RESEARCH DOSSIER
Longevity Risk in Retirement Portfolios
Abstract. Longevity risk in retirement portfolios arises when the number, timing, or duration of future retirement-income payments exceeds the assumptions used to fund them. At the portfolio level its defining feature is dependence: improvements in survival experienced jointly by many beneficiaries persist after the idiosyncratic variation in individual lifetimes has been diversified. This monograph develops an integrated actuarial, statistical, economic, and institutional account of the exposure. It distinguishes an individual's risk of outliving assets from the liability-side risk borne by pension plans and life-annuity providers; formalizes period, cohort, and systematic mortality uncertainty; derives present-value and covariance expressions for life-contingent payments; and explains how discounting, benefit indexation, survivor benefits, participant selection, and mortality-model uncertainty interact with long-duration obligations. A reproducible, entirely hypothetical 45-year cohort calculation illustrates both a 7.30% increase in the present value of fixed annuity payments following a uniform 20% reduction in annual death probabilities and the persistence of a common-state variance floor as portfolio size rises. The analysis then compares the capital and governance implications of longevity stress testing, annuity underwriting, pension asset–liability management, buy-ins, buy-outs, longevity insurance, customized swaps, and index-linked hedging. Institutional evidence is drawn from demographic databases, the actuarial literature, the OECD, the Continuous Mortality Investigation, U.S. Social Security projections, and the European insurance and United Kingdom pensions frameworks. The result is a source-grounded interpretation of longevity as a cross-generational balance-sheet exposure whose financial consequences depend on the identity of the risk bearer, the terms of the benefit promise, the joint mortality distribution, and the effectiveness of any transfer arrangement. The work provides original derivations, reconciliations, and quantitative illustrations without asserting a new theoretical system or empirical discovery.
Keywords: Longevity risk; systematic mortality; retirement portfolios; pension liabilities; life annuities; cohort mortality; mortality improvement; stochastic mortality; survival dependence; actuarial present value; asset–liability management; pension solvency; longevity swaps; longevity hedging; population basis risk; model uncertainty; demographic scenarios; retirement income.
Introduction
A retirement-income promise is a claim on future resources conditional on the survival of one or more persons. Its financial value therefore depends on mortality experience across years when payments are due, on interest rates and assets available to meet those payments, and on the specification of benefits. Longevity risk becomes financially material when realized or newly forecast survival is higher than the survival embodied in the original funding or valuation basis. An individual may bear this exposure as a shortage of resources during an unusually long life; a defined-benefit pension plan or a life-annuity insurer bears it as unexpectedly persistent payment obligations. Analysis must identify the risk bearer before describing the direction of loss.
The central economic problem is the joint survival of many people. The independent component of individual lifetime uncertainty can be pooled. Common changes in medical conditions, behavior, technology, social circumstances, and the demographic trajectory may affect entire birth cohorts, however, so that increasing the number of beneficiaries cannot remove this part of the liability. The OECD study of mortality assumptions documents the practical consequence for pension funds and annuity providers: mortality assumptions that insufficiently allow for future improvement can leave inadequate provisions. The question is therefore both probabilistic and institutional: what is the joint distribution of future benefits, and who is obliged to finance the unfavorable states of that distribution?
Longevity has become a major subject in demographic and financial research because the same observable facts can support very different valuation outcomes. A higher expected remaining lifetime may change the price of a new annuity, the funded status of an existing scheme, the size of the reserve held by an insurer, and the retirement consumption possibilities of a household. The OECD's Pensions at a Glance 2025 reports that, across OECD members, remaining period life expectancy at age 65 in 2024 averaged 18.5 years for men and 21.6 years for women. It also projects further gains to 2065, with substantial international differences. These summary indicators describe demographic conditions; a valuation of a specific retirement portfolio additionally requires the portfolio's own mortality basis, benefit terms, cash-flow horizon, and discounting assumptions.
