RESEARCH DOSSIER
Annuity Longevity Capital Stress and Reserve Adequacy
Abstract. Annuity insurers promise cash flows whose duration depends upon survival, contractual benefit form, and financial and demographic conditions. Systematic improvement in the longevity of insured cohorts increases expected benefit duration, can increase technical provisions and the capital required to support unexpected losses, and changes the economic usefulness of hedges and reinsurance. This monograph establishes the valuation object at the level of legal contract obligations; reconstructs the evidentiary chain from mortality exposure through projected survival and discounted cash flow; and compares the distinct treatment of longevity in Solvency II, the international Insurance Capital Standard (ICS), United States statutory reserving and risk-based capital, and IFRS 17 financial reporting. It distinguishes a best estimate, an explicit valuation margin, an unexpected-loss capital requirement, and the assets available to pay policyholders. The analysis separates an instantaneous proportional stress to mortality rates from the level, trend, volatility, and parameter uncertainty of future mortality improvements. A fully specified illustrative cohort calculation demonstrates how permanent prescribed mortality stresses, improvement assumptions, interest rates, and payment expenses interact. A 20% reduction in hypothetical mortality probabilities raises the model's per-contract expected present value by 7.32%, whereas a 17.5% reduction raises it by 6.32%; these are comparable calculations of one simplified liability, not comparative regulatory capital ratios. A joint longevity-and-discount shock produces a larger increase than either isolated shock. The international comparison shows that the scientific object must be held constant before apparently similar capital metrics can be compared. Historical annuitant selection, post-pandemic mortality change, data provenance, reserve-release dynamics, asset-liability mismatch, counterparty risk, and legal effective dates remain decisive to reserve adequacy and policyholder security. The conclusions are an original synthesis of verified research and existing supervisory principles; no new doctrine or proprietary empirical calibration is asserted.
Keywords: Annuity; longevity risk; systematic mortality improvement; technical provisions; reserve adequacy; Solvency II; Insurance Capital Standard; VM-22; principle-based reserving; risk-based capital; IFRS 17; annuitant selection; mortality model; risk margin; asset-liability management; stress testing; policyholder security; basis risk.
Introduction
A life-contingent annuity exchanges present resources or an accumulation value for a sequence of payments whose continuation is governed by an insured person's survival, by an agreed joint-life condition, or by contractually specified guaranteed periods. The insurer's object is the enforceable claim on its balance sheet: the payment timing, promised amount, escalation and refund rights, beneficiaries, and options of the exact contract cohort. Population longevity provides a potential statistical input, while the mortality of individuals who actually buy or receive annuities is the closer experience object. Public promises by the state, defined-benefit pension liabilities retained by a plan, and household self-insurance against running out of savings present related but separately owned obligations. Treating all four as one capital exposure would erase distinctions essential to both empirical evidence and prudential assessment.
The research questions are how longevity shocks are calibrated, how mortality expectations enter reserves and risk margins, how adverse shifts affect eligible capital and solvency intervention, and which international comparisons survive differences in valuation perimeter and effective legal rules. The starting proposition is that longevity stress and reserve adequacy are linked through future benefit cash flows but measured through multiple distinct institutional accounts. As the IAIS valuation and capital principles explain, a solvency assessment is a total-balance-sheet exercise: technical provisions, available regulatory resources, and required capital jointly determine the picture of security. This institutional principle provides a disciplined interpretation for every quantitative example below.
The source method favors legally operative provisions, the issuing standard setter's primary materials, insurance experience publications, official mortality data repositories, and verified scholarly publication records. The paper's empirical evidence includes the publicly described scope of the 2020–2024 annuitant mortality study, the Human Mortality Database, and the released CMI_2025 projection documentation; subscription-only insurer microdata are not represented as independently examined. Academic work is used for established findings and model identification within the accessible publisher-record or abstract scope. The numerical portfolio is constructed and completely disclosed, rather than fitted to an insurer or population dataset. All values below are nominal illustration units, not estimated regulatory returns or externally observed reserve figures.
The temporal reference date is November 15, 2026. This is crucial for the European reforms effective January 30, 2027 and the American VM-22 transition extending to 2029. The July 2026 Solvency II review announcement establishes the former effective date; the NAIC reserving transition summary establishes the latter schedule. A proposed, published, or adopted rule is reported with its applicable date rather than silently installed in the valuation as an already effective requirement. Throughout the manuscript, a capital requirement is a regulatory measure associated with specified coverage, confidence, calculation period and diversification rules. The realized ability to meet future claims additionally depends on the quality, duration, liquidity and legal availability of supporting assets and on operational and recovery conditions.
Historical, Mathematical, and Institutional Foundations
The economic obligation and its demographic exposure
A life annuity transforms uncertain individual survival into a pooled lifetime-income promise. For a policyholder it reduces exposure to outliving personally managed assets; for a provider it creates a stream of contingent obligations extending into older ages. Insurance pooling reduces idiosyncratic variation around expected claims but does not eliminate a shift in mortality common to all members of a cohort. If every member's mortality improves unexpectedly, pooled benefit payments can exceed the amount previously expected even in a large book. This is the central systematic longevity exposure.
Selection enters before projection. Finkelstein and Poterba's United Kingdom annuity selection study found longer-lived annuitants than non-annuitants and larger selection effects in the voluntary than in the compulsory market in their historical sample. Their later policyholder-level contract analysis associated observed mortality with certain contract features, including payment timing and estate payments, while finding no substantive selection by annuity size in that sample. These findings justify treating annuity source populations, purchase channels, contract options, and cohorts as valuation variables. They do not establish that a historical British numerical difference applies unchanged to a 2026 annuity portfolio in another country.
