RESEARCH DOSSIER
Retirement Annuity Mortality Basis Risk Across Populations
Abstract. Retirement annuity obligations depend on the survival of identifiable beneficiaries, whereas mortality evidence and longevity hedges commonly refer to other populations. This monograph investigates the resulting mortality basis risk across individual payout annuitants, group and occupational pensioners, deferred annuity holders, insured lives, and national mortality reference populations. Its central proposition is that the appropriate mortality basis is an empirically and contractually specified relationship between the covered lives, benefit-weighted exposure, observed deaths, projected improvements, and the financial settlement variable. A reference series becomes useful for pricing or hedging when its transfer to the actual insured obligations is demonstrated and its remaining error is measured. The analysis integrates original actuarial experience studies, comparative demographic research, stochastic mortality modeling, index-based longevity hedging, insurance valuation, and international prudential standards. It differentiates population level, age, cohort, socioeconomic, selection-duration, benefit-weighting, historical-period, and improvement-model mismatches; derives their effects on discounted lifetime payments; and evaluates basis risk in derivatives, swaps, reinsurance, and internal capital models. A fully reproducible synthetic annuity calculation and a variance-minimizing index-hedge example distinguish an improved hedge from a completely matched liability. International evidence from the United States, United Kingdom, OECD countries, Australia, and the Netherlands provides specific institutional comparisons while preserving source-access and jurisdictional limits. Original analytical synthesis identifies the conditions under which demographic similarity translates into financial protection. No new proprietary theory, new actuarial doctrine, or empirical discovery is claimed.
Keywords: Mortality basis risk; longevity risk; retirement annuities; pension mortality; annuitant selection; mortality improvement; population heterogeneity; demographic reference indices; actuarial valuation; survival probabilities; cohort projections; q-forwards; longevity swaps; hedging effectiveness; benefit weighting; stochastic mortality; insurance capital; reinsurance; mortality data provenance.
Introduction
An insurer selling a lifetime income promises a sequence of payments conditioned on survival. A pension scheme financing an occupational pension makes a related promise, though its membership, benefit formula, and legal allocation of responsibility can differ materially from an individual annuity contract. Both institutions therefore have exposure to the possibility that beneficiaries live longer than expected. National population mortality tables provide a widely observable reference for this exposure, but the contract is settled against the lives and benefits recorded in the insured portfolio. That difference is the starting point of mortality basis risk.
The OECD's retirement-income mortality study finds that the average survival experience of pensioners and annuitants frequently differs from that of the national population. Its cross-country synthesis describes selection effects equivalent to roughly two to two and one-half additional years of life expectancy at age 65 on average, while anticipated mortality improvements add further projected survival. These figures are comparative findings from the report's surveyed settings, not universal multipliers applicable to every country, product, or individual. Their significance is that a reference mortality table may be well constructed for its own population and still be materially inappropriate for a particular lifetime-income portfolio.
The concern is wider than a difference in average life expectancy. The precise age distribution of recipients, payment amounts, inflation escalations, survival contingencies, mortality improvement, cohort exposure, family status, and timing of contractual benefits all affect the present value and sensitivity of liabilities. A hedge written on a national index may offset the common component of longevity change while leaving a portfolio-specific component. The original basis-risk analysis by Coughlan et al. (2011) established a practical framework based on historical relationships, hedge calibration, and measured hedge effectiveness. Its case study concerned assured lives and a national population in the United Kingdom; it did not assert that all retirement annuity populations share the same relationship.
This monograph brings those strands together at the level of actual survival-contingent cash flows. It treats a population as a specified set of lives and exposure definitions rather than a convenient table name. It treats a mortality index as a measurable settlement process rather than a substitute for the contractual obligation. It also investigates the circumstances in which using a closely related reference can be defensible despite imperfect identity. The result is an analytical account useful to researchers, actuaries, insurance economists, supervisors, and readers of international retirement-income evidence.
Object, Boundaries, and Research Method
The object is mortality and longevity basis risk arising when retirement annuity or pension obligations are valued, projected, insured, or hedged using mortality data obtained from a population different from the population whose survival determines the obligation. A direct annuity payout, the retirement phase of a deferred income annuity, a defined-benefit pension payment, and a life-contingent survivor benefit fall within the object when mortality determines cash flows. A fixed-term payment promise with no survival contingency requires different analysis; the financial liability remains, but direct mortality sensitivity depends on its contractual conditions.
Four research questions organize the study. First, which measurable differences distinguish annuitants, pension members, insured-life policyholders, and national reference populations? Second, through which mathematical channels do those differences enter expected payouts and their uncertainty? Third, when does a population-referenced hedge reduce economic longevity risk, and how large is the exposure that it leaves? Fourth, which evidence and governance practices support a defensible mortality basis across jurisdictions?
The method combines authoritative actuarial and demographic studies, original methodological papers, official experience-program documentation, and regulatory sources. The OECD comparison of mortality tables supplies a comparative framework for selection and standard-table development. The SOA payout-annuity experience study supplies documented American industry coverage and observation years. UK CMI studies document pension and annuity investigation design. Academic publications supply the mathematical basis for coherent mortality forecasting and hedging. Accounting and solvency sources establish how mortality assumptions enter formal valuations.
The source audit separates directly accessible official or journal abstract findings from results requiring restricted subscription data or underlying individual records. The public summaries of several recent experience studies establish scope, participation, and stated methods, while the underlying exposure-level records and some full results remain access-restricted. Accordingly, the quantitative example below is explicitly hypothetical, constructed from declared input assumptions, and presented as an algebraic sensitivity demonstration. It is not a fit to SOA, CMI, Australian, or Dutch microdata and is not an estimate of a market hedge price.
Authorial synthesis consists of comparing these sources, deriving direct implications for financial exposure, and identifying failure modes in transfer of reference mortality. Established demographic models and variance-minimizing hedge mathematics are used as scientific tools. The analysis makes no historical priority claim for these tools.
The Survival-Contingent Contract
The contractual unit determines the economic object that a mortality basis must represent. A single-life immediate annuity may pay a fixed sum at each scheduled date while the named annuitant is alive. A joint-and-survivor annuity depends on the joint survival status of two lives and may reduce payments after the first death. A guaranteed-period annuity can continue payments to a beneficiary for a stated term after the covered life dies. A deferred lifetime annuity begins only after a deferral period or other stated vesting condition. Each produces a different functional dependence on mortality.
