Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Joint-Life Survival Dependence in Retirement Payouts

Abstract. Joint-life retirement income is governed by the survival states of two people whose lifetimes may be statistically dependent. The dependence affects the probabilities of simultaneous survival, survival of exactly one beneficiary, the first death, and the final death. These distinctions determine the expected timing and amount of lifetime income, survivor continuation benefits, annuity reserves, and household income security. This monograph derives the valuation of first-death, last-survivor, and participant-first reversionary annuities directly from bivariate survival probabilities. It examines household frailty, shared exposures, simultaneous mortality shocks, and the change in survivor mortality after bereavement; evaluates evidence from dependent-life annuity portfolios and population studies; distinguishes statistical dependence from causal bereavement; and compares United States and United Kingdom entitlement rules with international actuarial and financial-reporting frameworks. A fully specified hypothetical common-shock calculation demonstrates that increased positive lifetime dependence, holding marginal survival constant, reduces the expected value of a survivor-only continuation and a last-survivor pension while increasing the value of a both-alive benefit. The direction depends on contractual payment states; it is therefore incorrect to assign one universal dependence adjustment to every joint-life annuity. The contribution is an integrated, source-checked actuarial and institutional synthesis, with reproducible equations, model validation criteria, and precise limitations rather than a newly proposed theory.

Keywords: Joint-life mortality; dependent lifetimes; bivariate survival; reversionary annuity; survivor pension; last-survivor annuity; common shocks; frailty; bereavement; copula; actuarial valuation; annuitant experience; retirement income; IFRS 17; pension benefits.

Introduction

Lifetime pension payments may continue as long as a named participant lives, as long as both named persons live, or as long as either remains alive. Some benefits decrease after the first death; others begin only if the designated participant dies while a beneficiary survives. A valuation that is accurate for one promise may be inaccurate for another even when both are marketed as a joint-life annuity. The scientific object of this dossier is the dependence between the remaining lifetimes of the covered persons and its specific consequences for payments under those promises.

The seminal study by Frees, Carriere, and Valdez (1996), Annuity Valuation with Dependent Mortality, demonstrated from a large insurer's portfolio that statistical independence of the lives was consequential for valuation. Its abstract reported an approximately five-percent reduction in the annuity values examined when dependent mortality replaced independence; this is a study-specific estimate, dependent on the contracts, data and model, rather than a transferable universal loading. The literature subsequently distinguished temporal widowhood effects, underlying shared heterogeneity, catastrophic common shocks, and covariate-dependent association. Ji, Hardy, and Li (2011) developed Markovian approaches to joint-life mortality, while the 2026 study by Da Fonseca and Wong examined joint-survival annuity and option valuation using a dynamic Wishart process.

The practical stakes extend beyond insurer prices. The benefit formula determines how income changes for a surviving household member, and legal rules may prescribe a minimum continuation amount or require a spouse's consent to a waiver. Pension administrators must also preserve records of covered persons, designations, death dates, election dates and payment timing. Any comparative valuation must identify the contract and governing jurisdiction before interpreting the mortality model.

This research treats each source according to what was actually accessible: operative public law and institutional guidance support regulatory descriptions; publisher articles support findings available in their abstracts or accessible text; and actuarial identities are independently derived. Numerical examples below are hypothetical, deliberately separated from observations and market quotes. No new household mortality data were collected.

Scientific Foundations, Historical Development, and Contract Identity

The contractual object

Let A denote the pension participant or first designated annuitant and B the spouse or other named survivor. A lifetime pension payable to A alone terminates at A's death, apart from any explicitly separate guaranteed payment, refund, or death-benefit provision. A pure joint-life or first-death annuity pays while both covered persons are alive and ends upon the first death. A last-survivor annuity pays while at least one is alive. A reversionary annuity starts or continues a defined payment to B when A has died and B is still alive. A participant-first joint-and-survivor annuity may combine full payment while A lives with a reduced payment to B after A's death. A period-certain guarantee and a refund-of-premium payment have additional rules that require their own contract-state functions.

The UK MoneyHelper description of annuity options explicitly differentiates joint-life continuation, value protection, guarantee periods, and payments between regular settlement dates. It also identifies that guarantees can interact by overlap or nonoverlap. These are distinct promises with different timing and beneficiary conditions. No statement about dependence has actuarial meaning until the cash-flow state is specified.

