RESEARCH DOSSIER
Annuity Portfolio Mortality Improvement Assumption Governance
Abstract. Mortality improvement assumptions determine the projected duration of insured lifetime-income payments. Their governance requires reconciling observed annuitant mortality with population evidence, selecting age–calendar-year and cohort projections, documenting historical and future assumptions, testing structural uncertainty and maintaining an auditable record of revisions. This monograph analyzes that governance problem in international annuity portfolios, focusing on mortality base and projection choice, assumption approval and revision, stress documentation, valuation effects and the distinct requirements of actuarial professionalism, Solvency II and IFRS 17. It examines the most recent publicly documented mortality-improvement research available through November 14, 2026, including RPEC 2026, CMI2025 and the SOA/LIMRA 2020–2024 individual payout annuity study. A fully specified hypothetical 50-year cash-flow model demonstrates that an annual improvement rate of 1.0%, compared with 0.5%, raises the expected present value of a $10,000 annually payable survival annuity from approximately $149,266 to $152,875, given the stated mortality and discount assumptions. The analysis presents original source-grounded synthesis of the relationship among observational data, the projected longevity curve, model uncertainty, valuation, institutional authority and auditability. No new actuarial theory, named doctrine or proprietary methodology is proposed. Empirical limitations and the distinction between public summaries and restricted technical materials are documented.
Keywords: Annuity; longevity risk; mortality improvement; life-contingent benefit; annuitant selection; cohort projection; mortality base; actuarial assumption governance; actuarial valuation; life expectancy; Solvency II; IFRS 17; mortality experience; MP-2021; CMI2025; model validation; stress testing; audit trail.
Introduction
A lifetime-income insurance contract converts a survival contingency into a continuing financial commitment. If annuitants survive longer than the mortality assumptions imply, a pure life annuity generally pays more benefits. The magnitude of this exposure depends on when payments begin, whether their amounts escalate, which lives are covered, the presence of guaranteed payments or survivor benefits, how the portfolio has been selected and how the projected payments are discounted. Prospective mortality improvement is therefore a component of annuity pricing, technical-provision measurement, financial reporting, capital, reinsurance and asset-liability management.
The subject of this Research Dossier is the institutional governance of prospective mortality-improvement assumptions specifically applied to insured annuity portfolios. It studies the relationship between empirical mortality bases, future improvement rates, responsibility for approval and review, stochastic and deterministic uncertainty, model validation, and the ability of a third party to reconstruct the rationale for an assumption change. This subject is narrower than a complete theory of life insurance mortality, general retirement modeling or asset-liability management. The insurance contract and the selected insured population define its economic object.
Recent evidence illustrates the analytical problem. The Society of Actuaries' RPEC 2026 mortality update reported that U.S. mortality at age 65 and older had broadly returned to the level expected by projecting 2019 population mortality with MP-2021, yet it retained the older scale for retirement-plan use because a new long-term improvement trend could not be supported adequately by the post-pandemic record. The Continuous Mortality Investigation's CMI_2025 projection report incorporated England and Wales population mortality through 2025 and yielded changed cohort life expectancies. A current annual model release, a decision to retain an older scale, and an insurer's own mortality experience may all be relevant while answering different questions.
The study is based on official actuarial research descriptions, accessible regulatory texts, professional practice standards, publisher bibliographic records and openly described empirical studies. Original analysis combines those sources with transparent numerical sensitivity calculations. Subscription-only annuity datasets and restricted CMI technical workbooks are not represented as accessed. The resulting observations concern the proper structure and supporting evidence for an assumption decision, while the illustrative calculation remains a hypothetical demonstration rather than a real insurer's reserve or solvency estimate.
Scientific Foundations, History, and Research Identity
Insured lifetime benefits and survival contingencies
An immediate life annuity typically pays specified installments after purchase as long as the relevant life survives to the contractual payment dates. In a deferred income annuity, payments begin only after the elected deferment, potentially conditional on reaching the commencement age. Joint-life and last-survivor designs depend on two lives and may reduce income after the first death. Period-certain and premium-refund options introduce contractually guaranteed cash flows that may survive the annuitant's death. Guaranteed lifetime withdrawal benefits combine account values, policyholder choices and insurer-backed lifetime benefit conditions. These distinctions determine which future cash flows are exposed to longevity and in what direction.
The first governance task is contractual classification. An analyst must separate payments conditional on survival from death-triggered refunds and payments guaranteed regardless of survival. Applying an improvement scale uniformly to all cash flows can produce an incorrect result even if the scale itself is statistically sound. A benefit payable with certainty for ten years has no survival contingency during those guaranteed installments, whereas the life-only portion depends on remaining alive at each due date. Administrative settlement timing, whether payments are in advance or arrears, and the valuation convention for a death during a payment period also affect expected cash flows.
For a level annual end-of-year benefit B payable while one life aged x survives, let S_x(t) denote the probability of surviving t years under the projected mortality path and let v_t denote the discount factor. The simplified expected present value is
[ V_x=\sum_{t=1}^{T}B,v_t,S_x(t). ]
With mortality probabilities q_{x+j,y+j} at attained age x+j and calendar year y+j,
[ S_x(t)=\prod_{j=0}^{t-1}(1-q_{x+j,y+j}). ]
These equations describe only the life-contingent component under a finite horizon. They do not capture premiums, guaranteed or refund cash flows, reinsurance, expenses, taxes, policyholder options or nonlinear investment guarantees. A serious portfolio analysis establishes the contractual settlement conditions before filling the survival term.
Observed mortality and future improvement
Mortality improvement concerns the change of a mortality measure over time. For discrete annual death probabilities one possible relation is
[ q_{x,y+1}=q_{x,y}(1-m_{x,y}), ]
where m is an age- and year-specific improvement rate. A positive m reduces next year's death probability at that age; a negative m represents deterioration. Another model may project central death rates or mortality forces rather than death probabilities. The measure, reference year and conversion convention must therefore be identified: a percentage improvement in a central death rate is not mathematically identical to the same percentage improvement in a probability of death.
Period mortality describes deaths by attained age in a particular calendar year. Cohort mortality follows a generation as it ages through successive years and hence through different period mortality tables. An insured annuity payable over decades requires the latter trajectory in its expected survival calculation. The OECD's international life-expectancy analysis explicitly distinguishes period and cohort life expectancies and describes substantial international differences. A population life-expectancy estimate is a contextual statistic, not itself an annuity portfolio valuation basis.
An assumption must preserve a useful distinction among observed experience, a mortality base at the valuation date and the projection of future change. The observations consist of recorded deaths and exposure for selected people and times. A base table may be graduated from that experience, adjusted for small samples or borrowed from a relevant external table. An improvement scale transforms the base along future calendar years. Using a base already reflecting mortality changes through 2025 and separately projecting improvements from an earlier table date can double-count part of a trend unless a coherent bridging rule is applied. CMI Working Paper 214 directly emphasizes population and time consistency when combining base and improvement assumptions after the pandemic.