For an annual life annuity payable in arrears to a person aged , with an annual unit benefit and annual effective discount rate , the expected present value over horizon is
where is the probability that the covered person survives from valuation to the payment date . Under annual conditional death probabilities , survival is the product
The second subscript denotes calendar time or the relevant projection year, since projected mortality is generally a surface over attained age and calendar year. Actual contractual payments can involve months, escalation, tax withholding, benefit reductions, guaranteed terms, commutation, and administrative timing. The formulas therefore define an exact stylized liability, not every institutional contract.
The distinction between a selected covered life and a period-certain entitlement is crucial in basis-risk analysis. If a contract guarantees the first ten payments regardless of death, changing mortality during that period need not change those guaranteed payments. If the benefit is payable only on survival, the same mortality movement directly alters their expected value. A derivative whose index is a national death rate does not automatically reflect guaranteed benefits. Contract conditions are therefore the first data field in a proper population-basis analysis.
This accounting of contingent states also makes it possible to understand why two portfolios with the same average age might have different longevity sensitivity. A portfolio with heavily backloaded payments and generous survivor continuation weights survival in later years more strongly than a portfolio with modest level payments. The comparison must be between cash-flow-weighted exposures rather than demographic labels alone.
Populations and Their Observable Mortality
The national population is typically a residency- or jurisdiction-based statistical population from which deaths, person-years, life tables, and projected mortality are constructed. An occupational pension population consists of persons participating in a particular employer, industry, or plan arrangement; the pensioner subset is conditional on attaining pension payment status. An individual annuity population consists of purchasers or annuitants meeting product-specific eligibility, election, and payment commencement criteria. An insured-life population includes life insurance contracts assembled through underwriting, distribution, and policy persistency processes and is not necessarily comparable to annuitants.
The SOA Pri-2012 retirement tables distinguish status, sex, job category, and weighting basis. Their underlying private-sector retirement-plan investigation generated approximately 16 million life-years of exposure and 343,000 deaths. The study's availability of both benefit-weighted and headcount-weighted tables is a substantive reminder that the relevant exposure can be dollars of pension as well as counts of people. For a portfolio of high-benefit survivors, a dollar-weighted mortality rate can differ from the count-weighted rate even in the same calendar year.
The SOA 2020–2024 payout-annuity study covers 23 parent company groups, 26 individual companies, 3.1 million contract-years of exposure, approximately $33 billion of annual-income-years of exposure, and 143,190 recorded deaths. Its stated purpose includes comparisons with an earlier annuity mortality table, preceding experience, and US population mortality. These figures describe the study's dataset. They do not establish that participating insurers represent every US retirement contract or that its reported totals can be transplanted to a particular portfolio.
Deferred annuity populations introduce a further distinction. The SOA 2015–2024 deferred-annuity investigation reports 26 participating companies, 80 million contract-years of exposure, and more than 1.4 million deaths across a ten-year window. The study concerns a different set of contracts and risk stages from the payout-annuity investigation. Deferred ownership, accumulated account value, commencement election, and benefit vesting require separate analysis before treating deferred annuity mortality as a substitute for payout mortality.
An SOA individual life experience report provides another relevant comparator. Life insurance typically pays on death, while a lifetime annuity pays while the life continues. Underwriting requirements, policy continuation, face amount, and economic incentives create different sampling processes. A common mortality event therefore enters the two product liability equations with different signs and financial weights. Combining insured-life and annuitant data can be valuable for understanding shared mortality dynamics, but it requires explicit selection modeling.
Selection and the Meaning of Population Differences
Selection is the process through which a target population acquires a mortality distribution different from a broader reference because membership is conditional on observable or unobservable characteristics. An individual can enter an annuity pool by purchasing lifetime income, an occupational pension pool by having a particular employment history and satisfying plan rules, and an insurance pool through underwriting and policy continuation. The conditions are different, and so are their mortality implications.
The OECD's selection-factor discussion states that pensioner and annuitant groups generally exhibit lower mortality than the general population and that voluntary individual arrangements can show stronger selection than broad mandatory arrangements. It also notes changes by age and duration. An annuity purchaser's favorable private information about health can influence purchase timing; over subsequent years, survival-based composition changes and the effect of initial selection can attenuate. These propositions explain why a fixed scaling factor applied to every future death rate may be inadequate.
Selection is not a single permanent discount on mortality. If denotes a reference probability and the target annuitant probability, a simple representation is
where depends on attained age, calendar year, duration since entering the portfolio, and relevant covariates . This expression is a bookkeeping representation of a multiplicative relativity, not a claim that the true relationship is always proportional. If a researcher assumes a constant while the empirical relationship changes with purchase duration, the future discrepancy becomes part of basis risk.
The apparent mortality advantage of a portfolio may also reflect survivor composition, socioeconomics, benefit amount, selection into work, and incomplete death reporting. A low observed death rate is an estimate conditioned on the data capture mechanism. It is not, by itself, proof of causal health superiority. A rigorous portfolio comparison begins with matching observation age, calendar period, geography, and reporting definition; subsequent differences can then be investigated as hypotheses concerning selection and heterogeneity.
Occupation, Wealth, Geography, and Health
Retirement populations can differ systematically by occupation, lifetime earnings, education, locality, access to health services, and health status. These characteristics interact with pension participation and insured-benefit amounts. Accordingly, the relevant comparison may be between wealth-weighted retirees and a reference containing people who never acquired comparable retirement rights.
The Pri-2012 study's occupational categories distinguish blue-collar, white-collar, and unknown-collar experience. This is a data-defined stratification within a particular US private-plan investigation. Its existence establishes that occupational composition is recognized as a source of variation, while individual category rates require careful age-, sex-, status-, and weighting-standardization before interpretation.
The CMI pensioner mortality study for 2017–2024 reports analyses by pension amount bands and an area-level deprivation indicator, as well as calendar-year comparisons. The CMI self-administered scheme investigation specifies that its pensioner and dependant records arise from submitted schemes and that active and deferred members are outside that particular dataset. The scope is thus genuinely pensioner-specific. It cannot be treated as a full employment-life or whole-national-population dataset.
Geography adds a second dimension. The Australian Government Actuary's life-table program produces national mortality references for government and industry applications, while separately identifying investigations into regional and socioeconomic subgroup mortality. The existence of national and subgroup measurements demonstrates why a country-level rate is a starting reference rather than a fully conditioned estimate for every pension cohort. For an international annuity portfolio, migration, residency, and jurisdiction-specific recording can add further disagreement between the lives covered and the index on which a hedge settles.