A pension's funding arrangement matters as well. An insurer-issued annuity transfers specified longevity obligations to the issuing insurance entity. A defined benefit pension scheme may promise a survivor pension while the employer or pension trust retains the ultimate benefit obligation. A defined contribution account permitting discretionary withdrawals is exposed to asset depletion and beneficiary rules, but is not by itself an annuity payment promise. The IAS 19 classification of pension arrangements distinguishes defined benefit and defined contribution responsibilities according to the substance of plan terms. The benefit's actuarial cash-flow identity can be the same across arrangements while accounting, security, funding and legal enforcement differ.

Evolution of the scientific problem

Traditional multiple-life valuation was often constructed by multiplying single-life survival probabilities. That factorization makes calculation convenient and is correct when the two remaining lifetimes are independent conditional on the information and assumptions used. Married persons may share socioeconomic conditions, housing, environmental exposures, lifestyles and selection characteristics; death of one spouse may also change the other's subsequent hazard. These possibilities motivated dependent-lives research without eliminating the usefulness of independence as a benchmark.

Frees, Carriere, and Valdez (1996) used copulas to join marginal survival functions and calibrated dependence to insurance experience. Subsequent approaches sought a closer correspondence between observed temporal events and the mechanism producing association. Ji, Hardy, and Li (2011) distinguished instantaneous joint events, short-term effects of spousal death and longer-term association in joint-life mortality transition models. Gouriéroux and Lu (2015) combined a frailty component with a mortality-intensity change at bereavement in Love and Death: A Freund Model with Frailty. Lu (2017) further emphasized that causal bereavement and unobserved heterogeneity must be distinguished; her analysis of spousal mortality dependence found asymmetric responses by sex and additional positive association attributable to unobserved differences.

The resulting research domain involves three related questions: what joint distribution describes the observed dates of death; what mechanisms could explain that distribution; and how the distribution changes a specified contractual cash flow. The first question is statistical, the second causal and demographic, and the third actuarial and legal. Evidence for one is not automatically evidence for the others.

Definitions and scope of the covered population

Let T_A and T_B denote nonnegative future lifetimes measured from the same valuation date. Their distribution is conditional on age, sex or appropriately justified biological/demographic categories, observed medical underwriting information where lawful and available, cohort, calendar date, jurisdiction, and contract selection. Using spouses as the empirical reference does not mean that all two-life annuities cover married heterosexual couples: named beneficiaries, former spouses subject to pension orders and other eligible dependants require their own entitlement and statistical specifications. A population mortality table describes a different sampling frame from a jointly insured annuitant portfolio.

The precise object of estimation is the joint distribution of T_A and T_B for the covered and eligible lives, after conditioning on the information relevant to the contract. Estimation should not silently import a population bereavement hazard into a selected annuity book or treat observations of household deaths as independent exposures.

Joint Survival Distributions and Contract-State Valuation

The four survival states

At a payment horizon t define the marginal probabilities p_A(t)=P(T_A>t) and p_B(t)=P(T_B>t), and the joint survivor probability J(t)=P(T_A>t,T_B>t). The probability that at least one person remains alive is U(t)=p_A(t)+p_B(t)-J(t). The four mutually exclusive states are:

State at time t Probability
Both alive J(t)
A alive, B dead p_A(t)-J(t)
A dead, B alive p_B(t)-J(t)
Both dead 1-p_A(t)-p_B(t)+J(t)

These identities follow by inclusion and exclusion and hold under every valid bivariate lifetime distribution. Coherent probabilities require max(0,p_A+p_B-1) ≤ J ≤ min(p_A,p_B) at each t. The independence benchmark is J_ind(t)=p_A(t)p_B(t). Define the dependence deviation Δ(t)=J(t)-p_A(t)p_B(t). It is a horizon-specific probability difference. The sign and size of Δ may change with time and covariates; it is not a causal parameter or a constant correlation coefficient.