Development of mortality forecasting
Modern modeling extends historical life tables through statistical projection, first by deriving age-specific patterns and then by estimating how those patterns evolve. The influential Lee–Carter analysis represents logarithmic mortality as an age profile combined with a changing time index. Poisson log-bilinear mortality regression, described by Brouhns, Denuit and Vermunt, uses deaths and exposures explicitly in a count-data framework. The Cairns–Blake–Dowd two-factor model focuses on older-age mortality dynamics, longevity risk and parameter uncertainty. Their different structures remind the reader that model selection is a matter of population, purpose and available evidence.
Practical reserving also uses published reference tables. The Society of Actuaries' 2012 individual annuity table was prepared for a defined U.S. reserving application with projection scales and margins. The MP-2021 mortality scale reflects historical U.S. population experience through 2019. Each has a source population, derivation purpose and valuation context. A table intended for a regulatory minimum or pension measurement cannot automatically be equated with a best-estimate mortality trend for all individual payout annuity products.
An assumption's statistical validity and its actuarial applicability must be evaluated separately. One can correctly fit a national population time series and nevertheless predict poorly for an affluent, self-selected group of annuity purchasers. Conversely, an insurer may know its mortality level well but possess too little period exposure to estimate the improvement slope by cohort or sex. That combination creates a legitimate role for external trend evidence, coupled with documented portfolio adaptation and uncertainty rather than simple substitution.
Object, method and evidentiary scope
The unit of research is an approved assumption version for a particular insured portfolio and valuation date. It identifies the base mortality source and effective date, target population, projection parameters, long-run extrapolation, model edition, data cutoff, benefit design, purpose of measurement, governance owner, validation record and revision history. Statistical analysis should distinguish source validity, internal consistency, out-of-sample forecast performance and sensitivity of insurance liabilities. Regulatory analysis should identify jurisdiction and applicability instead of treating any one jurisdiction's rules as a global code.
This monograph reviews selected international sources across three types: primary legal or professional requirements governing documentation and valuation, institutional experience or model reports, and peer-reviewed statistical research. It conducts a deterministic valuation experiment to demonstrate sensitivity. The method is a documentary and analytical synthesis using established mathematics; no new theory is named, and its claims of originality are confined to the integrated comparison, interpretation and checkable example.
Mortality Base and Projection Choice
Building an annuitant experience basis
An insurer's observed death rate depends on accurately defining persons at risk. The data should identify date of birth, mortality observation dates, annuity issue and commencement dates, time exposed, death date or survival cutoff, sex where applicable, income amount, benefit form, contract status, reinsurance treatment, and linked lives under survivor benefits. Late death notification or inconsistent policy records can distort both the numerator of a death rate and the denominator of exposed person-time. A death count with a shifting exposure definition cannot be used to infer a reliable mortality improvement slope.
Count-weighted and income-weighted analyses have different economic purposes. A large contract has the same single-life death event as a small contract in a life count, but a different impact on expected benefit payments. An annuity liability calculated by income exposure may be more sensitive to the mortality of wealthy or high-payout lives than a count-weighted rate suggests. A source-data record must identify the weighting and portfolio segmentation in every observed-to-expected ratio. Selection of an external benchmark should reflect this weighting where the decision concerns the insurer's monetary liability.
The 2026 SOA/LIMRA individual payout annuity experience study describes about 3.1 million contract-years of exposure, $33 billion of annual-income-years exposure, and 143,190 deaths over 2020–2024 from 26 participating companies in 23 groups. The previous 2014–2019 experience study included approximately 4.3 million contract-years and more than 236,000 deaths. These are important empirical resources; their publicly reported totals alone do not imply that the same exact group of annuitants was observed across both periods, nor do they identify a unique industrywide future improvement rate. Changes in contributors, ages, benefit sizes, product mix and observation period require normalization.
Selection and population basis risk
The population purchasing lifetime income differs from the general population for economic and behavioral reasons. Financial capacity, retirement timing, private expectations of longevity, health, optional forms of pension transfer, distribution channel, product price and guaranteed features influence entry. Annuity selection may continue to affect observed mortality after purchase. Some contracts explicitly serve impaired lives, while bulk pension transfers may insure cohorts originating in occupational pension schemes. A mortality source's relevance therefore depends on how its covered population was selected.
A portfolio can have adequate evidence for a mortality level adjustment and inadequate evidence to estimate a long-term improvement trend. Credibility analysis should use actual exposure and death counts, sampling variability, relevant subgroups and appropriate external comparators. A low death count at the oldest ages creates substantial uncertainty precisely where long-lived guaranteed payments can remain material. Borrowing an external improvement scale may be a reasonable response when its relevance is explained, but the weight assigned to an external scale must be treated as judgment and reviewed.
The SOA's individual-life AG38/VM20 scale illustrates a domain distinction. A mortality improvement recommendation developed for individual life valuation has a different benefit direction and source population from payout annuities. It should not be relabeled as an annuity improvement standard. The CMI's mortality working-paper catalog separately includes pensioner and bulk-annuity experience studies alongside general population projections, reinforcing the need for source-population specificity.
Age, calendar year, and cohort
For an annuitant aged 65 in the valuation year, the model needs mortality at ages 65, 66, 67 and onward in successive future calendar years. Repeating one period table at all future payment dates assumes no prospective change. A projected mortality surface represents q as a function of age and calendar year, and it can reflect different improvement rates for different cohorts. A one-dimensional improvement average conceals changes at the ages where a portfolio's benefit cash flows are most sensitive.
Projection systems require precise versioning. CMI_2025 projection model uses England and Wales data through December 2025, along with user-specified projection assumptions. An appropriate governance record retains the exact release, calibration period, smoothing treatment, short-term adjustment, long-run improvement target, convergence and high-age rules. It is meaningless to approve “the CMI assumption” without those parameters. Similarly, MP-2021 is a named mortality scale whose version and target data must remain visible in the approval file.
The CMI March 2026 announcement reported that the core CMI_2025 model produced cohort life expectancy at age 65 about eight weeks higher for men and six weeks higher for women than CMI_2024, keeping other parameters comparable. This illustrates both sensitivity to newly observed data and dependence on modeling conventions. The exact change in an annuity insurer's liability depends on its annuitant mix, contract schedule and valuation assumptions; it is not the eight-week population result multiplied by annual benefit.
Pandemic disruption and the improvement curve
COVID-19 interrupted long-term mortality patterns with temporary and potentially persistent effects. An elevated observed base can interact with a prospective reversion assumption, while a pre-pandemic base can omit new information about current mortality. The central problem is to represent the initial level and subsequent improvement consistently, rather than to classify one period of data as automatically usable or unusable. The CMI's base-table interaction paper explicitly warns of over- or under-stating a pandemic effect through inconsistent inputs.
The RPEC October 2026 report places the July 2025–June 2026 age-65-plus actual-to-expected mortality ratio at approximately 98.7% for both sexes combined when compared with 2019 mortality projected using MP-2021. It describes the longer-run improvement trend as inconclusive and explains why MP-2021 remains a reasonable retirement-plan reference, particularly at older ages. These are U.S. general-population findings. The ratio is not a direct estimate of mortality or improvement in a particular insurer's block.
