Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Annuitant Mortality Selection and Purchase Timing

Abstract. Annuitant mortality selection is the relationship between a life-contingent retirement-income purchasing decision and the subsequent survival distribution of the people who make that decision. Purchase timing determines who can enter an insured group, the age and economic conditions at entry, the contractual survival period, and the statistical exposure that the insurer can subsequently observe. The subject therefore connects insurance economics, survival analysis, actuarial valuation, retirement policy, consumer behavior, and the empirical study of information asymmetry. This Research Dossier examines the original selection literature, evidence from the United Kingdom, the United States and Chile, recent institutional experience programs, and the cash-flow consequences of selected mortality. It analyzes the distinct mechanisms of entry selection, sorting among annuity guarantees, delayed purchase, survival conditioning and changing mortality composition by time since issue. The analysis also distinguishes the selection of initially healthier purchasers from a possible effect of assured lifetime income on later health and survival. Explicit mathematical identities and synthetic examples make the valuation and duration implications reproducible. Its primary conclusion is that mortality selection is an empirically meaningful quantity only after the relevant eligible population, annuity contract, entry date, observation period, risk classification and weighting convention have been defined. The research synthesizes established theory and published evidence without proposing a new proprietary theoretical system or claiming an original causal estimate.

Keywords: Annuitant mortality; life annuities; adverse selection; annuitization; retirement income; purchase timing; survival analysis; mortality heterogeneity; contract design; duration effects; delayed entry; immortal-time bias; enhanced annuities; longevity risk; actuarial valuation; pension economics; mortality improvement; money's worth.

Introduction

A purchaser of a life annuity exchanges capital for payments governed by a defined survival condition, subject to the terms of any guaranteed period, survivor provision, or refund. The insurer thereby assumes an obligation whose expected duration depends on the actual lives who purchase the product. Selection arises when these lives have different future survival from the population used as the reference for pricing or research. The decision may reflect private health information, expectations of longevity, risk tolerance, financial resources, pension rules, access to advice, or a choice among available guarantees. Purchase time adds the further condition that a person must remain alive until the chosen transaction date.

The uncertain-lifetime framework of Yaari (1965) provides an economic foundation for consumption and wealth decisions under uncertain survival. The annuity-welfare analysis of Davidoff, Brown, and Diamond (2005) extends the analysis across more general financial-market assumptions. Those contributions illuminate the value of transferring individual longevity risk but do not themselves establish the mortality experience of any particular insurer's actual customers. That question requires evidence about the people who enter the market, their contracts, and the dates at which they become exposed.

This monograph concerns the boundary between the decision to buy an annuity and the mortality distribution subsequently observed. It studies voluntary and institutionally shaped purchases, the sorting associated with payment schedules and estate guarantees, the effect of deferral, and the interpretation of mortality by duration. Its comparison includes historical UK individual annuities, US individual payout-annuity experience studies, and Chilean pension markets, with attention to the source jurisdiction and period of each finding. General pension design, annuity mathematics and longevity modeling provide essential foundations through the related Science areas Annuities and Retirement Insurance, Actuarial Science, and Insurance Economics.

The method combines documentary criticism, original historical and international comparison, established life-contingent valuation identities, and transparent numerical illustrations. Empirical claims are tied to original journal publications, research-center studies or official statistical programs. Where only publisher abstracts and study summaries are publicly available, the inference is limited to what those documents support. No insurer's confidential mortality microdata were available, and the synthetic calculations are explicitly separated from measured market results.

Scientific Foundations, History, and Research Identity

The scientific object and the comparison population

For a stated age, calendar date and set of observable characteristics, annuitant mortality selection denotes the difference between survival among people who purchase a specified annuity and survival among an explicit comparison group. Depending on the research objective, the comparison can be the general population, people eligible to buy an annuity, defined-contribution pension participants, retirees with sufficient accumulations, or holders of other contract forms. These are different denominators. Selection relative to the general population may reflect occupational and wealth differences that were already present before the annuity purchase was contemplated.

A mortality distribution may be described by annual conditional death probabilities q_x, survival probabilities {}_tp_x, a force of mortality, or an age-specific mortality rate estimated from deaths and exposure. A life-years-weighted estimate is different from a benefit-amount-weighted estimate. The latter gives greater influence to contracts paying larger amounts, which may have different purchaser characteristics. Likewise, one person may own multiple contracts; the count of policies need not equal the count of distinct lives. Every statement about an annuitant sample therefore requires the exposure denominator and weighting rule.

The economic meaning of adverse selection is relative to the risk carrier. In a standard life annuity, longer-than-assumed survival usually raises expected payments and can make the pricing mortality assumption unfavorable to the insurer. In death-benefit insurance, different mortality directions affect expected claims. In medically underwritten enhanced annuities, shorter expected survival can justify a greater income rate for a given premium. The term selection describes conditional population composition; the evaluation of whether its financial result is adverse depends on the contract and the initial price.

Four distinct dates

The date a consumer elects or applies for a contract, the date at which premium acceptance and legal coverage become effective, the promised commencement date of lifetime income, and the date of an individual installment may all differ. An immediate annuity generally begins income soon after issue. A deferred income annuity may be purchased years before its survival-contingent income begins. A payment-in-advance contract shifts the first benefit to the beginning of the applicable payment interval; a payment-in-arrears contract conditions later installments on entitlement at the payment date. The NAIC's annuity classification distinguishes immediate and deferred structures, while MoneyHelper's annuity guidance documents UK payment choices and enhanced-income arrangements.