Selection can affect the mortality baseline as well as sensitivity to improvement. A favorable-health buyer may begin with a lower death rate than a population benchmark. Meanwhile, duration since issue, income, underwriting selection, occupation, geography, sex and health information can alter the shape of the insured mortality curve. Contract-level death benefits alter the net effect of reduced mortality; a later death may prolong annuity payments while postponing or reducing other liabilities. The sign and size of a longevity shock therefore depend on the entire contract, not merely the existence of life-contingent cash flows.
Cash flows, survival probabilities, and the valuation operator
Let q(x+k,k) denote the conditional probability of death between successive policy years for a cohort initially aged x, with k elapsed years. Define p(0)=1 and p(t)=∏_{k=0}^{t−1}[1−q(x+k,k)] for t≥1. An annual annuity-immediate promises benefit B(t) at the end of policy year t only if the annuitant is then alive. Let e(t) be survival-contingent payment-administration expenses and d(t) the discount factor for the same maturity, currency, and economic basis. The elementary expected present value is
This is a transparent projection identity, not a substitute for the stochastic assumptions, contractual boundaries, discount curve, reinsurance treatment, or required margins of an actual valuation rule. The term “best estimate” here means the modeled conditional expected present value, subject to the exact reporting framework; an insurance accounting fulfillment cash-flow estimate may carry additional components. In the general case the sum includes scenario-specific benefits, policyholder exercise, indexation, guarantee exercise, tax or expense cash flows, and joint survival. Decrement rates must be internally consistent: a portfolio with deaths and surrenders requires competing-decrement probabilities rather than a death-only survival product.
For a mortality perturbation q^{(s)}=(1−s)q with 0≤s<1, survival probabilities rise at all remaining maturities. Where each benefit is nonnegative, survival-contingent, and other components are unchanged, the shocked annuity payment present value is nondecreasing with s. The first-order response depends on both the years of survival exposed and the discounted payments contingent on each surviving year. Duration in calendar years and sensitivity to the curve of age-specific mortality are therefore linked but are different derivatives. A guaranteed five-year period, for example, will not acquire a new survival condition merely because the longevity scenario changes.
Annuitant cash flows also exhibit a discounting exposure. Define liability discount duration approximately as D_L = −(1/V)(∂V/∂y) for a parallel yield change y, under the chosen discount basis and mortality model. An asset portfolio with a different duration and convexity can change in market value by an amount that fails to offset the liability movement. Higher expected survival generally shifts expected payments to later dates, potentially increasing the duration mismatch, but the result must be calculated under actual assets, guarantees, and curve assumptions. This explains why a mortality stress cannot be interpreted as a complete loss in own funds until the asset side and other liabilities have been considered.
Mortality evidence and projection history
The Lee–Carter mortality forecasting study modeled log age-specific death rates through age effects and a time index, demonstrating a powerful systematic way to forecast mortality patterns. Cairns, Blake, and Dowd's two-factor older-age mortality model represented distinct level and age-pattern movements and explicitly examined parameter uncertainty in longevity-risk valuation. These are established modeling architectures, not forecasts that remain valid automatically without new data and re-estimation. A forward mortality curve extrapolated from an old fitting period can be materially different from one calibrated to subsequent mortality disruptions or changing health systems.
The Human Mortality Database series provides harmonized international period and cohort data, deaths and exposures for participating countries. The accompanying HMD source and licensing guidance distinguishes constructed open estimates from underlying provider data with separate restrictions. Such population estimates support the study of general mortality trends and demographic plausibility. They cannot by themselves validate an insured annuitant's mortality assumption because selection and contract exposures differ.
The 2026 payout-annuity experience release reports that 23 parent groups and 26 individual companies contributed a 2020–2024 sample covering 3.1 million contract-years and 143,190 deaths, with income-weighted exposure also described by the authors. These published figures establish study scope and industry relevance, not an unrestricted complete mortality surface. The conditional mortality of a portfolio can be computed from its own death and exposure files when those files are made available and validated. They were not available in the open source material examined here.
The 2026 CMI working-paper release shows that its CMI_2025 model was fitted to population mortality data in England and Wales through December 31, 2025. The March 2026 CMI comparison reported a higher cohort life expectancy at age 65 under CMI_2025 than under CMI_2024, holding other parameters constant, and related the shift to the new mortality experience. The later CMI_2026 consultation concerns a proposed subsequent model, not a model already released as of this monograph's cutoff. The distinction matters in provenance, backtesting, and whether analysts characterize a projection as an established current version or a prospective change.
What counts as reserve adequacy
At the insurer level, reserve adequacy asks whether the recognized provisions represent the value required by the applicable regime for the actual insurance obligations and their risk, assumptions, guarantees, and reporting date. Capital adequacy asks whether assets and eligible own funds, after valuation and specified shocks, meet prudential requirements. Liquidity sufficiency asks whether cash, income and obtainable financing can pay benefits as they become due, including under collateral and surrender demands. Different mathematical measures can be simultaneously relevant; a high statutory capital ratio does not mechanically guarantee perfect cash matching, and a high book reserve does not guarantee that supporting assets are free of credit loss.
The IAIS ICP 14 valuation standard articulates this comprehensive picture, including margin over current estimate, reinsurance valuation, consistent treatment of risks, and explanation of accounting-to-solvency differences. International comparison therefore begins with a common specimen: one defined set of annuity contracts, one valuation date and currency, one economic and demographic scenario, explicit asset holdings, and carefully documented translations into each jurisdiction's measurement convention.
Longevity Stress Calibration
The event being stressed
A mortality improvement is a decline in age- and period-specific death probabilities or intensities, measured relative to an identified baseline. A longevity shock can refer to an immediate step change in the current mortality level, a persistent acceleration in the trend of improvements, an adverse stochastic realization of a mortality factor, a parameter revision, or the tail behavior of very old ages. Those operations have distinct interpretations and may produce widely different long-horizon cash flows. A level stress applied to every age can be replicated without forecasting an improvement time series; a trend shock must specify what happens across future calendar years and birth cohorts.