The methodological foundation combines deterministic cash-flow valuation, conditional-probability arguments, stochastic mortality research, official demographic descriptions, pension and insurance guidance, and a transparent numerical experiment. The foundational Lee and Carter (1992) mortality study established a widely used age–period forecasting framework. The later Cairns, Blake, and Dowd (2006) model addressed stochastic mortality at older ages and the valuation of aggregate longevity exposure. Neither a single fitted model nor a single stress test can settle uncertainty about distant mortality patterns, especially when the decision concerns liabilities spanning several decades.
The quantitative examples in this monograph use declared hypothetical assumptions rather than administrative observations or calibrated forecasts. They provide analytical counterexamples and reproducible arithmetic. Regulatory comparisons distinguish formal legal requirements from illustrative analogies. Bibliographic references disclose when full-text examination was available and when publisher abstracts or institutional summaries were the attainable verification boundary. The work is an international scientific study; differences in jurisdictional law, pension structure, mortality selection, and demographic data coverage remain part of its subject matter.
Analytical scope and method
The unit of financial analysis is a portfolio of retirement-income promises. A benefit is included when its present value depends on the survival of the person or persons specified in its contract. A defined-contribution account without a lifetime payment guarantee gives the individual the main financial consequences of extraordinary survival; a lifetime annuity moves an agreed component of that exposure to its provider. Defined-benefit liabilities depend on survivors' entitlements, contribution covenants, investments, funding rules, and permitted adjustment mechanisms. Social insurance systems face a related demographic and fiscal challenge but commonly operate under different financing and policy-adjustment arrangements.
The evidentiary method has four steps. First, basic probabilistic identities are derived under explicit assumptions. Second, their financial expressions are compared across contract forms and risk bearers. Third, observed institutional evidence is attributed to dated sources and its measurement basis is identified. Fourth, hypothetical calculations are used to isolate mechanisms that cannot be established by a country average alone. Numerical amounts are presented in nominal U.S. dollars for arithmetic convenience and imply no recommendation, market price, or observed portfolio estimate.
The chapter sequence follows the already approved scientific passport for this Research Dossier: Individual versus Systematic Longevity; Cohort Dependence; Aggregate Annuity Liabilities; Long-Term Capital Impact; Longevity Risk Transfer; and Scenario Robustness. Mathematical derivations and comparative evidence are connected to each claim rather than collected into unspecific institutional citation chains.
Individual versus Systematic Longevity
Risk bearers and contractual definitions
The meaning of longevity risk follows from the obligation whose value is being measured. For a retired household holding finite liquid wealth, unusually high survival can create a consumption shortfall after accumulated assets have been drawn down. The probability of such a shortfall depends on spending, investment returns, inflation, health-related expenditures, transfers, and the retirement horizon. At the insurer or pension-plan level, an unusually long lifetime ordinarily increases the value of benefits promised to an annuitant. The household and the provider thus hold opposite exposures to the same lifetime event in the stylized case of a fixed lifetime annuity.
Let denote the residual lifetime of person at retirement and let denote available resources at future time . A household's longevity shortfall event can be expressed as together with under a specified consumption and asset-return policy. For a provider, define , the indicator that the individual qualifies for a payment at time . Each year's survival indicator produces a contingent benefit. This direct definition accommodates a fixed life annuity, an inflation-indexed pension, a survivor annuity, and combinations of benefits when the relevant contingencies are specified.
Three distinctions organize the study. Individual longevity variability concerns the chance that one lifetime differs from an estimate. Systematic longevity variability concerns joint deviations of survival across a portfolio, particularly from a shared shift in mortality trajectories. Estimation and model uncertainty concern the extent to which the parameters or structure used to assign probabilities may differ from the mechanism that actually produces future mortality. The last distinction matters even when the portfolio is very large: a massive dataset can measure yesterday's rates precisely while leaving the shape of future mortality improvement uncertain.
The OECD's institutional assessment treats longevity risk as a provision and governance problem for pension funds and annuity providers. That analysis is consistent with the actuarial exposition by Pitacco, Denuit, Haberman, and Olivieri (2009), which frames future mortality improvements and uncertainty as determinants of pensions and annuity business. The broader consequence is that a mortality assumption is simultaneously a demographic forecast and a financial input. An optimistic or pessimistic forecast cannot be evaluated separately from the product whose promised payment stream is being valued.