The SOA's MIM-2026 explanation reports retrospective data updates through 2023 while retaining MP-2021 prospectively and excluding post-2019 experience from its prospective trend fitting. The distinction between refreshed observations and unchanged future assumptions is analytically important. The CMI_2026 consultation was published in October 2026, with an intended new release in March 2027; as of the research cutoff, CMI_2026 remains a proposal rather than an adopted scale. The assumption register must preserve these dated distinctions.
Long horizons and extreme ages
Lifetime liabilities can extend beyond the observable mortality history by several decades. Recent improvement rates may converge toward longer-run assumptions, vary across cohorts or cease to be informative at very old ages. Extrapolation must keep death probabilities admissible and reflect uncertainty arising from sparse high-age exposures, age misstatement, registration differences and cohort survival selection. An apparently precise model estimate can conceal substantial sensitivity to terminal improvement or the oldest-attained-age cutoff.
The OECD's 2025 longevity analysis indicates the extent of differences in life expectancy across countries and discusses slower old-age gains in recent periods. National averages provide useful contextual evidence but cannot settle the projection for a selected annuitant group. A defensible governance decision reports how different regional and socioeconomic cohorts are reflected in the insurer's source evidence, or explicitly states why available evidence does not support an additional differential.
Assumption Approval and Revisions
Governance as a chain of accountable decisions
Selection of a mortality improvement assumption has three logically distinct phases: formation of a scientific estimate, authorization of its use for a stated institutional purpose, and periodic reconsideration in light of observed results. The actuarial estimate concerns future mortality conditional on the data and methodology selected. Its authorization concerns whether that estimate is appropriate for a particular portfolio, valuation objective, and effective date. Its revision concerns what newly available evidence changes about that previously documented judgment. These phases require traceable records because a technically correct projection can still be used in the wrong insurance portfolio or in the wrong regulatory measurement.
The analyst proposing a basis should identify the previous approved parameters; all new or superseded evidence; the target covered lives, product forms and cash-flow periods; alternatives considered; effects of changing the mortality base, the improvement surface and any margins; and the reason for the recommended choice. Material assumptions need an accountable owner who can explain why the chosen table and projection are consistent. Independent model-validation, risk, accounting and control functions can challenge the assessment according to the insurer's organizational arrangements, and an authorized decision-maker records approval. The practical division of work varies by insurer and jurisdiction; the essential informational requirement is that evidence, judgment, model implementation and authorization are separately recoverable.
The professional ASOP 23 data-quality standard describes responsibilities for assessing the appropriateness and limitations of data and reliance. ASOP 56 on modeling addresses model purpose, model risk, assumptions, validation and documentation. The requirements of ASOP 41 on actuarial communications support reporting of scope, methods, results and their limitations. These are U.S. actuarial professional standards with defined scopes, rather than a universal corporate-approval statute. Nevertheless, their disciplined separation of data assessment, modeling and communication explains why a single undocumented rate pasted into a valuation system cannot constitute a complete governance record.
The current ASOP 27 pension standard addresses the selection, consistency, review and disclosure of pension assumptions, including mortality improvement. It took effect on January 1, 2025, replacing the separate demographic-assumption standard ASOP 35, which was formally repealed for work performed on or after that date. This is a pension-specific professional comparison rather than a direct annuity-insurer reserving rule. Its careful differentiation of base mortality, historical improvement and future improvement remains analytically relevant without changing the standard's scope.
Selection and formal approval
A responsible assumption memorandum opens with the precise valuation question. For example: “Expected survival probabilities at each future payment date for the closed portfolio of single-premium payout annuities in force on December 31, 2026, using contract-specific payment terms and the insurer's defined central estimate of future experience.” This description identifies the relevant population and objective more precisely than a title such as “Mortality improvement 2026.” The record then identifies whether the estimate will be used for statutory technical provisions, financial reporting, pricing, economic risk, or another purpose, since a prudential margin or prescribed basis may enter differently for each.
A decision package should include the mortality data inventory, base-table construction and reference date, the future improvement curve or surface, selection effects, graduation assumptions, extrapolation limits, parameter fitting, excluded or downweighted years, controls for unusual mortality events, fit diagnostics and uncertainty. It must specify whether different subgroups receive different assumptions and why. A full internal approval need not expose confidential policy-level data publicly; the governing institution can keep protected records while retaining complete independent audit access.
The selection process should explicitly address competing evidence. A general population model can show apparent mortality improvement while a specific insurer's count-weighted annuity block shows near-term mortality deterioration. The difference may reflect composition, observation lag, age distribution or statistical noise. Rather than selecting the series that produces the desired reported liability, the analysis should reconcile relevant denominators, definitions and temporal windows, then identify which source can genuinely predict the selected future cash flows. Where the insurer's history is insufficient to estimate improvement, the record should explain the external benchmark and any downward or upward adjustment rather than invoking an unsupported claim of portfolio precision.
Version control, criteria for review, and materiality
A stable version identifier should point to the exact base table, improvement surface, software model, data snapshot and technical assumptions used in a signed valuation. The policy records model and data version separately; replacing a missing death record should not silently change the previously approved future improvement assumption. An input revision, model change and prospective assumption change are three different events that can all move the liability. The approval file should maintain an effective date and the date of the underlying evidence, which may precede it.
A regular review calendar is necessary because the mortality path extends for many decades while new observations emerge continuously. Reviews can be annual or more frequent where applicable rules or portfolio materiality require it. Event-driven review is equally important. Triggers may include a new annuitant experience study, persistent actual-to-expected deviations, revision of national population data, a substantial shift in portfolio demographics, a change in contract guarantees or distribution, a new reinsurance transaction, significant pandemic or health developments, evidence of model implementation error, or a proposed change in supervisory requirements.
The strongest review process recognizes “retain the prior assumption” as a possible reasoned decision. The 2026 RPEC judgment to retain MP-2021 for retirement-plan use shows that data updates do not necessarily produce stronger support for an entirely new long-term trend. The underlying rationale—considering emerging experience, uncertainty and forecast stability—can inform institutional governance. It does not create an automatic entitlement to retain MP-2021 for an insured payout annuity portfolio. Conversely, immediately substituting the newest CMI or other population scale for every portfolio may introduce changes unsupported by its insured-life characteristics.
The most recent survey evidence also demonstrates institutional variation. The Society of Actuaries' 2025 Mortality Improvement Survey Report, released in October 2026, describes differing practices among U.S. and Canadian insurer respondents, including variation by product, function, attained age and projection horizon, and no general consensus on the post-pandemic path. Survey responses document practice diversity rather than proving that the most common assumption is actuarially correct. This distinction matters when a selection committee considers industry practice as one source among many.
Distinguishing a revision of mortality from a change in liabilities
A mortality assumption affects future expected cash flows without automatically changing the holder's contractual income entitlement. If the issuer promises a fixed amount each year while a covered life survives, a newly approved longevity projection affects the expected number of payments and the measured liability, not the amount promised by the original contract. The economic effect can instead enter new-business pricing, risk appetite, asset duration or reinsurance negotiations, subject to applicable contracts and laws. Joint-life, escalating, deferred and guaranteed structures can respond in distinct ways to the same assumed trend.