The conventional one-year insurance capital horizon also differs from the final year in which an annuity can pay. A one-year change in the economic value of obligations can reflect revisions to the expected cash flows for decades ahead. Conversely, a 30-year forecast distribution of cumulative payments is not automatically the one-year loss distribution that a capital rule requires. This horizon distinction is essential when interpreting calibration statistics and percentage mortality stress factors.
European and international prescribed shocks
Under Solvency II Delegated Regulation Article 138, the standard formula longevity-risk component measures a loss in basic own funds following an instantaneous permanent 20% reduction in mortality rates used for technical provisions. The rule directs this mortality reduction to contracts for which it increases best-estimate technical provisions before risk margin, with prescribed identification and grouping conditions. A mortality multiplier of 0.80 does not imply a 20% increase in liability, nor a 20% increase in required capital. The liability response is nonlinear in annual survival, years remaining, discount factors, contract features and the combination of cash flows. Article 138's loss metric also relies on the relevant own-funds valuation, beyond the simple liability-only exercise below.
The ICS longevity-risk technical specifications prescribe a 17.5% decrease in mortality rates at all ages for relevant policies where the reduction decreases net asset value. The standard's online Table 8 gives this factor for the listed geographic regions. The 2024 Insurance Capital Standard adoption identifies the ICS as the quantitative ComFrame measure for internationally active insurance groups. Its scope, valuation and aggregation architecture must be preserved in international capital comparison. The 17.5% and 20% factors describe prescribed input shocks in different supervisory systems, not directly comparable final solvency charges; this distinction is important even if both factors are fed into the same simplified liability model.
The EU requirement under Solvency II Directive Article 101 is calibrated to a 99.5% one-year Value-at-Risk of basic own funds across quantifiable risks. The stand-alone longevity component is combined with other risks under the applicable formula or approved internal model. The ICS calculation instead begins from its own permitted total balance-sheet framework, stresses and aggregation. Consequently, comparing the ratio of two isolated annuity liability increases does not rank capital stringency, policyholder security or aggregate group risk in either system.
Statistical calibration and identification
An empirical calibration must specify the estimation object before selecting an extreme quantile. Relevant coordinates include attained age, birth cohort, sex or permitted rating class, calendar period, policy issue year, benefit form, payment size, death-registration delay, portfolio size, and whether rates are per life or income-weighted. Let D(x,t) be deaths and E(x,t) central exposure measured on consistent time and selection definitions. The crude rate m(x,t)=D(x,t)/E(x,t) is a descriptive estimator whose stability depends upon exposure and data quality. A rate converted into annual q must use an explicit within-year hazard assumption; q≈m for small hazards is only an approximation.
Year-on-year improvements may be expressed as I(x,t)=1−m(x,t)/m(x,t−1) where both denominators are positive. These observations are noisy at older ages and can be distorted by compositional changes. An underlying model can include period and cohort effects or smoothing, but the intervention should then be tested against experience not used for fitting. The Lee–Carter age-period formulation and the Cairns–Blake–Dowd two-factor approach offer contrasting established families: the first estimates a broad mortality index, while the latter permits age-pattern movements in older populations. Both require interval estimates and attention to parameter estimation error.
Mortality shocks should distinguish process variation (who dies in a finite portfolio) from aggregate trend risk (the whole cohort's rates change), parameter estimation uncertainty, structural model risk and basis risk relative to a reference mortality index. Pooling reduces the first component at scale; it leaves systematic shifts material. An external hedge based on population death rates can fail to match the annuitant portfolio if their mortality improvements diverge. The degree of failure depends on product and population, so no universal hedge effectiveness can be inferred from a historical correlation alone.
Calibration's scientific controls
A defensible source-to-model lineage runs from mortality records to exposure validation, a selected base table, a projection law, a stress mapping, contract-level cash flows, and a final valuation measure. For each transition, the investigator records missing deaths, late registration, duplicate lives, censoring, product changes, pandemic-year handling, geographic shifts, cohort mix, and the frequency of assumptions approval. A high-age extrapolation should be shown as a modeling assumption when sparse deaths make the parameters weakly identifiable.
The CMI_2025 source documentation explicitly notes the distinction between dates of death registration and dates of occurrence; this supplies an unusually concrete example of why headline mortality movements cannot be compared without a common observation rule. The HMD data-provenance conditions likewise underscore the need to state original data owners and versions. Future improvements used to price a new contract need not be identical to adverse stresses used to assess capital adequacy, and an institution must record how it derives and governs both sets of assumptions.
Reserving Margins and Reserve Adequacy
The components must be defined before they are compared
Reserve adequacy is a conditional judgment made under an identified measurement system. In Solvency II, Directive Article 77 on technical provisions specifies the sum of a best estimate and a risk margin. The best estimate is a probability-weighted present value using a relevant risk-free term structure and realistic assumptions, with reinsurance recoverables calculated separately. That separation protects the intelligibility of gross contractual obligations when reinsurance is purchased. A mathematical increase in gross annuity benefits after a mortality shock is accordingly one ingredient of the valuation, rather than a declaration that an identical increase should appear in every reported net balance.
The risk margin serves a different measurement purpose from a deliberate conservative tweak to life expectancy. Under the pre-reform Article 37 formula, a projected path of reference-undertaking capital requirements is multiplied by a cost-of-capital rate and discounted. The formula's economic reasoning is the cost of providing capital over the run-off of insurance obligations; it refers to the future path of the capital support, and not to the face amount of an annuity contract. A change in the mortality base can alter projected underwriting risk, the run-off period, discounting, and the distribution of capital over future years. An apparently small change to a longevity assumption can therefore influence both the best-estimate component and the provision associated with future nonhedgeable risk.