For payments due at n=1,...,N and a deterministic discount factor d_n, the expected present value is Σ[n=1..N] d_n E[C_n]. The formulas also apply with nonuniform payments b_n, which can represent prescribed escalation. When economic discount factors are stochastic and correlated with mortality, valuation requires an expectation of the product of cash flow and discount factor, not merely a product of their separate expectations.

Survivor continuation with named primary and beneficiary

Suppose a fixed annual amount b_n is paid whenever A is alive, regardless of B's survival; if A has died and B is still alive, a proportion r of b_n is paid, where 0≤r≤1. At a payment time n the expected benefit is

E[C_n]=b_n{p_A(n)+r[p_B(n)-J(n)]}.

The bivariate dependence enters exclusively through the probability that the designated survivor B outlives the participant A. The value is

V_rev=Σ[n=1..N] d_n b_n{p_A(n)+r[p_B(n)-J(n)]}.

Holding marginal survival probabilities, payment terms and discount factors fixed, the difference between dependent and independence valuations is

V_rev(dep)-V_rev(ind)=-Σ[n=1..N] d_n b_n r Δ(n).

A positive Δ decreases this reversionary liability because it reduces the probability that B is alive after A's death; a negative Δ increases it. A pre-death payment contingent on A alone does not acquire dependence sensitivity merely because B's name is on the contract. If r=0, the valuation is exactly the single-life annuity on A under the assumptions stated. The effect also depends on which person is the primary annuitant: reversing A and B changes the survivor-only state to p_A-J and can change the economic result when their marginals differ.

Both-alive and last-survivor benefit designs

For a pure both-alive payment b_n, E[C_n]=b_n J(n). Positive Δ therefore increases its value relative to independence. For a level payment while either life survives, E[C_n]=b_n U(n)=b_n[p_A(n)+p_B(n)-J(n)]. Positive Δ decreases that value relative to independence. These two payment rules share the same underlying couples but have opposite valuation responses.

A generalized four-state benefit B_{11,n}, B_{10,n}, B_{01,n}, B_{00,n}, where 1 denotes alive and 0 denotes dead, has expected amount

E[C_n]=B_{11,n}J+B_{10,n}(p_A-J)+B_{01,n}(p_B-J)+B_{00,n}(1-p_A-p_B+J).

Collecting terms gives

E[C_n]=B_{00,n}+(B_{10,n}-B_{00,n})p_A+(B_{01,n}-B_{00,n})p_B+(B_{11,n}-B_{10,n}-B_{01,n}+B_{00,n})J.

Consequently the marginal-fixed dependence sensitivity equals the last coefficient multiplied by Δ. This expression provides a contract-state reading of dependence across multiple payout patterns. It is an algebraic synthesis of established joint-life valuation, not a new mortality model. A guaranteed benefit that continues after both deaths requires a nonzero B_{00,n} for the relevant contractual period and may introduce additional contingent beneficiary states.

Timing, conditions, and continuous-payment forms

A monthly payment in advance, a monthly payment in arrears and a continuous income stream have different valuation boundaries. The formulas above assume payment at the stated end-of-period horizon if the qualifying survival state holds then. Actual policy clauses may grant the full month's payment when a person was alive on an earlier determination date, pro-rate the final interval, or impose a proof-of-death administrative delay. Such provisions require valuation on the contract's true timeline.

For a continuous payable rate b(t) and discount factor d(t), a both-alive annuity has value ∫[0,∞) d(t)b(t)J(t)dt, and a last-survivor annuity substitutes U(t). A participant-first annuity with survivor fraction r substitutes p_A(t)+r[p_B(t)-J(t)]. Deferred commencement at a specified time s changes the domain to t≥s and may require survival of one or both designated persons at the vesting date. Early-death refunds, purchase-price guarantees, surrender provisions and beneficiary substitution produce extra states and cannot be inferred from the displayed survivor proportion alone.

A bounded one-date illustration

At one hypothetical future payment date suppose p_A=0.80, p_B=0.70, and J=0.60. The independence benchmark has J_ind=0.56. The actual assumed probabilities are coherent: only B survives with probability 0.10 rather than 0.14, and at least one survives with probability 0.90 rather than 0.94. If b=$10,000 and r=0.60, the expected participant-first payment is $8,600 rather than $8,840 under independence. A both-alive $10,000 benefit is worth $6,000 in expected undiscounted payment rather than $5,600, and a last-survivor $10,000 benefit is worth $9,000 rather than $9,400. The numbers describe one date and one assumed distribution; they are neither observed frequencies nor lifetime prices.