An original contract must be read before it is mapped onto a survival-state formula. Guaranteed installments payable after death are cash-flow obligations even though the original life has died. A return-of-premium rider may create a beneficiary claim and an associated death-triggered payment. A joint-life contract may continue in a reduced amount after the first life dies. A deferred annuity may provide a premium refund if the purchaser dies before commencement. These provisions alter the relationship between the insurer's observed biological deaths and its outstanding financial obligations.

Historical literature and evolving institutions

The market-based money's-worth study by Mitchell and colleagues (1999) illustrates how the present value of promised annuity payments changes with the mortality basis chosen by the analyst. The retirement-demand analysis by Brown connects annuitization decisions to health and household characteristics. Together they show why demand and actuarial valuation must be studied in parallel: the people who find lifetime payments attractive may have a systematically different survival distribution from those used to calculate a population average.

The empirical turning point in the historical UK literature was the distinction between voluntary and then-compulsory annuity markets. In their 2002 UK selection study, Finkelstein and Poterba found longer average survival among annuitants than among nonannuitants and a stronger selection pattern in their voluntary market sample. They estimated that the adverse-selection magnitude in the historically compulsory segment was about half that in the voluntary segment. This result belongs to the observed UK institutions and period; later pension reforms can change eligibility and the choice set. It cannot be treated as a universal contemporary ratio between compulsory and voluntary annuitization.

Their later policyholder evidence from 2004 compared mortality across annuity contract features, finding patterns associated with payment timing and estate provisions, while the mortality association with policy size was less significant. This is important because private information can be expressed through choices other than the amount of insurance. Contract-level underwriting and comparative research must preserve these product characteristics rather than treating all purchasers at the same age as a homogeneous group.

A different institutional environment appears in the Chilean retrospective evidence, where public pension provisions, information, and characteristics of retirees influenced the choice of payout. The example shows that the population from which annuitants emerge is constructed partly by retirement-system rules. Changing the available state pension or programmed withdrawal alternative may change annuity demand and subsequent observed mortality without an immediate biological change in the population.

Specialized mortality-experience programs developed alongside this literature. The SOA individual annuity studies register records successive US studies of payout-annuity mortality and related contract-owner behavior. The CMI annuities investigation describes distinct UK programs and product subsets. These data sources have different exposures, data contributors and definitions; they must be compared through their methodology rather than by importing one crude aggregate into another jurisdiction's pricing system.

Entry selection, contract selection, and survivor selection

Three processes are routinely conflated. Entry selection arises because some eligible people buy an annuity and others do not. Contract selection arises because buyers choose different income guarantees and payment forms, potentially revealing preferences and private longevity information. Survivor selection develops after issue, as lives with higher mortality tend to leave the surviving population more quickly. The third process can occur with no change to individual health or hazard; it follows from conditional survival within a heterogeneous group.

The mechanisms have different empirical requirements. Entry selection calls for a matched population of eligible purchasers and nonpurchasers. Contract selection requires detailed benefit elections, guarantee forms, and quoted prices. Survivor selection requires consistent issue-date cohorts, individual exposures, attained ages and observed durations. A study with only aggregate insurer death counts cannot uniquely identify all three. Nor can a lower observed mortality rate after annuitization automatically establish that the act of buying the annuity increased longevity.

Analytical boundaries and professional frameworks

The comparative analysis recognizes that the FCA's 2024 retirement-income review addresses UK advice and customer outcomes, the Solvency II Article 77 rule governs specified EU technical provisions, and IFRS 17 insurance accounting establishes accounting measurement and presentation for contracts in its scope. None prescribes the same market mortality rate for every policyholder group. Their proper use begins by distinguishing what an insurer is legally obliged to pay from what it expects to pay, the risk margin it must recognize and the price it offers before the contract is signed.

Annuity Purchase Selection

Choice under uncertain lifetime

The valuation of annuitization to a consumer depends on the interaction of lifetime income security, liquid assets, pre-existing pensions, uncertain health expenditure, bequest preferences and other risks. The Davidoff–Brown–Diamond welfare results are valuable because their theoretical conclusions specify assumptions about incomplete markets and resource allocation. They do not establish that every retiree should buy a particular annuity or that nonbuyers have a known mortality disadvantage. Observed demand is a consequence of preferences and opportunities as well as health.

Let A_t denote purchase of an annuity at time t, C the chosen contract, X observable risk and economic characteristics, and Z unobserved attributes affecting survival or preferences. The distribution of latent characteristics conditional on purchase is

f(z∣At=1,C,X)=Pr⁡(At=1,C∣z,X)f(z∣X)∫Pr⁡(At=1,C∣u,X)f(u∣X) du.f(z\mid A_t=1,C,X) =\frac{\Pr(A_t=1,C\mid z,X)f(z\mid X)} {\int \Pr(A_t=1,C\mid u,X)f(u\mid X)\,du}.

This is a conditional-probability identity, not a newly proposed economic model. It shows that a purchaser sample is the eligible distribution reweighted by the likelihood of choosing a contract. Even with precise mortality data on buyers, the original distribution of Z and the selection propensity are not separately identifiable without evidence about the eligible population, prices or other exclusion assumptions.

A healthier purchasing sample can reflect higher subjective life expectancy, access to retirement capital, systematic differences in education or occupation, insurer medical underwriting or a distribution channel that reaches healthier households. The observation alone does not establish superior private information. Similarly, a purchaser with shorter expected life may value period-certain benefits or dependent survivor income, so the relationship between health and any single annuity attribute is sensitive to available alternatives.

Contract menus as information channels

Product form is a potential sorting variable. In a life-only single-premium immediate annuity, a death soon after purchase ordinarily ends the remaining life-contingent stream. A period-certain guarantee preserves payments to designated recipients during a specified period, regardless of early death. A cash-refund or installment-refund contract protects a different part of the transferred premium. Joint-life products depend on a second covered life and the continuation ratio. None of these forms can be valued properly without the original entitlement and beneficiary wording.