Regulatory dates matter here. The 2025 Solvency II amendment specifies a cost-of-capital rate assumption of 4.75% from January 30, 2027 and moves the risk margin toward a time-adjusted capital path. The subsequent 2026 delegated amendment provides detailed changes to the regulatory framework. These revised provisions belong to the 2027 implementation period; a current November 2026 statutory analysis must not silently use them as if they had already replaced all operative assumptions. The full method also depends on the version and transitional provisions adopted in the relevant jurisdiction. This monograph therefore presents the earlier conceptual formula for explanation and explicitly distinguishes the future regime rather than calculating a purported legal risk margin from an incomplete subset of regulatory parameters.
An illustrative cost-of-capital calculation makes the distinction tangible. Suppose an otherwise specified portfolio requires a projected reference-capital amount of 20, 18, 16, 14, and 12 million monetary units at successive valuation years t=0,...,4, a flat annual discount rate of 3%, and a chosen purely illustrative cost-of-capital factor of 6%. The discounted sum, using timing t+1, is
The capital amounts are deliberately exogenous here. A true regulatory calculation must determine eligible risks, run-off capital, other appropriate commitments, discount curves, and the actual rate and time factor effective on the valuation date. The equation only shows the projection mechanics and should not be interpreted as a Solvency II filing calculation.
IFRS 17 and financial reporting reconciliation
The IFRS 17 measurement overview describes insurance contract groups through fulfillment cash flows and a contractual service margin for unearned profit, with insurance service and finance results presented separately. The IFRS 17 definition of risk adjustment characterizes the nonfinancial risk adjustment as compensation required by the insurer for bearing insurance risk. The financial-reporting treatment thus introduces an explicit nonfinancial-risk measure and a pattern of profit recognition. The risk adjustment is methodologically connected to longevity uncertainty but it is not identical to a Solvency II risk margin, an EU solvency capital requirement, or a United States statutory reserve margin.
A comparative spreadsheet must reconcile rather than pool these quantities. For each reporting basis, the investigator should show the in-force contract boundary, fulfillment cash flows, mortality and expense assumptions, locked or current discounting where relevant, any investment components, risk adjustment, contractual service margin where applicable, and recognition of reinsurance contracts held. A gross insurer payout under a life-contingent contract may appear in all the views, while the discount rates, risk margins, eligibility of recoverables and timing of recognition differ. Calling an IFRS liability a 'regulatory reserve' without a bridge would falsely imply identical financial measurement purposes.
An especially important distinction is the presence of a contractual service margin. It allocates unearned expected profit as the entity provides insurance contract services under IFRS 17 and does not represent additional loss-absorbing prudential capital in a mechanically universal way. The IFRS concepts support a disciplined reporting bridge; the actual effect on equity and regulatory capital depends on jurisdiction-specific adjustments, tax, reporting consolidation and the contract group.
United States statutory reserves and the VM-22 transition
The NAIC principle-based reserving framework explains the transition away from purely static formula-driven approaches toward prescribed minimums supplemented by risk-sensitive projections, credible company experience and economic scenarios. For non-variable annuities, VM-22 became an optional framework effective in January 2026, subject to transition provisions, and becomes mandatory for eligible newly issued contracts in January 2029. The NAIC 2026 Valuation Manual index also documents the renamed historical VM-V materials; this matters because an analyst can otherwise misread older references to VM-22 and confuse an immediate-annuity interest-rate chapter with the newer non-variable annuity principle-based rule.
Product class and issue date are indispensable metadata. Variable annuity guarantees, payout income annuities, fixed-deferred products, cash-surrender rights and embedded options do not all enter one uniform reserve calculation. A company may face different requirements for existing blocks and newer business during the transition. The amount of financial scenario sensitivity can be large when surrender behavior or minimum guarantees are market-dependent. The insurer's annuity longevity assumption must be connected to the form of benefit and the applicable valuation standard before interpreting a reserve increase as evidence of a regulatory shortfall.
A reserve adequacy test should contain, at minimum, a statutory calculation using the applicable rule, a current economic benchmark for future contract cash flows, a release-and-emergence analysis of projected surplus, and sensitivities to longevity, interest rates, expenses and policyholder behavior. The minimum statutory reserve can be sufficient under the specified rule while its economic value changes under a separate market-consistent scenario. Conversely, using overly optimistic yields for assets with credit and liquidity risks can generate an apparent excess of assets that is unavailable in a stressed run-off. The researcher should locate the discrepancy in assumptions and asset valuation rather than attribute it to 'longevity' by default.
A practical reserve-adequacy identity
For an explicitly defined valuation basis let A be admissible assets, TP the relevant technical provisions, O other liabilities and K the applicable capital requirement. Set free resources F=A−TP−O−K. The statement F≥0 is only a simplified balance-sheet sufficiency check under that same basis. It becomes misleading if A uses market values while TP is measured by unrelated locked statutory discount rates, if reinsurance recoverables are double-counted, or if K has already been deducted in a different account. The effect of a demographic shock is
with each component revalued consistently and legal restrictions on resource eligibility respected. For a pure mortality perturbation in a hypothetical unhedged asset portfolio, may be set to zero to isolate the liability component; this is a modeling choice, not a fact about actual assets or hedges. The sign of cannot be inferred solely from , because the required capital recalibration can itself depend on changed exposures, diversification and transitional constraints.
A robust decision also needs an emergence path. If provisions are adequate today under prescribed assumptions but claims increase unexpectedly in years 15–30, invested assets must still produce the correct cash flows; liquidity and reinvestment conditions matter. A margin kept today absorbs anticipated variability to the extent allowed by its design, but it must be funded by actual assets of appropriate quality and availability. The IAIS total-balance-sheet framework consequently treats asset-liability recognition, technical provisions, capital and investments as interdependent, with disclosure of the explanations for differences from general-purpose financial reporting.