A valuation movement should be decomposed into identifiable sources. One useful bridge starts from the previously approved liability calculated on the earlier inputs, then measures changes attributable to actual portfolio run-off, new mortality observations, base mortality, prospective improvements, financial assumptions, benefit and expense assumptions, and the calculation engine. Where the change order is material because of interactions, the bridge should disclose ordering or use a stated allocation convention. Presenting a single net change labeled “mortality” when interest rates and contract composition also moved is analytically incomplete.
For EU insurers, Solvency II's Article 77 technical-provision rule relates best-estimate liabilities to probability-weighted future cash flows using up-to-date credible information and realistic assumptions. A separate risk margin has its own prescribed measurement. The insurer's actuarial function under Article 48 has responsibilities concerning appropriate methods and assumptions, data quality, experience comparisons and reporting to management. These obligations support systematic mortality assumption assessment within the directive's relevant application.
Financial reporting has a different measurement objective. The IFRS Foundation's IFRS 17 insurance-contracts overview describes current valuation of insurance contracts with estimates of cash flows and distinct reporting of insurance service and financial effects. Demographic revisions can change fulfillment cash flows, but the detailed accounting allocation depends on contract groups, coverage terms, the standard's measurement approach and the nature of the assumptions. Neither a universal reserve-release rule nor a uniform income statement treatment can be deduced from mortality improvement alone. The full licensed standard and relevant implementation rules are required for detailed accounting conclusions.
Corporate roles, communication and professional judgment
A mortality improvement assumption is used across several functions. Actuarial teams analyze observations and estimate survival; investment and asset-liability management teams study the timing of payments; enterprise risk teams evaluate systematic longevity shocks; finance reconciles accounting results; policy administration resolves deaths, survivor status and payment eligibility; legal staff assess the governing contract; and internal audit tests controls and delegated authority. This creates a chain of dependencies that must be documented without confusing the responsibility to model with the authority to approve.
The international IAIS insurance principles discuss governance, risk management and internal control as elements of insurer supervision. The IAA actuarial standards index identifies internationally framed practice standards intended for use through applicable adoption arrangements. International harmonization of principles does not make each document universally binding in every local jurisdiction. An insurer's approval memorandum should therefore separate professional standards, binding prudential rules, accounting standards, internal policy and optional analytical good practice.
The final governance decision should be intelligible to readers who are not technical mortality modelers. It can state the chosen projected longevity basis, its most important evidence, anticipated effect on liability, principal uncertainties, sensitivity to plausible alternative paths, and date of reassessment. A technical appendix retains calculations, data definitions, model versions and judgmental overrides. The separation of an executive decision record from reproducible technical support strengthens both comprehension and later challenge.
Stress Documentation and Auditability
The analytical object of stress
A mortality improvement stress is a deliberately specified alteration of the future mortality path intended to reveal the financial exposure to longevity and model uncertainty. Its definition requires the affected population, valuation date, projection years, age ranges, mortality measure, size and shape of the disturbance, any dependence between lives, discount and benefit assumptions, and intended interpretation. The stress should be reproducible from a documented input file, rather than known only by an informal name such as “severe longevity.”
Mortality level stress and improvement trend stress must be separated. A 10% proportional reduction in death probabilities at all modeled ages changes the starting level and all subsequent probabilities. Increasing annual mortality improvement from 0.5% to 1.0% compounds over time. The two scenarios differ at the valuation date, in their timing of effects and in their influence on future payment years. A temporary elevated death-rate scenario representing a mortality shock is different again, because deaths occurring early remove potential later annuity payments even if rates subsequently return to the baseline.
Useful deterministic stress families include: immediate permanent reductions in death probabilities; accelerated improvement concentrated at ages relevant to late-life annuity payments; slower or faster long-term convergence; cohort-specific improvements; a temporary deterioration followed by partial or full recovery; a sustained adverse trend after a public-health shock; and a late-age tail scenario. Joint-life annuities can require further dependence assumptions, while deferred annuities require separate assessment of mortality before and after commencement. The mathematical interpretation is product-dependent: death benefits may move in the opposite financial direction from survival benefits.
Scenario names have limited meaning without parameter trajectories. A stress record must identify whether the change applies to central death rates, probabilities or mortality intensities; whether it multiplies the base or the projected rate; and whether it persists until the modeled terminal age. Applying a rate below zero or above one as a death probability is a calculation error rather than a plausible mortality scenario. Each stress should also be connected to a specific governance question: reserve sensitivity, capital adequacy, benefit-duration tail, emerging-experience exposure or model-form challenge.
Idiosyncratic and systematic mortality risk
A large insurer pooling many independent annuitant deaths gains statistical stability relative to a portfolio of very few lives. The random event of whether one person dies in a particular year becomes a smaller share of the entire pool's payments as the number of sufficiently independent lives grows. A change in the underlying improvement trend, however, influences many lives at once. Systematic longevity risk therefore remains even in a very large annuity portfolio. The demographic mix, benefit concentration and correlated exposures can intensify this risk.
Stochastic mortality approaches distinguish random variation conditional on a given hazard path from uncertainty about the hazard path itself and from the estimation of model parameters. The Cairns–Blake–Dowd study develops an older-age mortality model in which uncertainty and multiple factors matter. The Lee–Carter forecast approach and Brouhns–Denuit–Vermunt Poisson model provide further established statistical structures. An insurer may use a validated stochastic approach for scenario analysis, but it must distinguish uncertainty about future mortality trends from the independent randomness of individual deaths.
A mortality forecast can perform adequately at observed ages and yet have unstable extreme-age or long-run behavior. Calibration errors may come from inappropriate mortality smoothing, cohort extrapolation, limited high-age exposure, a changing population or misclassification of death dates. Models that fit the same five-year period can diverge materially over a forty-year annuity horizon. Governance should disclose what backtests can establish and what remains judgment about long-term demographic development.
Regulatory validation, documentation and audit evidence
The EU delegated regulation's Article 264 validation requirement calls for technical-provision validation at least annually and when indicators suggest that methods, assumptions, data or the resulting liability level are inappropriate. Its scope includes comparison with experience and assessment of data and approximations. The associated Article 19 data requirements require an established process for complete, accurate and appropriate technical-provision data. These obligations create a direct link between mortality source records and the validity of actuarial measurement for regulated insurers.
A particularly concrete legal basis is Article 265 on documentation. It addresses collection and quality of data, the choice of valuation assumptions, selection and application of methods, and validation. It also requires directories of relevant assumptions, justification of selection, inputs, objectives, limitations, processes to review assumptions, and justification and estimated impact of material changes from one period to another. The rule identifies documentation as a substantive part of technical-provision governance rather than a supplementary clerical practice.
A technically reproducible mortality improvement model therefore requires an inventory of policy and person records; data source and extraction date; mortality and exposure definitions; reconciliation of deaths and benefit records; exact reference table and calibration; projection algorithm and all assumptions; code or validated model version; test outputs; sensitivities and stress definitions; reviews and approvals; and a traceable calculation from assumptions to liability. Each item can have a stable identifier or cryptographic checksum. A checksum establishes identity of bytes, not their scientific correctness: the documentation must also show that input population, equations and application are appropriate.