Suppose buyers believe they have different prospects of early death. All else equal, a purchaser expecting shorter life may assign greater value to payments continuing to an estate, while a purchaser expecting long life may prefer a larger purely life-contingent installment. Actual choices can also be driven by bequest motives and family status. The 2004 UK contract-characteristics evidence is consistent with selection along these features but does not permit an unconditional causal reading of every guarantee choice. The insurer must distinguish a feature's effect on the contractual payments from its correlation with the mortality of those choosing it.

The value of an optional guarantee changes when the insurer changes the amount of lifelong income offered. A higher guarantee often entails a lower annual life-contingent amount for the same premium; policyholders choose combinations of contractual protection and price. A credible selection analysis therefore uses the offer set and quoted relative prices. Comparing guarantee purchasers and life-only purchasers without accounting for the price they faced can produce a misleading interpretation of the disclosed mortality relationship.

Voluntary and institutionally required purchases

Voluntary annuitization means that the decision to enter the market is itself selective. A historical requirement to purchase an annuity may draw in people who would have chosen investments or programmed withdrawals, which tends to dilute some entry-level self-selection. The UK voluntary–compulsory comparison supports that mechanism in the examined historical data. Yet mandatory purchase does not eliminate all choice: the annuity provider, starting age, guarantee, payout schedule and survivor benefit may remain open, and their relative attractiveness can still depend on private information.

The pricing and redistribution of these features are also endogenous to regulation. Finkelstein, Poterba, and Rothschild (2009) studied retirement-annuity market responses to restrictions on gender-based pricing, emphasizing that menu and sorting responses matter for incidence. One cannot infer the distributional effect of an annuity-rate rule merely by changing a single demographic rate while keeping every consumer choice fixed.

In Chile, the coexistence of private annuitization options and public pension protections changes the choice environment. The Cox Edwards–James study provides an institutional example of how public benefits can crowd out demand and how information and mortality expectations enter the decision. It would be inappropriate to transfer a resulting annuitant mortality level directly to a UK or US retail insurer without reconstructing the corresponding pool of eligible retirement savers.

Health underwriting and enhanced annuities

A standard annuity may use broad mortality classes, while an enhanced annuity can provide higher income to a person whose medical or lifestyle circumstances imply shorter expected survival. The public UK enhanced-annuity explanation describes medical conditions and smoking as relevant to quotations. When enhanced products are offered, individuals with material impairments may shift away from the standard segment. Its apparent favorable mortality can thereby intensify because of both voluntary sorting and underwriting reclassification.

A dataset recording only standard-rate payout contracts can underrepresent the medically impaired part of the original eligible pool. Conversely, an aggregate insurer study mixing enhanced and standard contracts without risk classification may overstate or obscure differences across guarantee forms. The key observable is the whole journey from eligibility through disclosures and quotations to chosen benefit. A declined applicant, someone never offered a competitive quote, and an eligible retiree who actively chose drawdown are distinct nonpurchaser categories.

Information, wealth, advice and distribution

An annuity purchase requires sufficient available capital, access to a product and understanding of a complex tradeoff. Retirees with very low wealth or large pre-existing guaranteed pension benefits may have little marginal reason to purchase additional lifetime income even if they expect a long life. Household structure changes demand for survivor protection, and health expenditure uncertainty creates demand for liquid savings. These economic mechanisms are compatible with the Brown annuitization study and the broader annuity-welfare literature, while their strength must be estimated in the relevant population.

Advice can influence both whether and when an annuity is bought. A purchaser comparing providers may reveal medical conditions to obtain an enhanced rate, choose a partial annuity to preserve liquidity, or defer purchase until a pension entitlement becomes available. The FCA retirement-income thematic review concerns the quality of such consumer decision-making in its UK regulatory context. For a statistical study, advice and availability of options form part of the selection process; ignoring the distribution process can make differences in purchases look like differences in private mortality information.

A rigorous observational design begins at the point when consumers are eligible to purchase. It records nonpurchasers, incomplete applications, accepted and refused quotations, purchase dates and product decisions. If those records are unavailable, inferences must be explicitly limited to comparisons across observed buyers or to a benchmark whose differences are not fully attributable to adverse selection.

Duration Effects and Self-Selection

Purchase timing and annuitization as a stopping decision

Purchase timing is the choice of when to commit assets to the annuity, while commencement is the contractual date on which life-contingent income begins. The choice of purchase date reflects changing market yields, quoted income, attained age, pension access, expected health, accumulated assets and alternative investment opportunities. The Stabile (2006) optimal-timing study treats annuitization as a combined stochastic-control and stopping problem, demonstrating that, under its preference and constant-force mortality assumptions, immediate purchase, deferral or no purchase can be optimal in different parameter regions. Its result establishes the relevance of timing, not a universal retail instruction.

The earlier a risk set begins, the more strongly its composition is affected by decisions made after that date. A person who buys an annuity at age 70 had to survive from age 65 to make that purchase at 70. Someone purchasing deferred income at 65 may enter an insurance obligation immediately but have no life-contingent payout due until age 70. These individuals can have the same first payment age while presenting different issue-date risk and different value transfers in the event of early death.

The observation file must record eligibility, application, quote, premium settlement, legal issue, income commencement, first installment and death. A mortality study beginning only with the first installment can exclude all pre-commencement deaths among people who previously paid for deferred annuities. If those omitted deaths affected the original insurer risk, the resulting payer-cohort mortality does not describe mortality from the contract's issue date.

Survival conditioning and delayed entry

Let q_x denote an annual conditional probability of death at attained age x on a specified mortality basis. Survival for m integer years is

mpx=∏j=0m−1(1−qx+j).{}_mp_x=\prod_{j=0}^{m-1}(1-q_{x+j}).