Pooling and the surviving common component
Consider pensioners receiving identical lifetime benefits. If the present values of their benefits were mutually independent with a common finite variance, the standard deviation of total liability would increase in proportion to , while its mean would increase in proportion to . Consequently, the coefficient of variation of the portfolio's average benefit would decline approximately as . This is the familiar risk-pooling mechanism.
The assumption of unconditional independence is implausible for multi-decade longevity exposure. Members of the same birth cohort experience common calendar time, widely shared medical technologies and public-health changes, and similar aggregate economic or environmental circumstances. Conditional independence given those shared circumstances may be a useful approximation, but it does not make future survival unconditional independence. Once the shared environment is integrated out, dependence reappears even if no two individual deaths directly cause each other.
The two-factor mortality analysis of Cairns, Blake, and Dowd (2006) provides a formal reason to expect exposure across older ages to common time-varying factors. Its central subject is the development of the post-age-60 mortality curve in the United Kingdom and the pricing implications for longevity instruments. It does not imply that the shape or size of these factors transfers unchanged to every retirement population. The useful general result is that a large portfolio can reduce deaths' idiosyncratic noise while remaining sensitive to mortality-curve shifts shared across ages.
A pension institution can therefore face a higher absolute financial loss from a small systematic survival change than from a larger relative change in one member's individual experience. A cohort with many similar annuity promises and a common longevity shift has a concentrated exposure to the direction of that shift. Combining many cohorts, countries, occupations, ages, and sexes may diversify parts of the demographic risk, but the degree of diversification follows from their joint distribution and benefit weights. It cannot be inferred from the mere number of people or from geographic labels.
Retirement products and the direction of exposure
A defined-contribution savings account is primarily a stock of invested assets. If it is spent without a lifetime-income guarantee, the individual carries the risk that the retirement horizon is longer than anticipated. Periodic withdrawals may be adjusted to investment performance and remaining expected life, but financial flexibility changes the distribution of consumption rather than eliminating uncertainty about lifespan.
A life annuity converts a stated capital contribution into payments governed by survival. The purchaser obtains protection against longevity-related depletion of that covered income stream. The provider accepts additional payment obligations when annuitants, in aggregate, survive longer. The provider may partially offset that liability with mortality-sensitive assets or other insurance obligations, and may reinsure or hedge the remainder. The size of the net exposure depends on contract design, expenses, reserving, selection, asset duration, reinsurance protection, and its own capital base.
For a defined-benefit pension scheme, the legal promise typically specifies a pension formula and payment contingencies. The scheme's longevity sensitivity depends on the active-to-retired composition, normal and early retirement terms, cost-of-living adjustments, spouses' benefits, commutation options, and funding arrangements. Some promises are fixed; others adjust according to a rule or depend on policy decisions. A cash-flow forecast that models only the principal pension and omits survivor and indexation obligations cannot represent the full liability.
A hybrid arrangement may divide exposure among participants, sponsors, insurers, and public protection systems. When benefit adjustment rules respond to life expectancy or funded status, the allocation of longevity risk changes over time. It is thus appropriate to identify both the primary legal obligor and the eventual financial bearer of an unfavorable survival realization. In economic analysis, longevity risk is an allocation problem as well as a forecasting problem.
Period and cohort longevity
Period life expectancy imagines exposure to the age-specific mortality rates prevailing in one calendar period. Cohort life expectancy follows an actual or hypothetical birth cohort through successive future ages, incorporating mortality assumptions for those future years. These are different estimands, and the distinction is essential for retirement liabilities extending several decades.
If is the probability of death between exact ages and for a person aged in calendar year , a cohort alive at age in year has projected survival to age given by
The benefit value depends on the diagonal path through the age–calendar-year surface, not merely on the period table for . A forecast that holds current rates fixed over future years embeds a particular mortality-improvement assumption, even if its author does not label it as one. The same observation applies when a projection uses an actuarial improvement scale or a stochastic demographic model.