A health-conditioned or enhanced annuity requires an especially explicit population definition: eligibility, medical evidence, underwriting, and conversion of health information into payout rates must be documented. Assumptions drawn from ordinary annuitant experience may be directionally unsuitable for an enhanced-income population. The sign of the error depends on the contract, selection, and available observations. Population labels never substitute for the specific terms defining inclusion.
Mortality Data and Statistical Provenance
The foundational empirical object is a death count observed over a specified risk set and exposure period. An age-specific central death rate may be estimated as , where counts deaths assigned to age and year , and is corresponding person-years exposed. The one-year death probability is distinct from ; converting between them requires an assumption about the distribution of deaths within the age interval or a specified hazard model. Applying a tabulated as though it were a central rate creates avoidable methodological error.
Contract records can report death, benefit cessation, beneficiary continuation, record exit, or insurance claims on different dates. Missing dates, late notification, inconsistent age conventions, duplicate lives, and administrative terminations can contaminate both numerator and exposure. A pension payment file may continue a survivor's benefit after the original annuitant's death, whereas a mortality study may count that death for one status and open a new dependent-status exposure. The research record must make such transitions explicit.
The public SOA payout-annuity study description identifies contract-years and annual-income-weighted exposure as separate denominators. For a reader estimating future dollar payouts, weighting by benefit amounts is conceptually pertinent. For a researcher comparing biological death probabilities, life-weighted exposure may be more natural. The two should be reconciled, not silently exchanged.
The CMI's annuity mortality investigation distinguishes pension annuities, bulk annuities, enhanced annuities, and other business categories. This classification matters before any merger of observations: product definitions determine whether an apparently similar record belongs in the valuation population. Careful provenance tracks company and scheme inclusion, observation periods, age definition, benefit status, currency, data revision, and permitted use.
Missingness deserves substantive treatment. If death-record linkage is less complete among people who moved abroad or whose benefits were transferred, apparent survival may be overstated. If frail participants are more likely to disappear from administrative exposures, the opposite bias can arise from premature removal. Any expected-to-actual mortality ratio therefore needs its full cohort denominator, event definition, and censoring practice. Comparability must be established before numerical precision can be interpreted.
Period, Cohort, and Improvement Basis
A period life table applies rates observed or estimated within a calendar period across synthetic ages. A cohort life table follows or forecasts mortality for people born in a specified year as they advance through calendar time. Lifetime annuity liabilities use prospective cohort survival probabilities; hence their valuation depends on mortality change over the remaining payment years rather than solely on today's cross-sectional life expectancy.
The classic Lee–Carter mortality model relates log age-specific mortality to an age profile and a time-varying mortality index. It offered an influential basis for mortality forecasting and uncertainty measurement. Yet its one-population index need not maintain a stable relationship with a selected retirement portfolio. A reference population may improve faster than the pensioners for a period, or the reverse may occur for certain age groups.
The Li–Lee coherent population approach extends the forecasting problem to related populations by separating common long-run movements from group-specific variation. For basis risk, the essential insight is methodological: a stable common mortality trend and an uncertain differential trend should be modeled jointly. Assuming that both populations experience identical future improvements removes a central source of hedge error by assumption.
The CMI Mortality Projections Model documentation describes a model calibrated to England and Wales mortality through the end of 2025, together with sensitivity to assumptions and new data. A population projection is useful evidence for broad improvements. Its direct application to selected pensioners still requires an explicit relationship between the population and portfolio improvement rates. This distinction is equally important when a model has separate parameters for convergence, smoothing, and long-term trend.
A difference in mortality basis can originate in a level effect, such as lower mortality among annuity buyers at current ages, or in an improvement effect, such as a different rate of change in their future hazards. Because annuity present value sums survival over decades, small sustained improvement differences may become economically significant even when current ratios appear modest. A model governance file should store base-table year, projection start, improvement model version, calibration span, long-run assumptions, and any age/country transfer adjustment.
Stochastic and Multiple-Population Models
Systematic longevity risk concerns uncertainty in mortality trends affecting many covered lives. It can be modeled by a stochastic factor shared across age groups, potentially combined with age-specific and cohort-specific components. A reference model can be adequate for general mortality forecasting while remaining poorly calibrated to cross-population liability covariance.
Cairns, Blake, and Dowd's two-factor stochastic mortality study analyzes post-age-60 mortality with two factors affecting ages differently. It shows why term and cohort matter to the assessment of aggregate longevity exposure. Its findings refer to the model, data, and historical setting examined; independent retirement populations still require their own calibration or linkage.
A stylized multiple-population model can express each log mortality surface as a common time factor plus a population-specific deviation:
Here is shared, describes additional population-specific evolution, and captures residual measurement or model error. This is a generic factor representation used for explanation; it is not introduced here as an original theory or fitted estimator. Identification, estimation, and extrapolation require actual data. Differences in weighting, vintage, and observation quality can render a naive joint fit misleading.
The assessment of a mortality model involves more than in-sample fit. The comparative study of model residuals evaluates stochastic mortality models through residual diagnostics; related mortality forecast comparisons warn that historical fit does not establish plausible long-range forecast distributions. For basis risk, these lessons imply an additional test: reproduce the correlation and relative scale of forecast errors across the liability and reference populations, not merely the trend in each series separately.
Age and cohort alignment are essential. An index referencing mortality at age 70 in 2030 may not hedge liability to an annuitant currently aged 65 whose relevant death events occur at many attained ages in later calendar years. Both instruments can respond to longevity improvement and still have different sensitivities to age, calendar shocks, and late-life tails.
Liability Cash Flows and Basis Translation
For a portfolio of lives , scheduled annual benefits , and survival-conditioned eligibility , a realized discounted payment stream can be written
with the appropriate discount factor. The expectation of replaces by the actual target-population survival probability, conditioned on contract states. A mortality table for another population supplies an estimator of this expectation only after a documented transfer model.
The effect of a change in a conditional one-year death probability is transmitted to every subsequent survival-weighted payment. For one annual level benefit, lower increases survival probability from onward and thus raises the liability value. A large benefit payable late in life can place substantial economic weight on mortality uncertainty even if few beneficiaries reach that age. This is one reason crude early-retirement death counts may explain only a portion of long-tail reserve risk.