Household Mortality Dependence and Shared Shocks

Shared frailty and persistent exposures

Household members may experience common or correlated determinants of mortality. A shared latent frailty Z can represent unmeasured household conditions, and conditional independence given Z need not imply unconditional independence: E[p_A(t|Z)p_B(t|Z)] generally differs from E[p_A(t|Z)]E[p_B(t|Z)]. If both conditional survival probabilities change in the same direction with Z, their covariance is positive. Observed association may therefore persist even without a direct causal effect of one death on the other.

A frailty model needs explicit assumptions about the distribution of Z, conditional hazards, selection and censoring. Married couples entering an annuity portfolio are also selected: both may have lived to a particular age, elected particular payment forms and satisfied the insurer's eligibility rules. Conditioning on this event changes the joint distribution. Gouriéroux and Lu (2015) set out the distinction between common frailty and widowhood-related hazard changes, a distinction that remains central to interpreting portfolio data.

Bereavement and state-dependent hazards

After the first spouse's death, the remaining spouse may experience a changed short-term or longer-term mortality hazard. The effect can differ by the sex, age and health of the survivor and by the time elapsed since bereavement. State-dependent hazard models therefore track both identities and the duration since the first death. Ordinary time-homogeneous Markov models may be too rigid if the postbereavement effect decreases with elapsed time; semi-Markov models allow duration dependence.

Shor and colleagues (2012) conducted a systematic meta-analysis of widowhood and mortality. Their published analysis found higher mortality for widowed persons in the included observational comparisons, with variation by sex, age and design. Those associations support the importance of investigating bereavement in joint-life valuation; they do not license inserting one meta-analytic hazard ratio as a causal multiplier into every insured pension cohort. A death can be preceded by a period of caregiving, illness and financial stress, and both spouses may share preexisting risks. The separation of those mechanisms is an identification problem.

Lu (2017), in Broken-Heart, Common Life, Heterogeneity, estimated a mixed proportional-hazards specification with treatment effects and unobserved heterogeneity using joint-annuity data. The accessible publisher abstract reports an asymmetric effect of spousal loss and additional dependence from unobserved heterogeneity. It also reports that missing either effect can produce pricing errors in either direction for different couples. This empirically reinforces the contract-state result that dependence cannot be captured by one universal adjustment.

Simultaneous household shocks

A catastrophe, accident, epidemic or shared environmental exposure may affect both lives at the same instant. An instantaneous joint-death event requires attention to the point mass on T_A=T_B in a joint lifetime distribution. Certain smooth bivariate copulas have no such diagonal mass; a common-shock construction can generate it explicitly. The probability of a common event matters for same-day death claims, simultaneous cessation of benefits, and the identification of surviving beneficiaries. Insurance contracts may need legal rules for uncertain order of death, although the precise consequence is jurisdiction- and wording-dependent.

A common shock also illustrates why marginal rates alone cannot identify joint-life liabilities. Different combinations of individual-specific mortality and common-shock intensity can produce the same single-life marginals and different joint survivor probabilities. A dataset with two marginal mortality tables and no matched couple records generally cannot distinguish those combinations.

Copulas, transition models, and dynamic mortality

A survival copula joins two calibrated marginal survival distributions into a joint distribution; at fixed t, J(t)=C_t[p_A(t),p_B(t)] under a suitable survival-copula specification. A valid lifetime model must be coherent across all pairs of time points, not only on the equal-time diagonal needed for some payment formulas. Dependence may differ by age gap and which spouse is older. Dufresne, Hashorva, Ratovomirija, and Toukourou (2018) investigated age differences in joint lifetime modeling using Canadian insurer data and reported better fit when age-related dependence factors were included.