A person who buys only after m years must satisfy the survival condition. Comparing eventual late purchasers with earlier eligible persons as if everyone had been an annuitant since the common initial date introduces immortal-time bias: the later purchasers had to remain alive during the period before the event defining their cohort. The correct risk sets use delayed entry for issue-date annuitant mortality and a separate eligible-population cohort for the purchase hazard.

For unit payments made each year in arrears after an m-year deferral, a simplified deferred life-annuity factor valued at initial age x is

ax(m)=∑k=m+1ωvk kpx,v=(1+i)−1.a_x^{(m)}=\sum_{k=m+1}^{\omega}v^k\,{}_kp_x, \qquad v=(1+i)^{-1}.

If a new immediate annuity is bought at age x+m, its unit value conditional on survival to that date is

ax+m=∑j=1ω−mvj jpx+m.a_{x+m}=\sum_{j=1}^{\omega-m}v^j\,{}_jp_{x+m}.

Under the same mortality and interest assumptions, and with no additional pre-commencement death benefit, these satisfy

ax(m)=vm mpx ax+m.a_x^{(m)}=v^m\,{}_mp_x\,a_{x+m}.

The identity follows from factorization of the same conditional survival curve. It is a valuation relation, not an assertion that the two contracts have identical premiums, guarantees, underwriting rules or legal obligations. A refund of premium following pre-commencement death would add a distinct expected cash-flow term to the left-hand side. A purchase made only after survival can be financed at a different interest rate and in different market conditions.

Mixture selection by elapsed duration

Even with individual mortality rates held fixed, surviving annuitants can become a progressively different mixture of risks. Suppose the purchase cohort includes a low-mortality class L and a higher-mortality class H, with initial shares w_L and 1-w_L and constant one-year conditional death probabilities q_L and q_H. After d complete years, the share of L among survivors is

wL(d)=wL(1−qL)dwL(1−qL)d+(1−wL)(1−qH)d.w_L(d)= \frac{w_L(1-q_L)^d} {w_L(1-q_L)^d+(1-w_L)(1-q_H)^d}.

Its next-year aggregate death probability is

q(d)=wL(d)qL+[1−wL(d)]qH.q(d)=w_L(d)q_L+[1-w_L(d)]q_H.

For q_L below q_H, the low-mortality type grows as a share of survivors. A change in the crude aggregate therefore arises without any individual change in health. In a real portfolio, individuals also grow older, cohorts encounter different calendar-year conditions, and mortality can improve or deteriorate. Those age and time influences can outweigh or mask the selection-mixture effect.

This mechanism demonstrates why the phrase duration effect needs definition. A first-policy-year mortality rate can differ from a tenth-policy-year rate because of survival composition, underwriting at purchase, distribution channel, lapse or portability, the age mix of issue cohorts, reporting lag or differing contractual dates. It does not automatically measure how an individual's force of mortality changes because a contract has been held for ten years. A select-and-ultimate mortality table may represent genuine statistical patterns by duration, but their causal interpretation requires more information.

Calendar year, attained age, birth cohort and issue cohort

Age–period–cohort structures arise naturally in mortality research. Purchase adds another dimension, issue cohort, because consumers can enter under different products, economic yields and underwriting rules. An observation in a given calendar year simultaneously belongs to an attained age, a birth cohort, a duration since purchase and a contractual product cohort. Any two aggregate ratios may differ because the weighting across these dimensions differs.

A practical exposure table therefore segments attained age, sex or another scientifically relevant and lawful rating classification, issue year, duration, benefit type, underwriting class and calendar period. Life-years or appropriate central exposure are calculated with dates of entry, exit and death. Broad calendar shocks, especially in the 2020–2024 interval, require special attention. The recent SOA 2020–2024 payout-annuity study and the earlier 2014–2019 payout study cover different calendar environments and participating portfolios. A crude comparison of their death totals cannot isolate post-issue select mortality.

Duration and contractual guarantees

Annuity benefit timing affects expected settlement costs independently of selection. For equal annual life-contingent installments and a consistent mortality and discount basis, a unit immediate annuity payable in advance has factor

a¨x=1+∑k=1ωvk kpx=1+ax,\ddot a_x=1+\sum_{k=1}^{\omega}v^k\,{}_kp_x=1+a_x,

while one payable in arrears has factor a_x. The identity presumes the same fixed unit payment schedule and that an immediate payment at issue is due with certainty. Contract provisions such as a certain period, premium refund, joint survivor payment or income escalation require their own cash-flow modeling.

A K-installment certain-and-life annuity with annual payments in arrears can be represented by

ax:K‾ guaranteed=∑k=1Kvk+∑k=K+1ωvk kpx.a_{x:\overline K}^{\,\mathrm{guaranteed}} =\sum_{k=1}^{K}v^k+\sum_{k=K+1}^{\omega}v^k\,{}_kp_x.

This assumes each of the first K installments is guaranteed to the appropriate entitled payee, with further payments only while the named original life is alive. Contract-specific death notification and settlement terms still matter. If the guaranteed form attracts a different mortality group from life-only business, price analysis requires both the different payoff structure and the different selected survival curve. Treating those as one undifferentiated mortality change would misattribute the source of the financial effect.

Mortality Mispricing Implications

The contingent payment promise

For a single-life level annuity that pays an annual benefit B at the end of each complete year while the insured life survives, a simplified issue-date liability value is

Vx=B∑k=1ωvk kpx,V_x=B\sum_{k=1}^{\omega}v_k\,{}_kp_x,

where v_k represents a discount factor to the kth benefit date and {}_kp_x is survival to that date on the assumed mortality basis. The amount and timing of every payment are defined by the contract. If survival probabilities have been understated, the estimated present value of life-contingent payments is generally too low for positive benefits and discount factors. If mortality has been assumed too high for the actual purchasing pool, an income quote determined using that mortality basis may promise more annual payment than the initial premium can support under the corrected assumption.