Basis risk can appear even if two portfolios have identical average life expectancy. One may be composed of a mixture of high- and low-mortality lives and another of more homogeneous lives. Their survival curves can cross, and later benefit escalation can amplify differences at ages where one population has stronger survival. The correct comparison is the time path of weighted expected cash flows and covariance with the settlement index.
A pricing exercise requires prospective rather than purely retrospective mortality. The difference between the present value using target survival and a reference-only basis represents a valuation-basis difference at a specified valuation date. The variation of that difference under uncertain future experience represents basis risk. These are related but distinct quantities: a deterministic bias is not the same object as a random future hedge residual.
Reproducible Annuity Valuation
Consider three hypothetical annual mortality bases for a person aged 65. They are constructed solely for illustration and are not calibrated to any published nation, pension scheme, insurer, or mortality study. The target annuitant basis begins at and increases by 7.7% per year of attained age; the reference-population basis begins at and increases by 7.0%; a third synthetic pensioner basis begins at 0.014 and increases by 7.3%. Each applies the stated age progression to 40 annual end-of-year survival-contingent payments, with a 3% annual effective discount rate and an annual benefit of one unit. The calculation sets and . These assumed progressions are age profiles, not separately estimated calendar-year mortality improvements.
Direct summation yields the following present-value factors to six decimals:
| Hypothetical mortality basis | Initial | Annual age progression | Forty-payment annuity factor |
|---|---|---|---|
| Target payout annuitants | 0.012 | 7.7% | 14.851301 |
| Reference population | 0.016 | 7.0% | 13.855013 |
| Occupational pensioners | 0.014 | 7.3% | 14.343741 |
The target annuity factor exceeds the population-reference factor by 0.996288 benefit-years, or approximately 7.19% of the reference value. If the annual payment is $10,000, these illustrative present values are approximately $148,513 and $138,550, a difference of $9,963 per contract at this discount basis. This is an exact result of the stated synthetic exercise, before expenses, policy options, taxes, capital, risk margins, or market pricing adjustments. It demonstrates the direction and scale that an assumed population mismatch can produce, not the typical difference between real annuity prices.
For a second sensitivity, multiply every one-year in the target and reference projections separately by 0.90, representing an illustrative 10% lower mortality level throughout the future horizon. The target factor becomes 15.351278 and the reference factor 14.386325. Their responses to this hypothetical perturbation are 0.499977 and 0.531313 respectively. Even the same proportional change in mortality rates creates different cash-flow responses across the two bases. A hedge notional calibrated to a simple ratio of population sizes or premium totals therefore need not offset liability value movements.
This example isolates mortality translation. Real valuations also require time-varying interest rates, payment timing, payment escalation, expenses, survivor states, multiple birth cohorts, and explicit limits on extrapolation. The reproducibility appendix gives executable pseudocode sufficient to regenerate the factors without importing external statistical datasets. It provides a diagnostic benchmark for implementation, not an economic forecast.
Systematic, Idiosyncratic, and Model Risk
For a finite portfolio, realized death counts fluctuate even if the underlying mortality probabilities are known. This idiosyncratic or sampling variation generally diminishes relative to portfolio size under an assumption of weak dependence. Shared changes in mortality level or improvement across all beneficiaries do not vanish merely because a portfolio becomes large. Actuarial analysis must separate these phenomena from structural disagreement between the target and reference populations.
Mortality basis risk is particularly associated with the portion of a target portfolio's systematic movement that is absent from, or imperfectly correlated with, the reference. If the two groups share a common mortality improvement factor but have different responses by age, they can exhibit high correlation at one horizon and poor protection at another. A population-specific factor may also trend independently over time, changing the long-run relationship between their mortality rates.
Model risk arises because the selected projection process, estimation window, group structure, and recalibration convention can change future liability sensitivities. Cairns's robust longevity-hedging research specifically examines hedge ratios under parameter uncertainty, Poisson risk, and future recalibration. It finds that some single-instrument strategies are sensitive to recalibration even when other examined uncertainty sources have less effect, motivating careful multi-instrument construction.
An observed high historical correlation between two death-rate series should be interpreted alongside sample period, detrending method, exposure size, common pandemic shocks, and confidence intervals. Correlation of mortality levels may largely reflect shared age or secular trend rather than correlation of unexpected changes in liability value. The relevant covariance is the covariance of hedge-relevant unexpected cash-flow values at the stated hedge horizon.
Index-Based Longevity Hedges
A population-linked q-forward exchanges payments depending on realized mortality rates for a specified population, age, year, and settlement convention. A longevity bond or survivor swap can reference survival or death rates of a defined cohort. A tailored indemnity longevity swap can be structured to follow a particular scheme's actual payments or experience more closely. Each choice trades off index observability, contract specificity, transaction costs, liquidity, counterparty exposure, and basis risk.
The foundational Longevity Hedging 101 framework requires an explicit account of the historic and demographic relationship between the liability population and the reference. Its reported UK study finds a stable long-term relationship between the investigated assured-life and national populations and models favorable hedge effectiveness for an appropriately calibrated index instrument. A reader should not convert this finding into an assumption of perfect hedging for unrelated retirement annuity contracts.
For a life annuity, lower-than-expected target mortality typically increases payout obligations. A hedge designed to compensate for longevity gains should yield positive value in relevant reference-survival scenarios. But a common movement in the index may be accompanied by a larger or smaller movement in the annuity population. The residual remains economically important even when the hedge offsets much of the aggregate trend.
Ngai and Sherris's study of retail annuity hedging compares longevity-derivative performance across different life, deferred, indexed, and variable annuity designs. Its results show that hedge effectiveness depends on product design and accompanying market exposures. In particular, a longevity hedge alone does not remove inflation risk from an inflation-indexed annuity or other nonmortality components of variable benefits. Contract design determines the boundary of what mortality-index hedging can accomplish.
Index construction is therefore a substantive research question. The index must specify deaths or survivors, calendar cutoff, geographic coverage, sex or age disaggregation, revisions, data publication lag, and fallback rules. An ambiguous index definition converts a measurable actuarial relationship into a possible contract dispute. A carefully specified index is more auditable and less subject to data-revision disputes, even though demographic basis risk persists.