Capital and Policyholder Security
The stress-to-capital transmission
A prescribed longevity shock is mapped through projected deaths and survival into contract-level payments, technical provisions and finally a change in solvency value. Under Solvency II Article 101, the one-year 99.5% VaR metric is attached to basic own funds over the specified capital horizon, subject to its scope and conditions. Its life underwriting module interacts with market, health, credit and operational risks. A prescribed mortality perturbation is a standard-formula building block. It is not an empirical prediction that mortality will actually fall by exactly that percentage in the next calendar year, and it is not the cumulative loss on the whole life of every annuity.
The ICS Level 2 longevity charge uses its own prescribed reduction and a net-asset-value concept. The global ICS standard applies to internationally active insurance groups within ComFrame, not automatically to every insurer in every market. The Insurance Core Principles address broader supervisory objectives and explain why group capital, local entity capital, gross technical provisions, risk-mitigated exposures and eligible resources must be analyzed on their own terms. A credible comparative table should identify legal entity or group, economic basis, stress factor, shock timing, participating contracts, margins, diversification, and eligibility of mitigating instruments before quoting numbers.
The NAIC risk-based capital explanation describes five broad life-insurer risk categories, supervisory intervention thresholds and 2026 consideration of appropriate treatment for longevity reinsurance products. The NAIC also emphasizes that RBC is a regulatory intervention tool, not a general-purpose ranking of insurer financial quality. It would be inaccurate to translate the EU 20% or ICS 17.5% factor into a supposed universally prescribed United States life RBC longevity stress. Under U.S. practice, the reserve test, statutory asset accounting, life RBC categories, asset adequacy analysis, reinsurance credit treatment and state supervisory judgment have to be examined jointly.
Comparative matrix
| Analytical dimension | European Union Solvency II | IAIS ICS for IAIGs | United States statutory system | IFRS 17 |
|---|---|---|---|---|
| Principal measurement purpose | Entity solvency valuation and required capital | Comparable internationally active group capital measure | State statutory valuation and regulatory intervention | General-purpose insurance contract financial reporting |
| Longevity shock | Standard-formula instantaneous 20% mortality decrease on qualifying contracts | Prescribed 17.5% mortality decrease on adversely affected policies | No single universal percentage inferred here; regime/product-specific reserve and RBC treatment | Measurement includes expected cash flows and nonfinancial risk adjustment; no universal 20% stress |
| Liability components | Best estimate plus risk margin | ICS liabilities and margins under ICS valuation specifications | Product- and issue-date-specific statutory/PBR reserves | Fulfillment cash flows plus or minus contractual service margin where applicable |
| Horizon and aggregation | SCR one-year 99.5% VaR calibration with legal aggregation | ICS risk charges, own valuation basis and aggregation | Statutory reserve and RBC calculations with prescribed components | Reporting measurement and service-period recognition |
| Primary official sources | longevity stress rule and technical provisions rule | ICS online specifications | PBR framework and RBC framework | IFRS 17 standard summary |
This table compares analytical structures. It does not equate required-capital amounts, nor assert that an international group requirement substitutes for an individual jurisdiction's separate rules. The same insurer may simultaneously prepare several valuations and maintain explainable bridges between them.
Asset-liability management, liquidity and reserves
A mortality improvement lengthens expected annuity payment duration; the associated liability value is sensitive to long-term yields, credit spreads embedded in asset returns, inflation indexation, and reinvestment. A bond portfolio can match a chosen dollar-duration but still differ in convexity, cash-flow timing, credit quality and liquidity. A longevity hedge linked to a national mortality index can reduce systematic risk while creating basis risk if the insured cohort evolves differently from that index. The resulting trade-off is a contract-specific financial analysis rather than a guarantee of perfect capital efficiency.
If a mortality stress coincides with a decrease in nominal yields, both changes can raise the present value of future fixed annuity payments. A simultaneous fall in asset market yields may lift high-quality bond prices, partly offsetting the liability increase, but credit spreads, defaults, collateral flows, illiquid asset marks, liability duration mismatch and tax can change the net result. Capital stress work therefore requires a defined joint economic scenario, not a sum of liability stresses interpreted as a solvency loss.
The insurer may transfer some longevity risk through reinsurance, buy-ins or risk swaps. Recovery depends on eligible contract wording, duration, counterparty credit, legal enforceability, collateral and termination triggers. Under IAIS risk-transfer and valuation guidance, reinsurance purchases generally do not extinguish the underlying obligations absent effective extinguishment or novation, and the insurer must account for counterparty and basis risks. A legally enforceable reinsurance receivable is not equivalent to cash already received, particularly under stress.
A policyholder experiences security through actual payments, contract enforcement, disclosure, priority under insolvency law, state-specific guaranty mechanisms where applicable, and the quality of insurer supervision. The existence or amount of a particular guaranty-association benefit cannot be assumed globally. Capital and margins help support these outcomes through defined balance-sheet and intervention channels, but their practical effect is conditioned by insolvency administration, recovery rights and timeliness of available assets.