Internal audit should be able to reconstruct a selected valuation result at its original effective date. A later correction to an erroneous birth date must preserve the historical snapshot and document the correction and its effect. An updated national mortality table must be distinguished from a revised projection methodology. Restricted insurer experience can be retained inside controlled systems while allowing authorized independent verification. Public source summaries alone cannot give an outside researcher the same level of access as an institution holding its insured-life records.
Sensitivity, disclosure and legal-time consistency
The materiality of improvement uncertainty depends on the present value and benefit duration of the covered portfolio. A change that appears small as an annual mortality percentage can generate a larger liability effect through many future survival-contingent payments. Conversely, an improvement assumption may have little effect on fixed guaranteed installments that will be paid irrespective of survival. Risk communication should report dollar changes, proportional liability changes and the benefit periods contributing most to the sensitivity, as well as changes under competing financial assumptions.
The reporting context also matters. Solvency II Article 296 establishes disclosure requirements about the valuation of technical provisions, including main assumptions and relevant valuation differences. An insurer's public report can be less detailed than its confidential validation archive, but the internal assumptions must remain defensible and traceable. Under ASOP 41, the actuarial communication should match its intended users and disclose material assumptions and limitations.
A dated governance record must distinguish present law from announced future changes. EIOPA's July 2026 revised guidelines report states that revisions become applicable on January 30, 2027, with changes including risk-margin calculations. An insurer preparing a November 2026 valuation may analyze these prospective changes, but cannot label them current legal requirements on that date. The temporal identity of assumptions should therefore include the legal and model versions used, as well as the scientific evidence cutoff.
The role of institutional challenge
Independent review adds scientific value when it tests the assumptions that most determine the result. A reviewer can reconstruct the data weights, test whether changing the calendar window alters fitted improvement, verify the connection between the mortality metric and contract survival formula, and compare the model's fitted historical predictions with excluded years. It should assess whether high-age extrapolation produces plausible survival and whether material subgroup differences have been obscured by aggregation. Merely verifying that software executes without error does not answer these questions.
A challenge can also be directional. Lower death probabilities should ordinarily increase expected pure life-annuity payments, all else equal. If a model reports the reverse, the investigation should examine whether it includes a death-benefit component, policyholder withdrawals, future premiums, non-linear guarantees, a formula error or inconsistent discount changes. A transparent sensitivity grid is useful because it makes the expected direction and materiality of each change visible before a large production valuation is approved.
Evidence Validation and Model Sensitivity
A controlled, reproducible valuation experiment
A transparent numerical example shows why the governance of a mortality improvement assumption cannot stop at choosing a percentage. Consider a single life exactly age 65 at a hypothetical valuation date, with a benefit of $10,000 payable at the end of each year if that life survives the year. To keep the experiment finite and independently reproducible, the model considers at most 50 payments, ending at age 115. There are no expenses, premiums, refunds, period-certain benefits, taxes, options, spouse benefits or reinsurance; all payments are nominal, and the annual effective discount rate is constant at 3%. These conditions define an illustrative life-only annuity model, not a representation of a market product or insurer-specific regulatory basis.
The baseline attained-age annual death probability is defined by the synthetic Gompertz-like expression
[ q^{(0)}_x=\min{0.5,;0.0115\times1.085^{(x-65)}}. ]
The first-year death probability at age 65 is consequently 0.0115, and baseline mortality increases geometrically with attained age. To apply a constant prospective annual improvement rate m, let j denote the number of years after the valuation year, with j = 0 in the first year. The death probability governing the year beginning at age 65+j is
[ q_{65+j,j}=q^{(0)}_{65+j}(1-m)^j,\qquad j=0,\ldots,49. ]
This expression makes the effect of improvement compound with elapsed calendar time while attained-age mortality still increases because of the age gradient. It deliberately holds all other inputs constant. The sequence of future survival probabilities is generated recursively from S(0)=1:
[ S(t)=S(t-1)\left[1-q_{65+t-1,t-1}\right]. ]
The end-of-year annuity present value is
[ V(m,i)=10{,}000\sum_{t=1}^{50}\frac{S(t)}{(1+i)^t}. ]
There is no rounding within this sum, and the table below rounds only the final result. All assumptions are declared so that a reader can reproduce each payment-year contribution from the equations alone.
| Annual improvement m | Expected present value | Change versus zero improvement | Survival probability at 30 years |
|---|---|---|---|
| 0.0% | $145,864.15 | 0.00% | 0.227414 |
| 0.5% | $149,266.03 | 2.33% | 0.263469 |
| 1.0% | $152,875.12 | 4.81% | 0.300209 |
| 1.5% | $156,700.07 | 7.43% | 0.337135 |
Under these assumptions, increasing the flat annual improvement from 0.5% to 1.0% raises the value by $3,609.09 per contract. A completely hypothetical homogeneous block of 20,000 otherwise identical lives would have a corresponding aggregate gross expected present-value increase of about $72.18 million before all contract, expense, diversification, reinsurance and regulatory adjustments that a real valuation would require. The illustrative block is simply an arithmetic multiplication of the single-life result. It is not a quantified claim about any actual insurer's portfolio or solvency position.
The effect is nonlinear across a multiyear horizon. For the 30-year survival probability, the synthetic path produces approximately 26.35% at a 0.5% annual improvement and 30.02% at a 1.0% improvement. Later survival is built from the product of every earlier mortality probability. Therefore an improvement at one early date affects the likelihood of payment at that date and all later dates. A single change in the improvement parameter alters a long sequence of expected benefits even though the amount of each individual benefit remains unchanged.
Level and trend are different valuation drivers
A mortality level adjustment and an improvement adjustment change the path differently. Keep improvement at 1.0% and multiply every baseline attained-age death probability by 0.90 before applying the prospective improvement factor. The resulting present value is $158,109.46. Instead multiplying the baseline by 1.10 produces $148,101.45. These stress paths modify mortality beginning with the first modeled year; increasing future improvement from 0.5% to 1.0% leaves first-year mortality unchanged and changes later years through compounding. An assumption-change memorandum should therefore label the changed element rather than reporting both as the same “longevity assumption” adjustment.
At a fixed 1.0% improvement rate, a 2% discount rate produces a present value of $172,604.70, while a 4% discount rate gives $136,574.67. These values demonstrate that changes to financial assumptions can produce movements that exceed a plausible mortality-trend variation in the chosen example. If an insurer changes a discount curve and mortality improvement in the same valuation, the bridge should hold one fixed while testing the other, and separately disclose any interactions. A reserve movement is a joint result of contract cash flows, mortality, financial assumptions, expenses and regulatory measurement rules.
A short mortality shock produces a third pattern. Multiplying the mortality probabilities in only the first two modeled years by 1.25, with the original 1.0% improvement path thereafter, gives $151,983.79, compared with $152,875.12 without the shock. The higher early deaths lower later survival probabilities even after mortality rates return to their previous path. This construction intentionally omits rebound, cohort-selection feedback and changes to future cause-specific mortality; adding these mechanisms would require new evidence and explicit model assumptions.