Hedge Ratios and Residual Risk
Let denote the unexpected change in present value of the liability over a defined hedge interval and let denote the value change of a unit hedge, oriented so that its positive payoff compensates a positive longevity liability shock. A portfolio with hedge units has unhedged residual . Under finite second moments, the variance-minimizing hedge ratio is
provided the hedge variance is positive. The minimum residual variance is
where is the correlation between the relevant liability and hedge value changes. These formulas are ordinary least-squares projection identities. They specify the optimal ratio for a narrowly defined quadratic criterion; they do not account automatically for liquidity, credit, nonlinear stress, or constraints on available hedge notionals.
For a purely synthetic example, suppose standard deviations of unexpected liability and unit hedge-value changes are 14 and 11 monetary units, with correlation 0.85. The minimizing ratio is hedge units. The remaining variance is , corresponding to a residual standard deviation of approximately 7.375. The fraction of variance eliminated under these assumptions is 72.25%. A lower correlation of 0.60 would reduce this fraction to 36%, regardless of the same marginal standard deviations. These numbers are analytical examples rather than observed correlations or documented market hedge performance.
A hedge may have a high percentage effectiveness yet leave a large absolute exposure if the original liability is substantial. Conversely, a modest variance reduction may still have value when the hedge is inexpensive or addresses a particular capital stress. Evaluation should report unhedged and hedged value-at-risk or tail measures where appropriate, absolute sensitivity by cohort, cash-flow mismatches, and the probability of a large residual, rather than only an attractive percentage.
A static index hedge calibrated today can deteriorate as the population mix, payout schedule, discount curve, and models change. Future recalibration may cause the liability factor loadings to shift. A dynamic hedge adds implementation costs and the risk of poor execution when data are delayed. A multi-instrument hedge can match more ages or maturities and reduce concentrated sensitivity, but only if its reference populations and settlement definitions are well understood.
Reinsurance and Contractual Settlement
Reinsurance can transfer some longevity exposure from an insurer or pension arrangement to a reinsurer under an individually negotiated contract. The settlement variable is central. A contract paying on actual benefits or observed experience of a specified ceded population can align closely with underlying obligations. An index-based reinsurance or derivative arrangement settles on a distinct reference and therefore creates population basis risk. The degree of transfer depends on exactly which deaths, life statuses, payments, exclusions, and reporting dates are contractually recognized.
An indemnity-style arrangement may reduce a narrow population mismatch while preserving counterparty, administrative, commutation, legal, and data-integrity risks. An index arrangement may facilitate standardization and external trading but leaves the gap between insurer cash flows and index settlements. It is unsound to describe either as automatically superior without specifying the hedger's objective, data quality, eligible risks, and financial costs.
Cross-border transactions require particular care. Pension trustees or insurers may report mortality by local administrative standards while a derivative uses a national statistical series published with a lag or revised after settlement. Currency differences, discount rates, and legal rules regarding benefit eligibility are outside the mortality index and must be valued separately. Changes in scheme membership, buy-ins, buy-outs, or transferred annuities can shift the mortality characteristics of the exposure after the hedge was entered.
A precise reinsurance audit should document the target-life schedule, death and survival evidence, benefit cash-flow schedule, benefit indexation, population reference, attachment and settlement date, claims notice rules, basis adjustments, collateral, recapture triggers, and dispute mechanism. This contractual analysis makes basis risk part of the actual promise rather than only an output of a stochastic model.
International Evidence and Institutional Comparability
The United States supplies detailed evidence on the distinction between industry annuitants and private pensioners. The SOA's individual payout-annuity investigation observes contracts already characterized by payout-annuity eligibility and market participation. The private-plan Pri-2012 data represent a different plan-based population. Each can be informative about longevity risk, but a direct comparison requires age, sex, period, weight, benefit status, and selection alignment. Neither dataset describes the entire US population.
The United Kingdom has multiple distinct experience investigations. The CMI annuities program records separate pension annuity categories. The CMI scheme-pensioner program concerns pensioners and dependants of participating occupational schemes. CMI Working Paper 210, reported in the CMI's March 2026 program update, compares bulk-annuity and self-administered pension experience over 2016–2023. A related CMI technical discussion of that comparison describes how dataset alignment alters the apparent difference. The restricted underlying working-paper tables were not independently recalculated for this monograph.
The Netherlands offers a contrasting projection tradition. The Royal Dutch Actuarial Association's AG2024 mortality projection table uses Dutch experience and comparable European data, with explicit treatment of pandemic-era excess mortality and long-run trends. A national projection basis can support pension assessment while the mortality structure of a particular pension fund continues to depend on its membership. Countrywide adoption of a reference does not itself eliminate scheme-specific mortality.
Australia illustrates the relevance of national life tables and market scale. The Australian Life Tables 2020–22 are based on a three-year national observation period and support multiple public and private financial applications. The OECD's country comparison explains that the Australian annuity market is relatively small and that a reference annuity table produced from limited provider data cannot automatically be regarded as fully representative. When annuity data are sparse, uncertainty in the selection adjustment can be economically as consequential as uncertainty in the broad population trend.
A nationwide pension projection is also institutionally different from an insured annuity contract. The US Social Security Administration's 2026 Trustees Report publishes long-range demographic assumptions for social insurance finance. Its projection population, financing, benefits, and legal framework differ from those of commercially issued annuities. Such public projections supply useful national context, but they do not establish insurance-specific mortality without an additional transfer and validation exercise.
International comparison therefore has a methodological condition: tables must be compared in terms of the observed lives, exposure measures, observation periods, benefit weighting, and social institutions that produced them. A published rate cannot be imported solely because its age, sex, and country are known. Differences in pension coverage, economic inequality, purchase incentives, and data practices alter the meaning of membership in the target population.
Pandemic-Era Data and Projection Consistency
The pandemic introduced abrupt mortality changes and intensified the problem of reconciling experience periods with long-term improvement assumptions. An unusually high death rate in a short observation window can affect the base table, while projection models make separate assumptions about whether and how that elevation declines. Combining an already adjusted base with an incompatible projection may double count or omit parts of the event's assumed future effect.
The CMI's Working Paper 214 overview explicitly discusses the need for consistency between the mortality base and improvement assumptions, including differences between general population and specific portfolios. This is a direct issue for basis risk: if the national index and the annuitant liability are calibrated to different pandemic observation windows or rebound assumptions, an apparent stable population relativity may be mechanically distorted.