The sign of the effect reverses for some death-triggered obligations and can change after contractual guarantees are added. A certain-period guarantee reduces the dependence of early payments on survival. A cash-refund annuity creates an estate payment in specified states. A joint-life continuation contract depends on two survival processes, their dependence, and the benefit reduction, if any, after a first death. An enhanced annuity may price an identified impaired-life risk deliberately and offer more income for the same premium. The insurer's challenge is not to assume that all annuitants have unusually low mortality, but to identify the conditional survival and payment obligations belonging to each contract class.

The 2012 Individual Annuity Reserving Report documents specialized US payout annuity mortality work and illustrates why annuitant assumptions can be distinct from general population tables. It is an institutional mortality-basis resource for valuation in its regulatory and historical context. A pricing actuary must separately consider current credible experience, prospective mortality improvements, the present product's selection conditions and relevant statutory requirements.

Synthetic illustration: selection at entry and lifetime payout value

Consider a deliberately synthetic eligible population at one attained age. Half are assumed to have constant annual death probability q_L=0.02 and half q_H=0.05. This population's initial weighted one-year death probability is 0.5(0.02)+0.5(0.05)=0.035. Suppose the purchasers of a life-only annuity are 70% low-mortality type and 30% high-mortality type. Their initial weighted death probability is 0.7(0.02)+0.3(0.05)=0.029. The 0.6-percentage-point difference comes entirely from the assumed participation mix.

To isolate basis sensitivity from individual heterogeneity, now calculate annuity factors for two homogeneous constant-q approximations, one calibrated at q=0.035 and the other at q=0.029. Let the effective annual interest rate be 3%, benefits be one unit annually in arrears, death occur under independent annual survival transitions, and no terminal maximum payment age be imposed. The geometric-series annuity factor is

a(q)=∑k=1∞(1−q1.03)k=1−q0.03+q.a(q)=\sum_{k=1}^{\infty}\left(\frac{1-q}{1.03}\right)^k =\frac{1-q}{0.03+q}.

The reference approximation gives a(0.035)=0.965/0.065=14.846153846. The purchaser approximation gives a(0.029)=0.971/0.059=16.457627119. This is approximately 10.85% larger. A premium equal to the lower factor would be inadequate to support one unit per year under the second homogeneous approximation, before expenses or profit allowances.

The use of a single average q is a teaching simplification. The exact expected payout cost under the genuine two-type mixture is 0.7a(0.02)+0.3a(0.05), rather than a(0.029), because the valuation is nonlinear in q. Computing the exact mixture gives 0.7×(0.98/0.05)+0.3×(0.95/0.08)=0.7×19.6+0.3×11.875=17.2825. For the reference 50/50 mixture, the exact factor is 0.5×19.6+0.5×11.875=15.7375. The true illustrative increase for these explicitly mixed risk classes is 17.2825/15.7375−1, or approximately 9.82%. These two sets of values demonstrate an important model-risk issue: inserting average annual mortality into a nonlinear lifetime annuity formula is not equivalent to valuing a heterogeneous mortality mixture. The selection effect on cash-flow value should be calculated over the survival functions of the actual classes, not just from a one-year average q.

The same synthetic example makes duration selection observable. After five full years the low-mortality share among survivors is

wL(5)=0.7(0.98)50.7(0.98)5+0.3(0.95)5≈0.7316.w_L(5)= \frac{0.7(0.98)^5} {0.7(0.98)^5+0.3(0.95)^5} \approx 0.7316.

The next-year selected mixture death probability is therefore 0.7316×0.02+0.2684×0.05≈0.02805. Each type retains its original hazard, but the surviving portfolio has shifted towards the lower-mortality class. This calculation is not a measured empirical duration effect. The model intentionally excludes attained-age increases, improvements, wealth differences, contractual withdrawals and selective exits.

Money's worth and individual welfare

A money's-worth calculation discounts promised annuity benefits under a specified mortality basis and divides by the premium. The Mitchell–Poterba–Warshawsky–Brown study is an important historical contribution to empirical money's-worth measurement. Its result depends on the market quotation and the mortality and discount bases used. The same quoted benefit can have different population and annuitant money's-worth ratios because its promised future payment stream is evaluated with different survival probabilities.

A financial expected-value ratio is not a sufficient welfare measure. An individual may value the insurance against outliving wealth, even when the expected present value of cash flows is below the paid premium. Lifetime utility incorporates volatility in old-age consumption, existing pensions, family obligations, bequest motives and the value of liquidity. The Davidoff–Brown–Diamond welfare analysis supplies an established theoretical context for these distinctions. In addition, a purchasing population can be selected in ways that distribute value differently across health and income groups even if average insurer pricing is sustainable.

An economic evaluation should state whose mortality is used. The insurer prices with a prospective conditional distribution over purchasers. A social comparison may value benefits over the whole eligible cohort. A consumer may use personal beliefs about survival and consumption. An actuarial valuation at the reporting date concerns remaining obligations among policyholders already in force. Differences among these perspectives can be legitimate and informative. Conflating them can produce the false conclusion that a premium is simultaneously fair, underpriced and overpriced under the same meaning of value.