Copulas are useful for flexible statistical association, common-shock models for simultaneous risks, and multistate or frailty models for time since bereavement and heterogeneity. Their distinct assumptions can yield different pricing outcomes even where fitted marginal mortality is equal. Da Fonseca and Wong (2026) developed a dynamic Wishart mortality formulation allowing dependence between mortality intensities and tractable joint-survival annuity and option calculations. Their Canadian portfolio application reported positive dependence relative to benchmark specifications. Its value for practice is a documented dynamic alternative; its calibration is not a universal representation of all retirement populations.

Mortality improvements and systemic shifts

Time-varying cohort mortality improvement changes both marginals. Dependence estimation must distinguish a persistent common household factor from shared national or regional mortality trends, health-system changes and shocks affecting many unrelated households. The resulting analytical decomposition matters in asset-liability management: diversification across couples can reduce idiosyncratic death risk while leaving aggregate longevity improvements or disaster shocks correlated at portfolio level.

A long-survival stress should therefore consider positive correlation of longevity improvements across policyholders, separate from the dependence between the two lives within each contract. The first is portfolio and population risk; the second is household dependence. Both affect capital, but their data, risk aggregation and hedging arguments differ.

Contract Valuation, Institutions, and Jurisdictional Entitlements

The value of the promise is defined by its states

For a participant-first continuation benefit, dependence modifies the survivor-only portion, and the liability difference is proportional to r when all else is fixed. When benefits escalate, the sign of the difference remains determined by the same state coefficient but late-horizon weights change. When a guarantee period pays even after both lives die, part of the liability is driven by the fixed guarantee schedule, so the total proportional effect of dependence can be diluted. When commencement requires both people to survive until a deferred date, the conditioning and eligibility terms must be rebuilt before applying a simple continuation formula.

The actuarial equivalence of alternative payment elections can be studied by setting their expected values equal under a stated mortality basis, discount curve and expenses. A joint-and-survivor election commonly pays a lower amount before the first death than an otherwise comparable single-life election because it extends some payments to a beneficiary. That is a conditional valuation outcome, not a claim that every specific insured quote or pension conversion rate is actuarially fair. Expenses, regulatory rules, guarantees, anti-selection, capital, profit margins and insurer competition can change the offered amount.

United States: qualified joint-and-survivor protections

For pension arrangements to which its requirements apply, 29 U.S.C. § 1055 defines a qualified joint and survivor annuity as a life annuity for the participant with a spouse's survivor annuity of at least 50 percent and no more than 100 percent of the amount payable during the joint lives, with specified actuarial equivalence. The IRS qualified joint and survivor annuity guidance explains participant and survivor payments and QJSA/QOSA distinctions. These are United States pension-law requirements for covered arrangements, not worldwide standard clauses for commercial annuities.

Entitlement and pricing answer different questions. A model can calculate the expected cost of a 50 percent or 100 percent continuation under a given mortality dependence without deciding whether the participant may lawfully waive survivor coverage or whether the beneficiary falls within a protected spouse definition. Qualification, waiver, consent and pension order rules are determined by the relevant plan and law. A reviewer should read the original election terms, date of retirement and beneficiary evidence before modeling the legally payable state.

United Kingdom: contract forms and dependant annuities

The UK MoneyHelper explanation of pensions after death distinguishes lifetime annuities with selected survivor options from account-based pension inheritance. The UK tax authority's beneficiary annuity guidance explains when a dependant's or nominee's annuity can be established as a related annuity at the member's retirement. These sources describe jurisdiction-specific product categories and tax treatment, rather than a common actuarial percentage applying to every UK pension.

An observed beneficiary designation is insufficient by itself to infer contractual continuation. The annuity might require a specific eligible dependant, evidence of survival at a settlement date, or a defined purchase window. An overlapping guarantee can also produce an interim payment different from the reversionary fraction. Mathematical valuation requires each such clause to appear as a state or time condition.

International pension design and professional standards

The OECD Recommendation on the Good Design of Defined Contribution Pension Plans, adopted in 2022, addresses the coherent design of accumulation and payout arrangements and protection against longevity risk. This is an international policy recommendation whose institutional application varies across adherents. It supports evaluating lifetime income and the financial position of survivors in context; it does not prescribe one household mortality copula.