The Australian actuarial review of 2020–2022 describes materially different mortality patterns by calendar year, including a notably different Australian timing of pandemic effects than in countries with earlier large mortality waves. That finding illustrates the danger of applying a common pandemic factor to unrelated countries or product cohorts. The appropriate mortality basis must carry its historical period and local epidemiological context.
The CMI pensioner study for 2017–2024 likewise emphasizes year-specific variation in the experience of participating pensioners. The observed response of a particular population can reflect its age and socioeconomic distribution, access to care, geography, and institution-specific record structure. An index-based longevity swap whose mortality observation period is shifted relative to the liability measurement date therefore bears both demographic basis and data-vintage risk.
The analytical implication is that a base-mortality relativity and a future improvement relativity should be tested jointly. A study that fits the first without the second may reproduce past A/E ratios and still give misleading projections of annuity duration and hedge covariance. Scenario exercises should therefore vary both level and improvement, and should make the permanence or transience of shocks explicit.
Prudential, Accounting, and Governance Implications
The IFRS 17 accounting standard requires insurance groups within its scope to measure insurance contract obligations using current estimates of fulfillment cash flows, a risk adjustment for nonfinancial risk, and the contractual service margin where applicable. Mortality expectations affect cash flows and the assessment of uncertainty. IFRS accounting measurement is a specific reporting framework; the existence of an index hedge does not automatically imply full accounting offset of the underlying mortality risk.
The International Association of Insurance Supervisors' Insurance Core Principles articulate requirements for reliable, decision-useful valuation and risk-sensitive supervision. Evidence quality, appropriate population data, internally controlled assumptions, and independent assessment are therefore relevant to a longevity basis adopted for solvency analysis. Local implementation remains governed by applicable national law and supervisory requirements; international principles do not supply one worldwide mandatory mortality table.
In the European Solvency II standard formula, Article 138 on longevity risk specifies a permanent instantaneous 20% reduction in mortality rates for contracts where lower mortality increases technical provisions, subject to the rule's precise terms. This is a regulatory stress definition rather than an empirical prediction that all populations will improve by 20%. A portfolio's true basis risk under a population index may not be fully described by that one prescribed stress.
Risk governance should begin with the identities of the data owner, valuation population, model version, and hedge settlement series. Its record should then show base mortality, projected improvements, demographic mix, benefits, uncertainty, sensitivity results, the approval decision, and future reassessment triggers. A model using annuitant experience without controlling for the payout amount distribution can systematically understate the economic sensitivity of a high-benefit subportfolio. Conversely, using dollar-weighted mortality for a purely headcount-dependent question can introduce an irrelevant financial composition effect.
Capital adequacy is affected by the distribution of residual loss. A hedge that reduces central forecast variance can leave tail dependence, counterparty risk, or a severe portfolio-specific mortality improvement unexplained. Solvency and accounting assessments should therefore record hedge results in their actual frameworks rather than relying solely on a generic index-correlation statistic.
Validation, Stress Testing, and Decision Audit
A reproducible population-basis validation begins with an exact population statement. The analyst specifies inclusion dates, ages, birth cohorts, sex where collected and lawfully used, pension status, product type, benefit amount, and geography. The reference is documented to the same precision. Every death rate must be supported by exposure and event-count definitions, with a record of record linkage, missingness, and revision history.
Historical fitting follows. A/E mortality comparisons are standardized for attained age, sex, year, and relevant statuses. The study separately estimates level differences and historical changes in the difference, with uncertainty intervals. If data volume is insufficient to estimate stable improvement relativity, the analyst records that limitation and evaluates several plausible future paths rather than presenting an unstable point estimate as settled science.
Forecast validation then tests whether models reproduce observed experience outside the estimation window and whether joint population forecasts capture the covariance of unexpected changes. The model-fit research in Insurance: Mathematics and Economics supplies a basis for residual testing, while Cairns's robust-hedging analysis underscores the importance of changes in model calibration. For annuity liabilities, validation should include survival at very old ages, where limited data and benefit concentrations may have disproportionate financial consequences.
Hedge testing connects the validated model to actual cash flows. At a common valuation date, the actuary reconstructs contractual benefits, estimates the value change under shared and population-specific shocks, applies the exact derivative payoff formula, and measures residual outcomes. Useful outputs include the covariance-minimizing hedge ratio, sensitivity by age and maturity, stress loss, cash-flow timing mismatch, and cost assumptions. The hedge can then be retested after future experience or model revisions.
A particularly revealing stress changes the relative improvement of target and index populations while leaving their broad common trend intact. A second varies selection decay by duration after annuity purchase. A third changes the distribution of large benefits among survivors. A fourth introduces a time lag or retrospective data revision in the index. Together these tests distinguish genuine population mismatch from contractual, demographic, or data-processing sources of residual loss.
Editorial QA requires a matched chain between each material statement and its cited work, including dates, cohorts, jurisdictions, and accessible evidence level. Numerical QA independently reproduces the calculations. The final passport should record the assumptions under which the findings hold, rather than reporting a single percentage as a general measure of longevity hedging quality.
Critical Discussion
The most common oversimplification is to treat annuitant mortality as a uniformly scaled version of national population mortality. A scaling factor can be a practical approximation when supported by actual portfolio data, yet different purchase ages, socioeconomic compositions, and calendar periods can change it. The OECD's cross-country discussion of selection is explicit that selection factors vary by product and jurisdiction. Any transfer across countries therefore includes judgment that cannot be eliminated by substituting one published national table for another.
A second simplification is to equate a good historical fit with a sound economic hedge. A joint model may track past mortality levels while understating the uncertainty of the difference between two populations. A liability hedge may also reference a limited range of ages even though its payments span several decades. Cross-population forecast quality, cash-flow weight, and model-recalibration robustness must be evaluated together.
A third error is to interpret differences in data-covered populations as differences in biological survival without examining the record-generation process. Retirement schemes and payout annuity providers record different benefit transitions, and voluntary product ownership is associated with economic selection. Benefit-weighted exposure adds a financial relationship absent from ordinary headcount mortality. The appropriate answer depends on the liability being valued.
A fourth simplification is to treat public data summaries as complete empirical access. Recent SOA and CMI investigations establish valuable research programs and data dimensions, but detailed paid or authorized-user material was not independently inspected here. It would be methodologically unsound to infer subgroup mortality coefficients or hedge correlations from their public descriptive summaries. The synthetic examples in this monograph use disclosed assumptions precisely because actual joined life-level data are unavailable.