Pricing, technical provisions and accounting

Commercial annuity pricing considers mortality, guarantees, asset yields, expenses, taxes, risk capital, competitive conditions and expected policyholder behavior. The regulator's assessment of an existing obligation follows different rules from the original selling price. Under the EU Solvency II Article 77, technical provisions consist of a best estimate and a risk margin, with the best estimate based on probability-weighted future cash flows using the relevant risk-free interest-rate term structure. A mortality basis that misstates survival affects the projected best-estimate cash flows; it does not imply the insurer's offered premium has the same numerical amount as that best estimate.

For reporting entities applying IFRS 17 Insurance Contracts, fulfilment cash flows, risk adjustments and contractual service margins perform distinct measurement and reporting roles. A revision to the expected pattern of survival can change recognized insurance-contract measurements subject to the standard's rules. The IAIS Insurance Core Principles provide a broader supervisory framework addressing valuation, risk management and conduct. These sources govern different questions, jurisdictions and institutions; they do not constitute one universal annuity pricing formula.

In an ongoing portfolio, reserve adequacy depends on the selected lives who have survived to the valuation date, not simply the original mix of purchasers. Mortality improvement may affect future periods differently across cohorts, and interest-rate risk can magnify the financial significance of later payments. Contract options can also alter duration of liabilities and the risk to be held by the insurer. Model governance should preserve separate records of the initial selection basis, actual emerging experience, future improvement assumptions, guarantee cash flows and reporting-specific risk margins.

Distributional consequences and insurer conduct

Selection has consequences for access and equity as well as for financial cost. A uniform rate across people with different survival expectations can transfer value between groups; a health-adjusted quote can redistribute available income towards purchasers with reduced expected longevity. Restrictions on using certain risk classifications can change product design and customer sorting. The Finkelstein–Poterba–Rothschild regulation study shows that evaluating redistribution requires market reactions rather than a purely static transfer calculation. The relevance of protected characteristics and permissible risk rating depends on applicable law.

The provider's explanation of terms matters because a customer may rationally value survivor protection more than an initially larger life-only installment. A quote should be assessed against its actual contractual promise, including guarantees, escalation, dependents and commencement. The UK retirement-income advice review is an example of consumer-outcome scrutiny in this context, while the NAIC's annuity explanation identifies important product forms in US markets. Neither source should be stretched into a universal legislative conclusion beyond its stated authority.

Evidence Validation and Model Sensitivity

Public evidence and restricted actuarial experience

An empirical mortality study is reproducible only with sufficient data about deaths, risk exposure, entry and exit, product form and the definition of the insured life. A public study overview may establish scope, company contributions, years and aggregate counts. It does not necessarily provide the microdata or fitted model coefficients needed to independently estimate selection by duration or product choice.

The SOA/LIMRA 2020–2024 payout study reports approximately 3.1 million contract-years of exposure and 143,190 deaths from 26 companies belonging to 23 parent groups. It is an important contemporary US evidence collection, while much of the detailed material requires paid access. The earlier SOA/LIMRA 2014–2019 payout study involved approximately 4.3 million contract-years and more than 236,000 deaths from 25 companies. These study windows differ by contributing portfolios and calendar-year mortality conditions. A raw deaths ratio is not a controlled estimate of mortality improvement, purchaser selection or the age-conditional death rate.

The separate 2015–2024 deferred annuity mortality study reports roughly 80 million contract-years and 1.4 million deaths with participation from 26 companies. The contract-owner population is distinct from people who have converted retirement assets to life-contingent payout. Its wider number of contract-years cannot be pooled with payout-annuity exposure without studying product states and the legal event that creates lifetime income.

The UK CMI annuity investigation studies relevant experience across product segments including standard, enhanced and bulk business. A March 2026 CMI notice refers to Working Paper 210 comparing bulk annuity and private-sector pensioner mortality, with restricted detailed access. Its existence and subject can be recorded, but a statistic hidden in subscriber material cannot be asserted from the notice alone. The 2020–2024 SOA data request is useful for understanding which product and exposure dimensions the institutional study sought to collect.

Source hierarchy and interpretation of academic claims

The study method distinguishes an official legal or regulatory text, an official institutional study announcement, a publisher's bibliographic abstract, a publicly available working paper and original data independently reanalyzed by the present researcher. These evidence types warrant different forms of statement. An official regulator page can establish a stated current requirement within scope. An academic abstract supports a faithful description of what authors claim. A research organization's summary establishes study scale and reported findings, but the omitted detailed dataset limits methodological reproduction.

The Finkelstein–Poterba 2002 article establishes the original study's historical UK evidence and design; its compulsory–voluntary result should retain the original period and institutional context. The 2004 policyholder study provides an independent empirical dimension concerning contract characteristics. The Stabile timing model answers a theoretical optimal-stopping question under its assumptions, not an empirical cause of later mortality. These results can be synthesized because their inferential objects are explicit, rather than treating any one finding as direct proof of the others.

A recent contribution changes the research agenda. In Larrain, Previtero, and Severino (2026), the authors analyze about 600,000 Chilean retirees over 2004–2022 and report that annuitization instrumented by recent market returns reduces subsequent mortality at five- and ten-year horizons. The authors suggest income stability, lower investment-related stress and increased health investment as mechanisms. The NBER working paper is a substantive reported causal claim, but causal interpretation rests on instrument relevance and exclusion restrictions; this dossier does not replicate the estimate or certify the exclusion restriction. Its key implication is analytical: favorable observed annuitant survival could reflect both pre-existing health selection and changes in subsequent life conditions induced by stable income. Future studies must distinguish those channels with credible research designs.

Data provenance and mortality observation

A credible annuitant dataset needs life-level or properly linked contract-level records of date of birth, sex or risk group where lawfully collected, eligibility, application, issue, premium payment, income commencement, benefit form, guarantee, health classification, exposure start and end, death reporting and contract exit. It should link multiple contracts owned by one life, joint-life beneficiaries, transfers, refunds and payment continuation after the original death. For confidentiality, identifiers can be pseudonymous while link integrity is preserved.