International reporting standards distinguish responsibilities by issuing institution. IFRS 17 Insurance Contracts requires measurements including expected future cash flows, financial discounting, adjustment for nonfinancial risk and, where applicable, the contractual service margin. A life-contingent cash-flow estimate should reflect the actual entitlement probabilities and assumptions consistent with the reporting framework. IAS 19 Employee Benefits is pertinent when a reporting entity has employee-benefit obligations and addresses actuarial measurement of defined benefit plans. Applying both standards indiscriminately to the same pension promise would confuse the identity of the liability holder. The IAA International Standards of Actuarial Practice register provides professional guidance relevant to assumptions, documentation and reporting in applicable assignments; regulatory incorporation is jurisdiction-specific.

Policyholder welfare and survivor security

Joint-life income protects a household against the loss of earnings or pensions after a first death, but benefit design matters for the welfare effect. A beneficiary with substantial independent income may prefer a different continuation fraction from a financially dependent survivor; a guarantee period may protect heirs even when both covered persons die early. These preferences interact with health information, premium constraints, inflation protection and survivor eligibility. The actuarial valuation of r is an input to plan and product comparison; it does not establish one optimal fraction for every household.

An insurer evaluating continuation payments must recognize that the same terms affect two distinct outcomes: the surviving household's income path and the value of the insurer's contingent obligation. Scientific interpretation keeps both outcomes visible, rather than treating a lower expected claim amount as proof that the product provides superior household protection.

Evidence, Reproducible Sensitivity Analysis, and Model Validation

A reproducible common-shock benchmark

Consider independent exponential latent times E_A, E_B and E_C with annual intensities α=0.035, β=0.045 and γ=0.005. Define T_A=min(E_A,E_C) and T_B=min(E_B,E_C). The resulting marginal hazards are λ_A=α+γ=0.040 and λ_B=β+γ=0.050. Their survivals are p_A(t)=exp(-0.040t), p_B(t)=exp(-0.050t), and joint survival is J_dep(t)=exp[-(α+β+γ)t]=exp(-0.085t). An independence comparison with exactly the same marginals uses J_ind(t)=exp(-0.090t). Positive dependence here is caused by the explicitly modeled common event. This stylized construction is an established shared-shock form; its parameters are chosen solely to illustrate mathematics.

Assume annual payments at the ends of years n=1 through 30, amount b=$12,000, survivor continuation r=0.60, and a deterministic continuously compounded discount rate of 3 percent per year so that d_n=exp(-0.03n). No expenses, fees, taxes, improvement, selection, inflation, extra guarantees, lapses or competing benefits are included. All amounts are nominal and the same in every year.

The assumptions produce the following exact-to-rounding values:

Horizon in years p_A p_B J_dep J_ind U_dep U_ind
1 0.960789 0.951229 0.918512 0.913931 0.993507 0.998088
10 0.670320 0.606531 0.427415 0.406570 0.849436 0.870281
20 0.449329 0.367879 0.182684 0.165299 0.634525 0.651910

The participant-first present value can be independently replicated by the following finite sum, with γ between zero and the smaller of the two fixed marginal hazards:

V_rev(γ)=12000 Σ[n=1..30] exp(-0.03n){exp(-0.04n)+0.60[exp(-0.05n)-exp(-(0.09-γ)n)]}.

The equivalent last-survivor formula is

V_last(γ)=12000 Σ[n=1..30] exp(-0.03n){exp(-0.04n)+exp(-0.05n)-exp(-(0.09-γ)n)}.

Evaluating the sums yields V_rev(0)=$168,908.85 and V_rev(0.005)=$166,635.59, a difference of −$2,273.26 or about −1.35 percent of the independence benchmark. For the same 30 payments at $12,000 while either life survives, V_last(0)=$184,693.29 and V_last(0.005)=$180,904.52, a difference of −$3,788.77. These are present values under stated numerical hypotheses, not premium quotes or estimated impacts for a real insured book.