Finally, risk reduction is not equivalent to risk disappearance. Even an accurately priced and collateralized index hedge can leave population-specific longevity risk, data revisions, and contract mismatches. A tailored transfer can reduce some of these differences while creating other costs and risks. The most defensible decision criterion is the distribution of financial outcomes after all relevant components and costs have been recognized.
Original Analytical Synthesis
The evidence supports a precise institutional conclusion. The mortality basis relevant to an annuity is the mortality experience of the beneficiaries to whom the payment promises apply, weighted by the contingent benefits and their timing. A mortality table from another population is a proxy whose usefulness is established by measured transferability. This conclusion follows directly from the life-contingent cash-flow equation and from the observed selection and heterogeneity documented in official retirement mortality research.
The first analytical implication is that population fit and liability fit are different tasks. An index can track the average death rate of its reference population accurately and nevertheless have an inappropriate age- or benefit-weighted relationship with the liability. A robust research design therefore links mortality parameters to the payment schedule before estimating hedge effectiveness.
The second implication is that the useful unit for comparing different mortality bases is the residual cash-flow distribution. An average life-expectancy difference explains some economic exposure but does not capture its horizon, age-dependent response, or correlation with an index instrument. The worked example demonstrates valuation sensitivity to alternative mortality paths; the variance example demonstrates that high but imperfect correlation leaves a measurable residual. These results are straightforward deductions from established actuarial and statistical mathematics, not claims of a new model.
The third implication concerns institutional data. A broad national statistical population, a private occupational pension dataset, and a selected individual annuitant dataset can each be internally valid. Their relationship is contingent on membership and record construction. An explicit source-to-population map is therefore part of scientific reproducibility, especially when longitudinal experience covers changing product and pension markets.
The fourth implication is temporal. Current mortality level and future mortality improvement cannot be governed separately without testing consistency. Pandemic-era observations, shifting socioeconomic composition, and revised projections can alter both the base and its projected change. A stable historical difference provides evidence about the past; a hedge decision concerns the future joint distribution.
The fifth implication is contractual. A population reference becomes an enforceable hedge settlement only when the contract defines observable events, timing, revisions, and contingencies. Mortality basis risk then has a measurable legal and operational dimension alongside its demographic and statistical dimensions. Sound reinsurance and derivative design links all of these elements to the same life-contingent obligation.
Together these conclusions define an evidence-based research practice for comparing retirement mortality populations. Their contribution is the integrated explanation of how demographic differences, actuarial valuation, index construction, and regulatory assessment meet in the same financial liability. No separate new theory or doctrine is asserted.
Conclusion
Retirement annuity mortality basis risk arises whenever the population whose survival determines payments differs from the population whose mortality data determine estimates, projections, prices, or hedge settlements. The difference can concern membership selection, age and birth cohort, occupational and socioeconomic mix, benefit weighting, improvement trends, data periods, record quality, and contingent contractual rights. Each dimension can modify expected cash flows and the distribution of unexpected loss.
The strongest evidence for this interpretation comes from the combination of OECD comparative mortality research, the SOA's distinct annuity and pension experience programs, the CMI's population-specific investigations, foundational stochastic mortality studies, and academic analyses of index-based longevity hedging. Those sources support the possibility of effective population-referenced hedges when relationships are appropriately measured. They also establish the need for explicit calibration, validation, and recognition of residual exposure.
The monetary value of a lifetime payout depends on survival over the full contractual payment schedule. Consequently, similar current mortality rates or high historical population correlations are insufficient evidence of interchangeable annuity liabilities. The precise target is the benefit-weighted cash-flow process under relevant future mortality states. A useful mortality reference is one whose transfer to that target has been justified and whose error has been made visible.
The practical scientific conclusion is to specify the insured population and benefit contract first, then select, adjust, and jointly validate mortality evidence before pricing or hedging. Annuity mortality basis risk becomes assessable through transparent exposure definitions, reproducible calculations, tested cross-population relationships, and contractually defined settlement. The remaining uncertainty is an ordinary part of actuarial judgment, capital assessment, and risk transfer.
Reproducibility Appendix
The following algorithm regenerates the 40-year annuity factors in the synthetic example. All amounts and assumptions are invented for the demonstration. There is no random-number generator and no external dataset. Here is the index of an annual payment date after issue and is the corresponding annual mortality interval.
def annuity_factor(q65, age_progression, discount=0.03,
number_of_payments=40, mortality_scale=1.0):
survival = 1.0
present_value = 0.0
for t in range(1, number_of_payments + 1):
q = q65 * ((1.0 + age_progression) ** (t - 1))
q = min(0.95, q * mortality_scale)
survival *= (1.0 - q)
present_value += survival / ((1.0 + discount) ** t)
return present_value
target = annuity_factor(0.012, 0.077)
reference = annuity_factor(0.016, 0.070)
pension = annuity_factor(0.014, 0.073)
target_low_mortality = annuity_factor(0.012, 0.077,
mortality_scale=0.90)
reference_low_mortality = annuity_factor(0.016, 0.070,
mortality_scale=0.90)
print(target, reference, pension,
target_low_mortality, reference_low_mortality)
Expected results before rounding: approximately 14.85130096185526, 13.85501253844253, 14.34374051730801, 15.35127835426044, and 14.38632543485190. Values can be reproduced in a standard Python interpreter. The upper bound of 0.95 on annual death probability is included to keep the routine well defined under other input choices; it is not binding in these five evaluated scenarios.
For the hedge illustration, set , , and . Compute and residual variance . These yield 1.0818181818 and 54.39, respectively. The fraction of risk variance reduced is . None of these inputs are inferred from a mortality experience study.
Reproduction of a real portfolio study would require an independently authorized and privacy-preserving exposure dataset, specific benefit terms, an observation definition, an appropriate actuarial life table and calendar-year improvement model, and the exact hedge settlement data. Such an empirical replication is outside the evidence accessed for this manuscript.