The study must distinguish biological death from contract termination and from termination of payments. A living person may transfer or surrender some contract types; a dead original annuitant may leave certain-period installments payable to a beneficiary. Late notification of death can cause administrative payments after death and require correction to the liability record. A study counting contract termination as death will therefore distort mortality. Cleaning rules should be documented with counts of retained and excluded lives, duplicate links, unresolved dates and missing outcomes.

Let D_g be deaths in a homogeneous risk stratum g and E_g its properly defined exposure to risk. The simple central mortality rate D_g/E_g is a descriptive measure under its chosen exposure convention. A one-year conditional probability is not identically equal to the central rate without assumptions about death timing or the force of mortality. An actual-to-expected ratio can be expressed as

A/E=∑gDg∑gEgqgexpected,A/E=\frac{\sum_g D_g}{\sum_g E_g q_g^{\mathrm{expected}}},

for a consistent definition of annual fractional exposure and the appropriate expected-probability approximation. More exact implementations allocate expected deaths across each life and interval. The analyst must also distinguish counts weighted by policy, life, premium or benefit amount. A mortality table used for reserves should be assessed against the economic weighting of the actual liability and should not quietly mix incompatible exposure definitions.

An in-force book observed only after a database extraction date is left-truncated relative to original issue. People who died or terminated before the extract may be invisible. Living lives at the final study date are right-censored for subsequent death; withdrawn or lost-to-follow-up lives require appropriately documented handling. A complete dataset retains the exact possible period of observation for each person rather than assuming all individuals remained continuously at risk in the same manner.

Design for estimating purchase selection

To study the decision to buy, construct a cohort of people eligible to purchase, with the date eligibility was first established. Track annuity quotations, offers accepted or rejected, income options, financial wealth and subsequent outcomes, with consistent privacy protection. The transition to purchase can be described by a cause-specific or discrete-time hazard conditional on being alive and eligible. Mortality after purchase is a separate transition. An observational comparison of early and late buyers should adjust for attained age, birth cohort, calendar period, financial conditions, prior health, and the possibility that some people could never access the offered product.

A multistate representation can track: alive and eligible without a payout annuity; alive after an annuity issue but before income commencement; alive with income payments due; dead; and relevant contract exits. This uses established survival and event-history methods rather than inventing a new framework. It prevents the analyst from treating the years before a future annuity purchase as exposure to purchased coverage. It also separates whether the person survived to buy from whether the product itself changed later survival prospects.

Selection into a health-rated quote must be studied as part of the process. If an enhanced annuity program is launched, a change in the average mortality of standard-rate purchasers can occur because of reclassification. The existence of the program and which applicants were eligible are then required explanatory variables. Without a record of applicants, offered quotes and rejected contracts, the insurer's in-force portfolio cannot identify population-wide selection effects by itself.

Sensitivity, backtesting and governance

A useful sensitivity study varies one conceptual axis at a time. The main axes are comparison population, voluntary versus institutional purchase obligation, first-year versus subsequent-duration mortality, standard versus enhanced underwriting, timing of purchase and benefit commencement, amount-weighted versus life-weighted exposure, guarantees, age and birth-cohort mix, calendar shocks, interest rates, mortality improvement and discount-rate methodology.

The OECD's 2025 pension report provides an international period-life-expectancy context, including OECD averages for people aged 65. These population figures cannot be used as selected annuitant assumptions without cohort adjustments and a documented population mapping. Period life expectancy reflects contemporaneous rates at different ages; a cohort projection incorporates expected future calendar-year conditions during the cohort's life. Confusing the two can bias projected duration of lifetime payments.

For US defined-benefit pension obligation measurements, the 2025-effective ASOP No. 27 provides practice guidance on appropriate selection of assumptions, including mortality and retirement and election assumptions within its scope. It is a pension measurement standard, not a universal commercial individual-annuity tariff. The IAA international actuarial standards register offers a professional standards framework for different assignments. A technical report should state which standard actually applies to the work performed.

An actuarial backtest should compare predicted and realized deaths and, where possible, promised income cash flows by age, contract form, issue cohort and duration. Mortality fit and present-value fit are not the same; a model can reproduce aggregate death counts while misallocating deaths toward contracts with larger benefits. Prediction intervals should reflect parameter uncertainty, systematic uncertainty in mortality improvement, and changes in product mix. Documentation must preserve the exact data cut, mortality-table edition, projection scale, interest-rate basis, valuation date and contract-version mapping.

Integrated Research Findings and Open Scientific Questions

The complete observation process determines meaning

The first integrated finding is that annuitant mortality cannot be interpreted independently of the system that generates its observation. A member of an eligible retirement cohort must survive to an opportunity to buy, encounter a contract and price menu, elect a product, pass any underwriting requirements, and remain observable until death or censoring. Every stage may select a different mixture of health, wealth and preferences. The original UK empirical studies identify contract-menu and market-entry sorting; the Chile retirement-market findings connect the observed pool to a different national pension opportunity set.

The scientific consequence is that a mortality comparison has a precise interpretation only when the inclusion criteria, purchase risk set, contract form, age, calendar period, exposure unit, and benchmark are stated. A very large administrative dataset can produce a precise but substantively misidentified estimate when any of those definitions are inconsistent. Precision must therefore be evaluated together with relevance to the original contract and risk.

Timing changes eligibility, valuation, and counterfactuals

The second finding concerns the temporal origin of analysis. A delayed buyer has survived until the decision to purchase. A person who buys deferred income earlier assumes, through the contract, a different series of future claims and possibly a different pre-commencement death benefit. An in-force payout beneficiary may have survived from issue to commencement and thus belongs to a conditional survivor pool. These statements use the same survival mathematics but different denominators and valuation dates.

