whose measurement starts from the legally and actuarially correct state of an already incurred claim. Occupation affects the physical and economic conditions under which work capacity and insured eligibility are evaluated. Elapsed disability duration changes the distribution of surviving claims, their remaining benefit horizons and the operative contract rules. Together they determine the structure of future payment probabilities, although their individual effects cannot be identified from aggregate comparisons alone.

Primary source evidence from the September 2026 SOA/LIMRA group LTD termination study confirms that actual-to-expected mortality and recovery differ across duration and broad collar groups. The evidence also indicates the importance of own-occupation contractual transitions and the limited transferability of an experience study from fully insured U.S. group policies. A/E comparisons establish what happened relative to a chosen reference table. Individual reserve assumptions require the baseline rates, matching risk sets, relevant population, and coherent payment-state models.

The prospective reserve formula and reproducible illustration establish the mathematical consequence of heterogeneity: the mean of seriatim occupation-sensitive reserves generally differs from the reserve produced by an average hazard, and a reduction in recovery can increase expected payments without generating a proportionate increase in the reserve. Financial measurement then requires jurisdiction-specific treatment of cash flows, discounting, margins, recoverables and reporting boundaries.

A scientifically credible reserving process therefore combines exact contractual interpretation, duration-conditioned competing risks, suitable occupational information, credible calibration, independent numerical checks, and retrospective cash-flow validation. These requirements give the practical significance of the research: better specified estimands and stronger evidence-to-assumption discipline permit more transparent, comparable and testable disability income claim liabilities.

References