References
Actuaries Institute (Australia). 2024. “Release of Australian Life Table 2020–22.” https://www.actuaries.asn.au/research-analysis/release-of-australian-life-table-2020-22
Australian Government Actuary. 2024. Australian Life Tables 2020–22. Released December 12, 2024. https://aga.gov.au/publications/life-tables/australian-life-tables-2020-22
Australian Government Actuary. n.d. Australian Life Tables. Official series and subgroup investigations. https://aga.gov.au/publications/life-tables
Cairns, Andrew J. G. 2013. “Robust Hedging of Longevity Risk.” Journal of Risk and Insurance 80 (3): 621–648. https://doi.org/10.1111/j.1539-6975.2013.01525.x
Cairns, Andrew J. G., David Blake, and Kevin Dowd. 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
Cairns, Andrew J. G., David Blake, Kevin Dowd, Guy D. Coughlan, David Epstein, and Marwa Khalaf-Allah. 2011. “Mortality Density Forecasts: An Analysis of Six Stochastic Mortality Models.” Insurance: Mathematics and Economics 48 (3): 355–367. https://ideas.repec.org/a/eee/insuma/v48y2011i3p355-367.html
Coughlan, Guy D., Marwa Khalaf-Allah, Yijing Ye, Sumit Kumar, Andrew J. G. Cairns, David Blake, and Kevin Dowd. 2011. “Longevity Hedging 101: A Framework for Longevity Basis Risk Analysis and Hedge Effectiveness.” North American Actuarial Journal 15 (2): 150–176. https://doi.org/10.1080/10920277.2011.10597615
Dowd, Kevin, Andrew J. G. Cairns, David Blake, Guy D. Coughlan, David Epstein, and Marwa Khalaf-Allah. 2010. “Evaluating the Goodness of Fit of Stochastic Mortality Models.” Insurance: Mathematics and Economics 47 (3): 255–265. https://doi.org/10.1016/j.insmatheco.2010.06.006
European Insurance and Occupational Pensions Authority. 2024. “Longevity Risk Sub-module.” Solvency II Single Rulebook, Delegated Regulation (EU) 2015/35, Article 138. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-5758_en
IFRS Foundation. 2017 (as amended). IFRS 17 Insurance Contracts. Issued May 2017; effective for annual reporting periods beginning on or after January 1, 2023. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/
Institute and Faculty of Actuaries, Continuous Mortality Investigation. 2026. “CMI Working Paper 209: Mortality Experience of Pensioners for the Period 2017 to 2024.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/self-administered-pension-scheme-mortality/cmi-working-paper-209
Institute and Faculty of Actuaries, Continuous Mortality Investigation. 2026. “CMI Working Paper 211: CMI Mortality Projections Model, CMI_2025.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/mortality-projections/cmi-working-paper-211
Institute and Faculty of Actuaries, Continuous Mortality Investigation. 2026. “CMI Working Paper 214: Interaction between Mortality Base Tables and Improvement Assumptions.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-working-papers/mortality-projections/cmi-working-paper-214
Institute and Faculty of Actuaries, Continuous Mortality Investigation. 2026. “CMI News: Annuitant Mortality, Working Paper 210.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-news
Institute and Faculty of Actuaries, Continuous Mortality Investigation. n.d. “Annuities Investigation.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-investigations/annuities-investigation
Institute and Faculty of Actuaries, Continuous Mortality Investigation. n.d. “Self-administered Pension Scheme (SAPS) Mortality Investigation.” https://www.actuaries.org.uk/learn-and-develop/continuous-mortality-investigation/cmi-investigations/self-administered-pension-scheme-saps-mortality-investigation
Institute and Faculty of Actuaries. 2026. “Pensioners under the Microscope: Zooming in on the Mortality of Bulk Annuitants and Private Sector Pensioners.” Actuaries Blog. https://blog.actuaries.org.uk/pensioners-under-the-microscope/
International Association of Insurance Supervisors. 2024. Insurance Core Principles and Common Framework (ComFrame), updated December 2024, online standards tool. https://www.iais.org/icp-online-tool/
Lee, Ronald D., and Lawrence R. Carter. 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
Li, Nan, and Ronald Lee. 2005. “Coherent Mortality Forecasts for a Group of Populations: An Extension of the Lee–Carter Method.” Demography 42 (3): 575–594. https://pmc.ncbi.nlm.nih.gov/articles/PMC1356525/
Ngai, Andrew, and Michael Sherris. 2011. “Longevity Risk Management for Life and Variable Annuities: The Effectiveness of Static Hedging Using Longevity Bonds and Derivatives.” Insurance: Mathematics and Economics 49 (1): 100–114. https://doi.org/10.1016/j.insmatheco.2011.02.009
OECD. 2023. Mortality and the Provision of Retirement Income. Paris: OECD Publishing, February 2, 2023. https://www.oecd.org/en/publications/mortality-and-the-provision-of-retirement-income_a10a6c09-en.html
OECD. 2023. “Modelling Choices for the Development of Standard Mortality Tables.” In Mortality and the Provision of Retirement Income. https://www.oecd.org/en/publications/mortality-and-the-provision-of-retirement-income_a10a6c09-en/full-report/component-5.html
OECD. 2023. “Global Approaches to Developing Standard Mortality Tables for the Provision of Retirement Income.” In Mortality and the Provision of Retirement Income. https://www.oecd.org/en/publications/mortality-and-the-provision-of-retirement-income_a10a6c09-en/full-report/component-6.html
Royal Dutch Actuarial Association (Koninklijk Actuarieel Genootschap). 2024. “Prognosetafel AG2024.” https://www.actuarieelgenootschap.nl/actueel/koninklijk-actuarieel-genootschap-presenteert-prognosetafel-ag2024
Society of Actuaries. 2019. Pri-2012 Private Retirement Plans Mortality Tables. Retirement Plans Experience Committee, report and mortality tables. https://www.soa.org/resources/experience-studies/2019/pri-2012-private-mortality-tables/
Society of Actuaries Research Institute. 2021. “2017 Individual Life Insurance Mortality Experience Report.” https://www.soa.org/resources/research-reports/2021/2017-mortality-experience/
Society of Actuaries Research Institute and LIMRA. 2026. 2020–2024 Individual Payout Annuity Mortality Experience Study. Public study overview and restricted detailed data package. https://www.soa.org/resources/experience-studies/2026/2020-24-individual-payout/
Society of Actuaries Research Institute and LIMRA. 2026. 2015–2024 Deferred Annuity Mortality Experience Study. Public study overview and restricted detailed data package. https://www.soa.org/resources/experience-studies/2026/2015-24-deferred-annuity-mort/
United States Social Security Administration. 2026. The 2026 Annual Report of the Board of Trustees of the Federal Old-Age and Survivors Insurance and Federal Disability Insurance Trust Funds. https://www.ssa.gov/oact/TR/2026/