Reproduction and independent checks
The model can be reproduced by a short deterministic loop. Initialize survival probability at one and the liability at zero. For each payment year t from 1 through 50, set j=t−1, evaluate the baseline death probability at 65+j, apply the factor (1−m)^j and any declared level or event stress, update survival by 1−q, and add the discounted $10,000 payment weighted by survival. A second implementation can first generate all fifty survival factors and then sum expected discounted payments. Agreement between two computational approaches to displayed precision is a meaningful arithmetic check.
Model validation should check that all annual q values lie within [0,1], that survival is nonincreasing with payment year, that an increase in the favorable improvement m produces lower annual death probabilities after the initial year, and that a life-only annuity's value increases when the same death probabilities are reduced, all other assumptions held fixed. Tests of this kind are logical implications of the stated equations. They would fail to validate a production actuarial system unless that system also handles exact policy payment schedules, monthly or continuous mortality conventions, leaps and partial years, multiple insured lives, death notifications, recovery after overpayment, contract benefits, expenses, asset-related cash flows and relevant regulatory requirements.
The numerical example has a narrow purpose: it connects an improvement parameter to survival and to liability in a transparent way. Its artificially chosen mortality level and age slope are not calibrated to contemporary population or insurer data. It has no statistical confidence interval for the parameters because no data have been used to estimate them. Applying the results directly to an actual portfolio would be unjustified. The mathematical mechanism, however, is fully reproducible and demonstrates the necessary connection between a governance decision and measurable cash-flow sensitivity.
Actual-to-expected validation
In a real insurer's retrospective test, the numerator of an A/E ratio must be a documented observed death count or benefit amount, and the denominator must be the expected value derived from the appropriate mortality assumption over the same exposure window and population. If the expected count is computed using a table calibrated to a different age distribution, contracts or years, aggregate A/E can be misleading. The record should include attained-age, year, product, benefit amount and other relevant strata; explicit definitions of censoring; handling of missing or late-reported deaths; and uncertainty from small counts.
A historic out-of-sample test must preserve the information available at the time of the original forecast. Suppose a model was approved in 2022 and subsequently predicts outcomes through 2025. Testing that forecast requires freezing its 2022 parameters and its original population definition, then comparing with later results. Recalibrating the historical prediction with data first observed in 2025 is a revised-model comparison, not a true 2022 forecast evaluation. Both exercises have value, but they answer different questions. Record a model's pre-specified evaluation dates, forecast windows, and intervention or censoring decisions.
Observed death-count ratios are more informative about level calibration than about long-term improvement when the observation window is short. A persistent A/E ratio above 100% might reflect an incorrect base, an incorrect temporal trend, selection or exposure recording. Slope tests require multiple periods with comparable populations, and even then selection changes and systematic shocks can confound trends. Confidence intervals and uncertainty should accompany A/E results where the data support calculation, especially at the ages responsible for long-tail payments.
The public SOA/LIMRA 2020–2024 payout experience illustrates the difference between a large study headline and its granular evidence. The summary establishes the extensive combined participant experience and its exposure and death totals. It does not allow an independent author to reconstruct insurer-by-age improvement rates without the detailed data package. Similarly, international population sources such as the WHO Mortality Database supply broad demographic comparisons subject to country-year reporting and data-method limitations; population and annuitant populations remain distinct.
Source revisions as model-governance events
A scientific source is not completely identified by the title of its report. Model software, numerical workbooks, errata and working paper revisions may alter detailed results without changing the principal publication name. The CMI mortality-projections news record states that version 02 of Working Paper 211 corrected calculations in Tables 5.5, 5.6, 5.9, 5.10, 5.15 and 5.16 because the earlier tables described future-improvement sensitivity but inadvertently varied only a specified historic improvement interval. The correction matters directly to the interpretation of prospective mortality stresses.
The published CMI_2025 working-paper page remains an appropriate identifying source for the released model. An institution relying on numerical tables from the accompanying working paper, however, must retain the exact document version and determine whether the error affects its own calculations or communications. Model governance extends to third-party source corrections: they should generate a documented assessment, rather than an automatic assumption change or an ignored correction.
The distinction is equally relevant to administrative mortality data. A statistical agency can revise death-registration completeness or population denominators, changing A/E results for years previously reported. The RPEC 2026 update explains its data sources, completeness adjustments and limitations. A future forecast assessment must retain the data vintage it used. Without this provenance, a numerical discrepancy between two reports may be attributed to mortality change when it actually comes from the revision of historical exposure.
Methodological boundaries of sensitivity results
A deterministic sensitivity table isolates the effects of chosen changes, but does not assign probabilities to those changes. A 1.5% annual improvement case is a scenario, not automatically a statistical upper confidence bound, insurer capital requirement or extreme longevity quantile. Probabilistic outputs require assumptions about the distribution of mortality-trend uncertainty and an appropriate validated model. Alternative statistical structures, parameter uncertainty and incomplete portfolio evidence should be described separately.
For policies with death benefits, guaranteed terms, periodic income increases or policyholder options, the relation between mortality and total contract value can be more complex than the pure survival annuity. A higher survival probability may also affect expenses, premium inflows before retirement, guaranteed option exercise and investment exposure. Consequently, a model reviewer should examine benefit components and interactions before generalizing the monotonicity illustrated by this paper. This distinction is particularly important when the same improvement assumption is used across different annuity product designs.
Integrated Research Findings and Open Scientific Questions
The identity of the assumption is larger than a published scale
The first integrated finding is that a mortality improvement assumption is a dated proposition about an identified insured population and its future survival-contingent cash flows. Its identity comprises the source data, mortality base, future improvement model, time conventions, selection adjustments, contract design, discount and benefit assumptions, intended valuation purpose and approved version. Two actuaries may both reference MP-2021 or CMI_2025 while adopting different defensible bases because their covered populations and projection objectives differ. Conversely, citing the same named model does not establish that two valuations are directly comparable.
The second finding is that external population models and internal annuitant experience tend to provide different kinds of evidence. An insurer may have credible current mortality levels but insufficient historical observation to determine a separate improvement trend. A broad population study has greater experience for systematic temporal movements but does not represent the same selection characteristics. The relationship is therefore one of calibrated transfer and documented uncertainty. The SOA/LIMRA individual payout annuity study informs annuitant experience; the RPEC population update informs U.S. mortality trends; the CMI_2025 projection model informs England and Wales cohort forecasting with user-defined modeling choices. Their different coverage is a scientific property to be preserved, not an inconvenience to be averaged away.
The third finding is the importance of base–projection coherence. A current mortality base can include some of the same historical changes that a projection method intends to explain. A temporary shock can change measured levels without providing sufficient evidence to change the long-run slope. The CMI's base-and-improvement discussion addresses these problems directly. Approval must test the implied cohort trajectory as a whole, rather than approving the source table and the future improvement percentage as isolated administrative entries.