The optimal-annuitization research by Stabile makes purchase time an explicit decision variable, while the actuarial identities in this monograph show why delaying issue and deferring first income have different risk exposures. Empirical work that lacks the dates of eligibility, contract issue and income commencement cannot reliably isolate private-information selection from survival conditioning. The most valuable future data infrastructure would link complete quote and issue histories to death registries and benefit schedules.

Duration selection is mechanically possible without changing hazards

The third finding follows directly from the heterogeneous-mixture calculation. In a cohort with low- and high-mortality types, the relative share of lower-mortality lives rises among survivors over time. An apparently favorable select duration can arise because the pool changes, even if no individual health benefit is created by annuitization. Conversely, age-related hazard increases or an adverse calendar shock can cause aggregate observed mortality to rise despite this ongoing within-cohort selection. Any empirical duration analysis must decompose these simultaneous influences.

The SOA payout-annuity experience program and the CMI research investigations are institutionally relevant settings for that measurement. Their public summaries establish the population and date framework, but detailed tests must be based on properly accessible policyholder exposures. The absence of such confidential data in this dossier limits the analysis to published findings and reproducible theoretical illustrations.

Potential effects of annuity income on actual survival

The fourth finding identifies a question that cannot be resolved by conditioning alone. Annuity purchasers may have longer initial life expectancy because they select into purchase. They may also experience changes in consumption and stress after buying assured income, potentially changing their future mortality. The authors of the 2026 Chile NBER working paper report a causal result based on an instrumental-variable strategy; independent scrutiny must examine whether the market-return instrument affects health only through annuitization and whether the source population supports transfer of the conclusion to other markets.

This empirical possibility makes simple observational comparisons insufficient as evidence either that annuitization itself prolongs life or that favorable annuitant mortality is entirely caused by prior selection. An appropriate identification strategy should preserve pre-purchase health history, exogenous financial variation if credible, benefit differences, follow-up periods and migration between retirement-income states. Competing mechanisms can coexist and may differ by household wealth and welfare regime.

Price, policyholder welfare, and reporting are related but separate evaluations

The fifth finding concerns economic interpretation. Mortality selection affects expected lifetime payments for a given annuity promise and therefore commercial rates, risk capital and existing liabilities. Yet the insurer's pricing basis, the buyer's own expected value, a social-distributional comparison, an EU solvency measurement and an IFRS accounting number answer distinct questions. The Solvency II valuation rule and IFRS 17 measurement requirements govern specified institutional calculations, while Davidoff and coauthors' welfare research evaluates benefits of annuitization under economic assumptions.

A useful original analytical contribution is to place all of these valuations on one disclosed timeline: eligibility, quote, issue, first survival-contingent entitlement, subsequent exposure, death and settlement of guarantees. This is a transparent synthesis and reporting discipline derived from established models. It does not establish a novel theory or presume a single legal standard for all annuities.

Open questions and feasible research design

A stronger empirical study could connect a representative eligible retiree panel with anonymized health, financial and quotation records, annuity provider offers, actual contract elections, benefit schedules and linked death registrations. Its outcome design would separately estimate the hazard of purchase, the distribution of selected contract features, and mortality conditional on being alive at the appropriate issue or payment commencement date. A robust analysis would vary comparison populations and control for calendar-year shocks, risk classifications and market yields.

Another priority is cross-country harmonization of benefit definitions. An individual immediate life annuity, a medically enhanced contract, a deferred payout promise, a group pension annuity and a programmed withdrawal arrangement have distinct financing and survival exposure. A common description of the contract states and payment obligations would make comparisons more meaningful, while preserving the legal uniqueness of each jurisdiction. Each original mortality assumption should be versioned against its source and supported by a clearly documented cohort.

Further work should examine how changes in public minimum pension protection alter the composition of private purchasers, how mortality selection shifts under expanded enhanced-annuity underwriting, and whether rate options reveal health expectations after controlling for family and bequest preferences. A final priority is prospective liability backtesting by issued benefit form and amount, so that the test of mortality selection is tied to the insurance obligation rather than to crude deaths alone.

Conclusion

Annuitant mortality selection is created by the conditional composition of people who buy life-contingent income and by their choices among contractual payout structures. Purchase timing determines the age, calendar environment, selection window and survival conditioning under which that composition enters the insurer's risk pool. After purchase, mortality heterogeneity can alter the composition of survivors by duration without any change in individual hazards. Institutional pension rules, medical underwriting, benefit guarantees, financial opportunities and the purchaser's beliefs further determine the resulting survival and benefit distribution.

The historical UK research provides direct evidence of voluntary-market selection and product-feature sorting. US SOA/LIMRA and UK CMI studies demonstrate how specialized annuitant mortality experience is assembled, while the accessible public record places clear limits on independent analysis of restricted policyholder data. Chilean research contributes evidence about pension-system incentives and a contemporary, carefully qualified working-paper claim concerning possible post-annuitization effects on survival. These contributions are complementary when their population, period and identification limits are respected.

The insurer's principal financial implication is that a life-contingent benefit must be valued against survival probabilities appropriate to the lives, payment forms and durations under observation. The economic implication is that expected cash-flow value, annuitant welfare and distributional incidence require explicitly differentiated mortality assumptions and perspectives. The methodological implication is that eligibility, purchase, income commencement, contract guarantees and survival observation must occupy separate positions on the data timeline. These conclusions constitute an original synthesis of established scientific methods and verifiable research sources; no novel theoretical system, original insurer dataset, independent causal replication or publication claim is made.

References