Sensitivity to the shared-shock intensity

Keeping marginal hazards at 0.040 and 0.050 while changing γ requires setting the individual-specific intensities to α=0.040−γ and β=0.050−γ. For the selected examples γ=0, 0.0025, 0.005, 0.010 and 0.020, both specific intensities remain nonnegative. The corresponding participant-first values, using the same payment and discount structure, are:

Annual common-shock intensity γ Participant-first present value
0 $168,908.85
0.0025 $167,791.49
0.0050 $166,635.59
0.0100 $164,201.11
0.0200 $158,786.24

The monotonic decline is an algebraic implication of this fixed-marginal construction, not a forecast that changes in actual catastrophe exposure leave each spouse's life expectancy unchanged. Increasing γ in an unconstrained real-world system would ordinarily change both marginals and dependence. This exercise isolates one effect for sensitivity analysis, a valuable diagnostic precisely because it states its counterfactual.

First-death information and incomplete data

An empirical insurer typically observes payment commencement, original annuitant identifiers, beneficiary election, exit event, and sometimes the reported death of the primary insured but not the complete subsequent lifetime of the beneficiary. A last-survivor contract may record payments through the final death but provide limited information about which spouse died first. Records can also be left-truncated, right-censored or split across administration systems. One must model the likelihood of the observed histories and their censoring pattern, rather than treating every policy as two complete independent lifetimes.

The Society of Actuaries 2009–2013 Individual Payout Annuity Experience Study illustrates an operationally important issue. Its December 2022 update corrected certain actual-to-expected ratios by contract to exclude the secondary annuitant from the count while the policy remained in joint status. The amount-weighted ratios already reflected the intended treatment. This finding demonstrates that the definition of exposure and the definition of a counted life can materially affect apparent experience even before a dependence model is estimated. The SOA 2012 Individual Annuity Reserving Table materials describe annuity valuation mortality tables and projection elements. A marginal valuation table cannot, without an additional assumption or data, provide a joint distribution.

Validation sequence

An auditable analysis should begin with the contract and population, not with the selection of a copula family. The evidence record should identify the data controller, observation and calendar periods, original retirement and death-date fields, coverage and beneficiary definitions, quality rules, reporting lags, censoring, missing spouses and exclusions. It must separately describe which records contribute first-death, last-death and matched two-life information. The exposure denominator must agree with the reported claim event.

Next, estimate or justify age-, cohort- and calendar-conditioned marginal mortality, and test fit across entry ages, benefit types, duration since annuitization and relevant demographic groups. Estimate the dependence component with a model matched to the observed information, then compare independence, plausible copulas, shared frailty and state-dependent hazard alternatives. For any empirical claim about a causal widowhood effect, specify the identification strategy and control for shared confounding and prebereavement health trajectories. Standard errors, model risk and censoring uncertainty must be disclosed.

Validation should include out-of-sample cohort or calendar-period evaluation where data permit, joint-state probability calibration, observed versus expected payments, parameter stability, sensitivity to age gap and partner ordering, and the amount-weighted as well as policy-weighted effect on liabilities. Test the contract-state payoff under both first-death orderings and simultaneous death. A model may predict one-year marginal deaths well yet fail to forecast the probability of a surviving spouse; this is a substantive diagnostic, not a small presentation issue.

Finally, separate the independent marginal stresses from dependence stresses. Longer life expectancy increases survival-based liabilities under many payout forms; higher positive association with marginals held fixed may lower last-survivor and reversionary values. The net financial effect of a realistic pandemic, socioeconomic shift or improvement program requires the combined scenario and may differ from both partial derivatives.

Accounting, uncertainty, and reproducibility controls

The numerical illustration has no empirical confidence interval because its parameters were stipulated rather than estimated. An empirical valuation would propagate uncertainty from mortality tables, dependence estimates, discount curves and benefit data; it would also document any risk-adjustment or capital methodology outside the expected cash-flow calculation. Under the IFRS 17 fulfilment-cash-flow framework, risk adjustment and future-profit recognition have accounting roles separate from the best-estimate survival-state expectation. The displayed values should not be labeled IFRS 17 liabilities, statutory reserves or market prices.

The mathematical checks for this monograph are reproducible using the specified exponential function and finite summation. The sources verify the scientific context and institutional rules; they are not represented as endorsing the hypothetical hazard rates or the resulting dollar amounts. Independent code or spreadsheet replication should use all 30 annual terms, not a truncated integral approximation, and retain the stated end-of-year timing.