A reproducible portfolio sensitivity exercise
To isolate the liability effects, consider an illustrative annuitant aged exactly 65 receiving USD 10,000 at the end of each year if alive. The projection has 55 possible payments and no refund, spouse's benefit, future premium, surrender, tax, expense (unless added in a scenario), mortality correlation or asset revaluation. Annual conditional death probability for policy year k, beginning at k=0, is
where p_t(s,g) is generated by replacing with , s is the instantaneous permanent fractional mortality reduction, g is a further annual prospective multiplicative reduction in mortality, y is the flat annual discount rate, and e is annual expense payable only during survival. In the following rows only one parameter changes unless stated. The cap q≤0.6 is illustrative, and the age curve was chosen for calculation rather than fitted to actual mortality experience. It must never be represented as a valid national annuitant table.
| Constructed scenario | s | g | y | e (USD/year) | PV per contract (USD) | Change from baseline |
|---|---|---|---|---|---|---|
| Baseline mortality and discount | 0 | 0 | 3% | 0 | 143,844.06 | 0% |
| Permanent 20% mortality decrease | 20% | 0 | 3% | 0 | 154,369.33 | +7.32% |
| Permanent 17.5% mortality decrease | 17.5% | 0 | 3% | 0 | 152,928.41 | +6.32% |
| Prospective mortality improvement 1% per year | 0 | 1% | 3% | 0 | 150,751.78 | +4.80% |
| Prospective mortality improvement 2% per year | 0 | 2% | 3% | 0 | 158,583.26 | +10.25% |
| Discount rate 2% | 0 | 0 | 2% | 0 | 160,910.61 | +11.86% |
| Combined mortality decrease and 2% rate | 20% | 0 | 2% | 0 | 174,034.06 | +20.99% |
| Combined plus annual administration cost | 20% | 0 | 2% | 150 | 176,644.57 | +22.80% |
The 20% and 17.5% rows borrow the stress amplitudes from European longevity stress calibration and ICS prescribed mortality shock respectively, while the entire model itself remains author-constructed. A pool of 10,000 otherwise identical illustrative contracts gives baseline benefit PV about USD 1.43844 billion and a +20% longevity-stress uplift of about USD 105.25 million, with no asset offset. Diversification across independently dying persons would reduce year-by-year idiosyncratic fluctuations about this expectation, but could not diversify away the common change in q_k assumed for the pool.
For clarity, the following pseudocode specifies the experiment without external actuarial software:
function value(s=0, g=0, y=0.03, e=0) {
let p=1, pv=0;
for (let k=0; k<55; k++) {
const q=Math.min(0.6,0.012*Math.pow(1.085,k));
p *= 1-(1-s)*Math.pow(1-g,k)*q;
pv += p*(10000+e)/Math.pow(1+y,k+1);
}
return pv;
}
Both direct recursive summation and an independent array-based recalculation were used as consistency checks; the table reports values rounded to cents and percentage changes rounded to two decimals. The examples are liability PVs, without required margins or assets. They cannot be read as Solvency II SCRs, ICS longevity capital charges, US statutory VM-22 reserves, or IFRS 17 risk adjustments. They reveal the mechanism of long cash-flow tails and the potentially large combined sensitivity.
For the cost-of-capital illustration above, direct discounted summation of the five stipulated SCR values produces 4.429 million hypothetical monetary units after rounding, a distinct margin-like number. Neither toy exercise contains the entire legal contract valuation of a particular insurer. The coherent scientific result is comparative sensitivity, not estimated regulatory compliance.
Scientific Validity and Sensitivity
Observation design, measurement and model choice
The credibility of an annuity longevity reserve is limited first by the observation design. A death record must be tied to the correct life, benefit status, attained age and date, with known periods at risk. Exposure should include insured lives present for part of the year and distinguish terminations, commutations, suspended benefits, unpaid amounts, delayed death reporting and benefit payment made after death. Fraud, fraud prevention, overseas residence, and data-system changes can affect record completeness and the time when a claim becomes known. The correct measure of risk for life-count mortality may differ from the income-weighted effect on an insurer paying heterogeneous benefits.
For an observed death count D with exposure E, a crude estimate D/E needs a confidence interval and a disclosure of heterogeneity. A high count of participant-years alone does not guarantee high credibility at ages 100 and over, where exposures may be sparse. The SOA and LIMRA 2020–2024 study offers verified metadata on the scale of contributed experience, but its paid detailed data are not a reproducible open sample in this research. The HMD population mortality records are available for wider comparisons and can identify broad period and cohort patterns, while the HMD usage agreement clarifies constraints on redistributing original provider inputs.
A valuation model should be fitted to a documented training window, validated on a genuinely withheld period and periodically assessed as new experience arrives. A favorable historical fit does not imply unbiased forecasts after economic, epidemic or compositional changes. The study design should state calibration years, holdout years, eligible age range, smoothing, observation weights, constraints on future improvement, and the handling of temporary shocks. For a mortality improvement model, reporting a single central trajectory without plausible alternative paths understates the uncertainty relevant to benefits payable several decades ahead.
The released CMI_2025 mortality projection basis illustrates both recency and measurement risk: it uses death-registration dates, while a different occurrence-date definition can change results. Its subsequent CMI_2026 consultation was not an already implemented model by November 15, 2026. Analysts should version both the input death records and the projection code. They should also report any new changes to the forecast level from changing long-run improvement assumptions, especially after years with unusual mortality. The revised CMI 2025 cohort-life-expectancy results are evidence for data sensitivity in a specified British population, not empirical proof that every international annuitant's future mortality improves at the same rate.
Identification and competing explanations
A rise in payments to surviving annuitants can reflect a larger insured cohort, greater benefit amounts, lower-than-expected death rates, benefit escalation, late claim processing, a changed surrender mix, or longer average contract duration. Separating these mechanisms requires decomposing observed cash flows into exposure, survival, contractual benefit and expense components. A retrospective comparison that attributes all payment growth to longevity will generally overstate the demographic effect.
The same identification problem arises in reserve changes. A reduction in interest rates raises the present value of later benefit payments even if mortality assumptions remain fixed. A reassessment of credit spreads or matching adjustments can change valuations separately. A stronger longevity projection can raise reserves while a reinsurance transaction creates new recoverable assets. A change in reporting basis can shift recognized margins without an equally sized change in underlying expected payments. For a defensible longitudinal analysis, fixed-basis and current-basis reserve bridges should be reported side by side.