Scientific interpretation of post-pandemic evidence
The fourth finding is that a new observation, a new model version and a new prospective assumption are separate events. In 2026, RPEC retained MP-2021 based on its assessment of available post-pandemic evidence, while MIM-2026 updated retrospective information without replacing that projection scale. CMI_2025 updated population calibration and published a new released projection edition; CMI_2026 was still a consultation at the November research cutoff. A scientifically governed portfolio may reasonably respond differently to each event. The owner should record the source's release status, version, observation window and the specific judgment about relevance.
The October 2026 SOA mortality improvement survey offers an additional observation: insurers reported different approaches by product and business function rather than a single settled response to pandemic experience. A practice survey is not itself an optimum-selection study and does not identify which method will forecast future annuitant survival most accurately. It does, however, show why governance must articulate a portfolio-specific rationale rather than invoking a presumed universal industry default.
A fifth finding concerns scientific corrections. The CMI correction to Working Paper 211's selected sensitivity tables demonstrates that even an established institutional model source can change materially after issue. Governance must distinguish a correction to supporting analysis from a change to the model's actual assumptions. A controlled response identifies which figures or calculations were affected and whether those changes alter the insurer's selected projection or disclosure. This is an inference from documented source-version events and standard validation responsibilities, not a claim that the released CMI software itself was defective in the same respect.
Liability materiality and institutional accountability
The hypothetical valuation identifies a sixth finding: a small-looking difference between annual improvement percentages can produce a nontrivial increase in a lifetime-income liability because survival probabilities compound. Under the fixed illustrative assumptions, raising the improvement rate from 0.5% to 1.0% increases present value by about $3,609 per policy. This result has no empirical status beyond the experiment, but demonstrates the direct connection between selected trend and contingent cash-flow duration. An assumption memorandum must report liability effects, not only the nominal rate.
The seventh finding is institutional. The end-to-end chain from death and exposure records to mortality base, projection, model execution, cash-flow sensitivity, approval and future validation has to be traceable. EU Article 265 documentation requirements explicitly demand records of data, assumptions, methods, validation, rationale and material changes. U.S. actuarial modeling and communication standards support related professional controls in applicable assignments. These requirements favor a stable source snapshot, an explainable decision and independent numerical reconstruction.
An approved improvement basis should therefore preserve the population and contract inventory, death/exposure definitions, base-table version, improvement model and all selected parameters, treatment of atypical periods, terminal-age behavior, projection scenario alternatives, contractual benefit cash-flow engine, validation diagnostics, quantified liability impact, prior-versus-new assumption bridge, approval actors, and date of adoption. This is an analytical consolidation of established evidence and governance responsibilities. It is not presented as a new proprietary governance framework.
Policyholder outcomes and risk transfers
The immediate contractual effect of a longevity-assumption change is often zero for an existing fixed annuity benefit: the issuer remains obligated to make the promised payments subject to the contract. The present value of that promise can change, as can the risk margin and insurer capital assessment under the applicable regime. Future offered annuity prices or options may change, which may alter consumers' choices and distributional outcomes. These consequences must be evaluated with actual contract language, market conditions and the law governing the business.
For reinsured annuity blocks, the allocation of longevity risk between insurer and counterparty determines whose balance sheet responds to a better-than-expected survival trend, and any experience-based contractual adjustments should be considered separately. An asset-intensive arrangement may also link the balance-sheet effect to interest-rate risk, investment spread and counterparty security. The actuarial mortality basis remains a necessary input but is only one part of the total transaction model. A dossier on mortality assumption governance should retain those interfaces while maintaining its defined scientific subject.
Open empirical and comparative questions
A first open question is the extent to which population improvement curves can be transferred to selected annuitant groups across countries, socioeconomic strata, purchase arrangements and guaranteed benefit types. Another concerns persistent versus temporary components of post-pandemic changes, including the roles of improved treatment, changed cause-specific risks, delayed death registration and population denominator revisions. The public evidence reviewed here does not establish a universal, long-horizon direction for these effects.
A further question is how much forecast accuracy would improve from combining national mortality statistics with increasingly granular but confidential annuity experience, and under which methods of statistical credibility, model regularization and privacy protection. Research using paid annuity datasets might compare base-level calibration with slope calibration separately and pre-register holdout windows to avoid retrospective overfitting. It would need actual data-access permission and full reproducible statistical definitions.
At extreme ages, improved registration and age validation may reveal different mortality shapes from those assumed in historical projections. In joint-life and deferred-annuity portfolios, dependence between lives, selection at commencement and behavior linked to financial conditions may be important. The relative weight of trend uncertainty, parameter uncertainty and idiosyncratic life variation over long horizons also requires empirical assessment. These are open questions with identifiable data and methodological requirements; the current study does not claim to settle them with a deterministic example.
There is also a governance-effectiveness question: do institutions that preserve source data versions, evidence-to-assumption links, model tests, formal approval records and frozen forecasts detect and correct assumption errors earlier than otherwise comparable institutions? Causal evaluation would require original institution-level evidence and a research design that addresses selection of governance practices. The monograph identifies this investigation as a future empirical direction without asserting an unverified result or naming a new doctrine.
Scope of conclusions
The combined evidence supports a coherent substantive conclusion. Insured annuity mortality improvement governance is the controlled estimation, application and revision of prospective survival assumptions for identified benefits and insured populations, supported by source-appropriate experience, explicit model parameters, quantified liability sensitivity, independent validation and accountable approval. The validity of the process is inseparable from the identity, version and limits of its evidence. Neither a large source dataset nor a sophisticated model replaces the need for a documented population-and-contract suitability argument.
These findings derive from a targeted source-grounded documentary review and mathematical synthesis. They are not an exhaustive systematic literature review, direct empirical analysis of insurer microdata, a professional opinion on the adequacy of a named insurer's reserves, an audit of a particular management system, or external peer review. They provide a full analytical treatment of the defined research object within those transparent evidentiary bounds.
Conclusion
Prospective mortality improvement is a determinant of insured annuity obligations because it changes the probability that payments will be made at future dates. This mechanism makes the choice of a mortality basis and its projected evolution economically consequential over horizons spanning decades. The scientifically defensible assumption is a coherent projection for the actual covered lives and contractual benefits, with explicit historical evidence, reference dates, selection adjustments, methods, uncertainties and measurement purpose.
The international evidence demonstrates the distinct functions of insurer experience and population projections. The 2020–2024 SOA/LIMRA annuitant experience study documents the insured population, while RPEC and CMI provide influential population-oriented mortality improvement research subject to their own calibration periods and methodological judgments. The contemporaneous 2026 SOA and CMI developments show why evidence updates, projection releases, consultations and corrections require separate governance decisions. A long-run projection cannot be chosen by title or publication date alone.
The transparent 50-year example demonstrates the mathematics linking mortality improvement to payment duration and present value. It also distinguishes a mortality level adjustment from trend acceleration, temporary mortality shock and interest-rate effects. Those differences have direct implications for scenario design, assumption-change bridges, capital interpretation, accounting and reinsurance analysis. The numerical experiment can be replicated using the stated formulas and has been expressly separated from any claim about a real insurance portfolio.
The institutional finding is that mortality improvement governance must preserve a complete chain from source evidence and data definitions through the approved model, stress tests, revision rationale and liability effects to independent challenge and authorization. Professional actuarial standards and relevant regulatory requirements support this chain in their respective scopes. It enables an insurer to decide when to retain a basis, when to revise it, and how to communicate both a chosen estimate and its uncertainty.