Integrated Findings and Research Agenda

Findings supported by theory and evidence

First, all standard two-life payment states can be expressed using the two marginal survival probabilities and the joint survivor probability at the payment date. This is an exact implication of probability algebra. The coefficient on joint survival in each contract-state payoff identifies the direction of its dependence effect at fixed marginals. A both-alive payment, a last-survivor payment and a named-beneficiary reversion can therefore produce different effects from the same couple-level dependence change.

Second, actual spousal dependence has empirical support beyond a single pricing study. Dependent mortality in annuity valuation, multistate mortality methods, bereavement and unobserved heterogeneity evidence, and age-gap-dependent association estimates identify distinct features of the problem. Their samples, likelihoods, assumptions and contract types differ, so their effect sizes should not be pooled mechanically.

Third, household dependence and systemic mortality uncertainty occupy different layers. Correlation within insured couples changes the first-death and survivor-state probabilities. Correlated mortality trends across many couples affect the entire liability portfolio and may be driven by longevity improvement, epidemics or shared macroeconomic conditions. A reserve or capital analysis that merges these into one undifferentiated dependence parameter cannot adequately attribute sources of risk.

Fourth, legal entitlement rules and payment definitions constrain the valuation. The U.S. qualified joint-and-survivor framework and UK dependant annuity guidance make clear that a contract's valid beneficiary and surviving payment fraction cannot be deduced from mortality data. A dependence model evaluates the entitlement that the plan or contract actually creates.

Fifth, data provenance is substantive actuarial evidence. The SOA's correction to joint-life exposure reporting shows why exposure counting, secondary life records and amount weighting must be inspected before interpreting observed mortality or dependence coefficients.

Unresolved empirical questions

Several questions merit further research with appropriately consented and governed paired-lifetime data: the stability of within-household dependence in modern annuity portfolios; the identification of bereavement effects separately from caregiving and latent health conditions; changing dependence among nonspousal designated beneficiaries; the relationship between household common shocks and population-level mortality crises; and the effect of dependence calibration on contractual pricing errors across alternative age gaps, continuation fractions and guarantee periods. More detailed study is also needed of how income discontinuities after the first death interact with actual living expenses, social insurance entitlements and behavioral selection into particular annuity options.

Testing these questions requires actual matched observations and legally accurate benefit specifications. The hypothetical results here establish what follows from the chosen payout functions and mortality assumptions. They do not establish empirical magnitudes for any specific pension system.

Scholarly and editorial scope

The original analytical contribution is the integrated contract-state sensitivity derivation, the explicit reconciliation of statistical and causal forms of dependence with entitlement rules, and the transparently reproducible common-shock comparison. The underlying probability identities and mortality mechanisms belong to established actuarial science. No claim of a newly discovered stochastic law, original mortality theory, or field-data innovation is made. This research dossier is designed for cross-disciplinary use in Annuities and Retirement Insurance, Actuarial Science, and Life Insurance, while retaining one canonical scientific record.

Conclusion

Dependence between two covered lifetimes is an essential determinant of the survival states that trigger joint-life retirement benefits. A valuation based only on two marginal mortality tables leaves the joint distribution unspecified. For a participant-first continuation annuity, the conditional economic exposure includes the possibility that the beneficiary is alive when the participant has died; for a both-alive contract, it is the simultaneous-survival state; for a last-survivor annuity, it is the event that at least one remains alive. Each state has a distinct dependence sensitivity derived exactly from its cash-flow promise.

The actuarial literature establishes multiple empirically relevant channels: persistent household heterogeneity, shared shocks, postbereavement mortality changes, age-gap covariates and dynamic mortality dependence. The scientific and commercial implications depend on the selected population, observation history, discounting and contract terms. The reproducible example quantifies a controlled difference between independence and an explicit common-shock dependence structure, with equal marginal hazards and a stated payment schedule. Its result is illustrative and internally consistent.

International application requires the same rigor in institutional interpretation. Benefit entitlements are established by the valid contract and governing law; funding and reporting are established by the actual obligation holder and applicable accounting framework. The durable conclusion is that credible joint-life retirement valuation begins with a verified payment-state specification, estimates the dependence supported by matched evidence, and reports mortality, contractual and financial sensitivities separately. This sequence improves scientific transparency, administrative auditability and interpretation of survivor income security.

References