Empirical selection must be tested by contract characteristics. The Finkelstein–Poterba annuity evidence found selection on several contract dimensions within one historical market; the earlier voluntary-compulsory comparison additionally demonstrates that purchase regime can alter the selected risk pool. Those studies establish a reason to stratify annuitant cohorts and an identification method, while their old sample periods limit any direct 2026 numerical extrapolation. A portfolio undergoing new underwriting, distribution or product redesign should treat the affected cohorts separately or explain why pooling remains defensible.
Parameter uncertainty, tail behavior and structural breaks
A stochastic model with uncertain future mortality-factor paths differs from one that also propagates uncertainty over estimated parameters. The Cairns, Blake, and Dowd parameter-uncertainty research explicitly distinguishes these sources, and the Lee–Carter long-term forecasting model supplies a benchmark historical structure. Backtesting should compare both average predictive fit and tail calibration. A model can fit moderate ages and still misstate the extreme-old-age survival tail most important to very late annuity payments.
An adequate sensitivity set therefore includes alternative mortality models, data endpoints, base tables and long-run improvement assumptions; plausible extreme-age hazard caps or deceleration; alternative discount curves; and variations in policyholder and expense assumptions. The modeler should avoid treating a mechanical prescribed stress as a substitute for a statistical model-diagnostic exercise. Prescribed stress explains regulatory compliance inputs; independent stochastic and deterministic scenarios probe reserve fragility. Reporting both is more informative than presenting one extreme stress as exhaustive.
A transparency test is to reconcile all scenario outputs to the same base cash-flow record. If one model applies benefit escalation from issue while another applies it only after retirement, the difference is contractual as well as demographic. If one simulates each life independently and another imposes a shared mortality improvement factor, their measured tail risks can diverge sharply even when central rates agree. Explicit equations and exposure definitions allow the discrepancy to be located and tested.
Quantitative validation and numerical safeguards
The table above was recalculated with independent recursions that implement and cumulative products followed by discounted summation. The finite-horizon tail has an explicit 55-payment cap; no hidden terminal benefit is inserted. End-of-year payments are always conditioned on survival through the corresponding year. The discount factor uses nominal annual compounding. Mortality improvement g modifies the age-dependent baseline in year k; it is not silently compounded to paid benefits. Reinsurance recovery, annual premium inflows and options are zero by assumption.
Checks include monotonicity (a lower death probability does not reduce PV for nonnegative survival benefits), a lower discount rate increases PV, expense increases PV proportionally to expected survival payments, and combined shocks reproduce the full recomputed scenario rather than sum stand-alone percentage changes. One-sided finite differences also quantify the sensitivities, but a 20% mortality shock is large enough to require full revaluation rather than reliance on a derivative-only approximation. The experiment is reproducible from the displayed inputs, and its results should be independently calculated if the manuscript is later used as a worked example in technical instruction.
A separate source quality test concerns the regulatory comparison. Cross-check the numeric shock factor on the actual Solvency II longevity-risk provision and ICS technical table, not a secondary infographic. Check that the effective-date confirmation for the EU revision and 2026 Valuation Manual edition listing correspond to the dates under discussion. Check that IFRS risk adjustment terminology is not confused with a regulatory risk margin. A numerically accurate scenario can still yield a scientifically incorrect conclusion if its regulatory interpretation is mislabeled.
Limits of this investigation
The international primary-source and scholarly literature review is grounded in accessible official standards, public issuer descriptions and publisher records. It does not contain restricted insurer contract-level policies, private annuitant microdata, capital-model calibration or claims accounting, nor does it establish that an extreme scenario is probable. The SOA study's detailed paid tables were not audited independently. The CMI proprietary workbook was not rerun. National transposition and precise applicability must be established for a specified legal entity before a real regulatory capital filing could be calculated. These limits directly delimit the claims: the comparative regulatory factors and analytical cash-flow sensitivities are supported; claims about a named company's reserve sufficiency would require a separate data engagement.
Integrated Results and Future Research Agenda
The analysis yields six linked findings. First, systematic longevity improvement is a common-factor exposure within annuity pools, while observed deaths of individuals also contain idiosyncratic variability. The economic importance of the common factor grows with the number and duration of future survival-contingent benefits. Second, an annuity reserve is a model-based representation of legal obligations under a specified reporting basis; it responds jointly to demographics, discounting, contractual options and administrative expenses. Third, an explicit risk margin and a capital requirement address different positions in the solvency assessment, and their accounting recognition need not match. Fourth, the EU, ICS and U.S. rules employ different legal perimeters, measurement tools and timetables. Fifth, selected annuitant mortality is empirically distinguishable from population mortality; historical selection and improvements require different control variables. Sixth, policyholder security ultimately involves admissible assets, capital coverage, funding liquidity, enforceable reinsurance, supervision and claims payment conditions as well as technical provisions.
These findings are original here in their integrated linkage to a common contract valuation and date-specific international regulatory comparison. They are derived from existing actuarial mathematics, demographic evidence and prudential standards and carry no claim to a new theoretical system. A useful synthesis equation is . The expression is a statement of necessary analytical inputs, not a proposed calibrated proprietary score or new policyholder security index.
The OECD Pensions Outlook 2026 situates the financing of retirement consumption within broader pension assets, governance, effective saving and use of assets to deliver income. For insurer-funded annuities, the implication is that a lifetime-payment promise converts one retirement-income risk into an institutionally managed reserve-and-capital obligation. The household values a reliable stream of payments, while the insurer must finance, measure and manage the entire survival distribution and contractual guarantee set. These two accounts are connected by the annuity contract and by prudential protection mechanisms.