The present monograph completes its scientific analysis with a sourced comparative interpretation and a reproducible demonstration. It establishes no new proprietary theory, makes no claim to an unavailable insurer dataset or restricted model output, and leaves its identified empirical research questions open for separately designed investigations.
References
Actuarial Standards Board. (2010 revision, online current). Actuarial Standard of Practice No. 41: Actuarial Communications. American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/actuarial-communications/
Actuarial Standards Board. (2016 revision, online current). Actuarial Standard of Practice No. 23: Data Quality. American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/data-quality/
Actuarial Standards Board. (2019). Actuarial Standard of Practice No. 56: Modeling. American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/modeling-3/
Actuarial Standards Board. (2023 adoption; effective January 1, 2025). Actuarial Standard of Practice No. 27: Selection of Assumptions for Measuring Pension Obligations. American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/adopted-asop-no-27-selection-of-assumptions-for-measuring-pension-obligations/
Actuarial Standards Board. (2024 repeal; effective January 1, 2025). Repeal of ASOP No. 35: Selection of Demographic and Other Noneconomic Assumptions for Measuring Pension Obligations. American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/asop-no-35-documentation-and-disclosure-in-property-and-casualty-insurance-ratemaking-loss-reserving-and-valuations-repealed/
Brouhns, N., Denuit, M., & Vermunt, J. K. (2002). A Poisson log-bilinear regression approach to the construction of projected lifetables. Insurance: Mathematics and Economics, 31(3), 373–393. https://www.sciencedirect.com/science/article/pii/S0167668702001853
Cairns, A. J. G., Blake, D., & Dowd, K. (2006). A two-factor model for stochastic mortality with parameter uncertainty: Theory and calibration. Journal of Risk and Insurance, 73(4), 687–718. https://doi.org/10.1111/j.1539-6975.2006.00195.x
Continuous Mortality Investigation. (2026, March 10; subsequently corrected working-paper edition). CMI Working Paper 211: CMI Mortality Projections Model, CMI_2025. Institute and Faculty of Actuaries. https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/mortality-projections/cmi-working-paper-211
Continuous Mortality Investigation. (2026, September 16). CMI Working Paper 214: Interaction between mortality base tables and improvement assumptions. Institute and Faculty of Actuaries. https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/mortality-projections/cmi-working-paper-214
Continuous Mortality Investigation. (2026, October 15). CMI Working Paper 216: CMI_2026 consultation. Institute and Faculty of Actuaries. https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/mortality-projections/cmi-working-paper-216
Continuous Mortality Investigation. (2026). CMI News: Mortality projections, CMI_2025 corrections and working-paper updates. Institute and Faculty of Actuaries. https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-news
Continuous Mortality Investigation. (2026). Numeric listing of CMI working papers. Institute and Faculty of Actuaries. https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/numeric-listing-working-papers
European Commission. (2015, as amended). Commission Delegated Regulation (EU) 2015/35, Article 19: Data used in the calculation of technical provisions. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2432_en
European Commission. (2015, as amended). Commission Delegated Regulation (EU) 2015/35, Article 264: Valuation of technical provisions—validation. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-5927_en
European Commission. (2015, as amended). Commission Delegated Regulation (EU) 2015/35, Article 265: Valuation of technical provisions—documentation. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-5928_en
European Commission. (2015, as amended). Commission Delegated Regulation (EU) 2015/35, Article 296: Valuation for solvency purposes—disclosure. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-5959_en
European Insurance and Occupational Pensions Authority. (2026, July 15). Final report on revised guidelines on valuation of technical provisions—Solvency II Review. https://www.eiopa.europa.eu/publications/final-report-revised-guidelines-valuation-technical-provisions-solvency-ii-review_en
European Parliament and Council of the European Union. (2009, as amended). Directive 2009/138/EC (Solvency II), Article 48: Actuarial function. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2131_en
European Parliament and Council of the European Union. (2009, as amended). Directive 2009/138/EC (Solvency II), Article 77: Calculation of technical provisions. EIOPA Solvency II Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2160_en
IFRS Foundation. (2023 effective; online current). IFRS 17 Insurance Contracts. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/
Institute and Faculty of Actuaries. (2026, March 10). CMI publishes CMI_2025 mortality projections model. https://actuaries.org.uk/news/cmi-publishes-cmi_2025-mortality-projections-model/
International Actuarial Association. (2026 online register). International Standards of Actuarial Practice (ISAPs). https://actuaries.org/publications/international-standards-of-actuarial-practices/
International Association of Insurance Supervisors. (2026 online access). Insurance Core Principles and ComFrame: Online tool. https://www.iais.org/icp-online-tool/
Lee, R. D., & Carter, L. R. (1992). Modeling and forecasting U.S. mortality. Journal of the American Statistical Association, 87(419), 659–671. https://doi.org/10.1080/01621459.1992.10475265
LIMRA & Society of Actuaries Research Institute. (2026). 2020–2024 Individual Payout Annuity Mortality Experience Study. https://www.soa.org/resources/experience-studies/2026/2020-24-individual-payout/
LIMRA & Society of Actuaries Research Institute. (2022). 2014–2019 Individual Payout Annuity Mortality Experience Study. https://www.soa.org/resources/experience-studies/2022/14-19-payout-annuity-mortality/
OECD. (2025). Pensions at a Glance 2025: OECD and G20 Indicators—Life Expectancy. OECD Publishing. https://www.oecd.org/en/publications/pensions-at-a-glance-2025_e40274c1-en/full-report/life-expectancy_02c88784.html
Society of Actuaries Research Institute. (2012). 2012 Individual Annuity Reserving Report & Table. https://www.soa.org/resources/experience-studies/2011/2012-ind-annuity-reserving-rpt/
Society of Actuaries Research Institute, Retirement Plans Experience Committee. (2021). Mortality Improvement Scale MP-2021. https://www.soa.org/resources/experience-studies/2021/mortality-improvement-scale-mp-2021
Society of Actuaries Research Institute, Retirement Plans Experience Committee. (2026, October 22). RPEC 2026 Mortality Improvement Update. https://www.soa.org/resources/research-reports/2026/rpec-mortality-improvement-update/
Society of Actuaries Research Institute, Mortality Improvement Model Advisory Group. (2026, June 23). Overview of the 2026 Updates to the Mortality Improvement Model (MIM). https://www.soa.org/resources/research-reports/2026/2026-updates-mim/
Society of Actuaries Research Institute, Mortality and Longevity Oversight Advisory Council. (2026, September 22). Individual Life Insurance Mortality Improvement Scale for Use with AG38/VM20, 2026. https://www.soa.org/resources/research-reports/ag-38-mortality-improvement/
Clark, K. (2026, October 15). 2025 Mortality Improvement Survey Report. Society of Actuaries Research Institute. https://www.soa.org/resources/research-reports/mort-improvement-survey-series/
World Health Organization. (2026 current online database). WHO Mortality Database. https://www.who.int/data/data-collection-tools/who-mortality-database