A reproducible future research agenda follows directly from remaining evidence gaps: build accessible age-by-age annuitant exposure and death panels with contract-form metadata; map them to population reference cohorts and quantify selection and basis risk; backtest multiple stochastic improvement models with separate period and cohort holdouts; calculate actual solvency valuation bridges across IFRS 17, applicable Solvency II versions and U.S. statutory products; evaluate asset-liability matching under joint longevity, interest-rate and credit stresses; and compare counterparty risk and collateral flows in longevity reinsurance. These are distinct empirical studies with explicit data requirements rather than an assertion that their answers have already been obtained.
Research should also consider how the full benefit schedule affects the marginal sensitivity to mortality. Annuities with refund or survivor provisions can have different signs and time profiles of component cash flows; group versus retail purchase, underwriting and demographic characteristics can alter the base. High-quality future work would report exposure, model code, original observation period, parameter estimates, identification tests, error bands, currency and discount conventions, and both zero-stress and stressed asset-liability movements. It would make the replication package accessible wherever legal rights permit.
For cross-jurisdiction comparisons after January 2027, the investigator must verify the fully effective amended Solvency II Directive and related delegated regulation, including national implementation, and compare them with the appropriate edition of NAIC principle-based reserving guidance and the active ICS specifications. Improvements to mortality projection models can shift the base even if prescribed stress percentages remain stable. The research protocol should thus separate changes to underlying data, models, reporting rules and risk exposures, and identify which change causes each numerical movement.
Conclusion
Annuity longevity capital stress arises from a durable fact of life-contingent insurance: payments extend when insured beneficiaries survive longer than assumed. The resulting balance-sheet effect is governed by how the insurer defines its contract obligations, estimates mortality, discounts and projects cash flows, values margins, holds assets, recognizes reinsurance and measures unexpected losses. A stress factor applied to mortality rates is an input to this chain and only becomes a capital measure under the complete applicable framework.
The present analysis confirms the different prescribed factors in the EU Solvency II and IAIS ICS materials, the distinct U.S. statutory and RBC architecture, and the different reporting function of IFRS 17. It also documents the legal transition to EU rules effective in 2027 and the U.S. VM-22 implementation path through 2029. These timing distinctions materially affect international comparisons. Historical and contemporary mortality studies establish the relevance of selection, cohort projections and data provenance, while the fully disclosed numerical example demonstrates the nonlinear movement of annuity present values under separate and combined mortality, interest-rate and expense scenarios.
Reserve adequacy requires sufficient, appropriately measured liabilities and actual support from assets and own funds across plausible outcomes. Policyholder security adds liquidity, enforcement, supervision and recoverability. The strongest comparative analysis therefore holds the contract and economic facts constant while changing one valuation or supervisory convention at a time, and then explains every residual difference. This monograph supplies that common analytical basis, identifies its empirical boundaries and records an executable research agenda without converting the synthesis into an unverified new doctrine.
References
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. Source publication
Continuous Mortality Investigation (2026, March 10). CMI Working Paper 211: CMI_2025. Source publication
Continuous Mortality Investigation (2026, October 15). CMI Working Paper 216: CMI_2026 consultation. Source publication
European Commission (2015, consolidated). Commission Delegated Regulation (EU) 2015/35, Article 138: Longevity risk sub-module. Source publication
European Commission (2015, historical article version). Commission Delegated Regulation (EU) 2015/35, Article 37: Calculation of the risk margin. Source publication
European Commission (2026). Commission Delegated Regulation (EU) 2026/269 amending Regulation 2015/35. Source publication
European Insurance and Occupational Pensions Authority (2026, July 15). EIOPA completes Solvency II review mandate. Source publication
European Parliament and Council (2009, consolidated). Directive 2009/138/EC (Solvency II), Article 101: Calculation of the Solvency Capital Requirement. Source publication
European Parliament and Council (2009, consolidated). Directive 2009/138/EC (Solvency II), Article 77: Calculation of technical provisions. Source publication
European Parliament and Council (2025). Directive (EU) 2025/2 amending Solvency II. Source publication
Finkelstein, A., & Poterba, J. (2002). Selection effects in the United Kingdom individual annuities market. Economic Journal, 112(476), 28–50. Source publication
Finkelstein, A., & Poterba, J. (2004). Adverse selection in insurance markets: Policyholder evidence from the U.K. annuity market. Journal of Political Economy, 112(1), 183–208. Source publication
Human Mortality Database (2026). Human Mortality Database: International harmonized mortality data. Source publication
Human Mortality Database (2026). User Agreement and Data Citation. Source publication
IFRS Foundation (2026, standard overview). IFRS 17 Insurance Contracts. Source publication
IFRS Foundation (2026). IFRS 17 Insurance Contracts: Key terms. Source publication
Institute and Faculty of Actuaries (2026, March 10). CMI Model shows further rise in cohort life expectancy. Source publication
International Association of Insurance Supervisors (2024). Insurance Capital Standard Online Tool, Level 2, longevity risk L2-146–147 and Table 8. Source publication
International Association of Insurance Supervisors (2024). Insurance Capital Standard: Level 1 and Level 2 texts; adoption documents. Source publication
International Association of Insurance Supervisors (2026, online current text). Insurance Core Principles and ComFrame, ICP 14 Valuation and ICP 17 Capital Adequacy. Source publication
Lee, R. D., & Carter, L. R. (1992). Modeling and forecasting U.S. mortality. Journal of the American Statistical Association, 87(419), 659–671. Source publication
National Association of Insurance Commissioners (2025, current page). Principle-Based Reserving. Source publication
National Association of Insurance Commissioners (2026, June 30). Risk-Based Capital. Source publication
National Association of Insurance Commissioners (2026). Principle-Based Reserving Data: 2026 Edition of the Valuation Manual. Source publication
Organisation for Economic Co-operation and Development (2026). OECD Pensions Outlook 2026. OECD Publishing. Source publication
Society of Actuaries Research Institute and LIMRA (2026). 2020–2024 Individual Payout Annuity Mortality Experience Study. Source publication