The 2026 U.S. Trustees cohort tables distinguish intermediate, low-cost, and high-cost assumptions for remaining cohort life expectancy. The table is useful evidence that public retirement forecasts incorporate multiple demographic scenarios. Its program-specific assumptions do not automatically represent a private insured annuity portfolio. Similarly, the OECD's 2025 pensions methodology uses country-specific prospective mortality projections in stylized defined-contribution annuitization calculations. The two resources demonstrate distinct institutional uses of cohort projection, not a single universal mortality basis.
Demographic evidence and scope
The Human Mortality Database overview describes harmonized age-specific rates and life tables based on vital-statistics, census, and population data for countries with sufficiently complete records. Its transparency facilitates demographic comparison and the evaluation of projection methods. It also establishes a coverage boundary: the countries included are selected for data quality, and national population rates need not describe selected annuitants, particular industries, or financially advantaged pension cohorts.
The CMI_2025 explanatory note makes the distinction between current mortality tables and assumptions about future improvement explicit. The Continuous Mortality Investigation blends short-term observed improvement with user-selected long-term improvement assumptions. It emphasizes that users should assess the suitability of starting assumptions for their actual portfolios. The CMI framework is relevant beyond the United Kingdom as a methodological example; its estimated parameters and data refer to England and Wales.
International longevity gains are heterogeneous across countries, sexes, ages, cohorts, and historical periods. The OECD's 2025 life-expectancy chapter also notes a slowdown in the rate of gains at older ages since the earlier period of rapid improvement. A slower recent trend does not settle the future rate for a particular retired cohort. The actuarial implication is the need to test multiple defensible trajectories and to attribute the results to the specific demographic and portfolio populations used.
The main finding of this chapter is a contract-sensitive distinction: individual retirement longevity is uncertainty about the duration for which resources must sustain a life, while systematic portfolio longevity is uncertainty about the joint duration of payments promised to many lives. Their link is the survival process; their financial consequences depend on who guarantees income and who retains the residual obligation.
Cohort Dependence
Survival filtration and probability structure
A retirement portfolio is exposed to a vector of future survival indicators rather than to a single random life expectancy. Let denote an entire stochastic demographic path: a joint description of future common mortality factors across calendar years and ages. For pensioner , let if the entitlement condition requiring survival at payment date is fulfilled, and otherwise. Given and the person's observable characteristics , define
The annual survival probabilities are determined by a mortality law appropriate to age, time, cohort, and member characteristics. A common demographic path can alter all of these conditional survival probabilities simultaneously. An individual's death time then reflects both the common path and the residual individual randomness around that path.
Conditional independence of different individuals given is a useful benchmark, especially for formal sensitivity calculations. It is a modeling assumption rather than a natural law. Close relatives, shared occupations, local epidemics, and group-specific dependencies can create residual association after a broad national mortality factor is accounted for. Conditional independence should therefore be tested when the portfolio is concentrated, when lives share salient circumstances, or when dependent joint-life contracts are material.
For a single life, payments at successive dates are intrinsically dependent: survival to date entails survival to date . Conditional on one fixed mortality scenario, the covariance of survival indicators is
The formula follows by observing that the product equals the indicator of survival to the later date. A valuation of a life annuity cannot treat payments to the same person in different future years as independent Bernoulli events. That mistake understates lifetime payment variability even when mortality for different persons is independent.
A variance decomposition for life-contingent portfolios
Suppose each member has a discounted lifetime payment random variable
where is the deterministic discount factor and is the fixed payment amount at date , absent further contractual contingencies. The aggregate present value is . Conditional on the common demographic path, write and . The law of total variance gives the exact identity
When and are conditionally independent for distinct members, the first term reduces to . The second becomes
The off-diagonal elements are economically critical. Common mortality improvements raise the conditional expected values of many annuity obligations together, producing positive covariance under the stipulated exposure. Where annuity payouts, ages, populations, or benefits differ, the covariance can vary in size or direction. Total variance is therefore an attribute of the entire portfolio and its common-state exposure.
For conditionally independent, exchangeable members, let and be the shared conditional moments. Then
Dividing by yields
The first term converges to zero as the portfolio grows, subject to the stated finite-variance conditions. The second term persists wherever common demographic uncertainty materially changes the conditional mean benefit. This is the analytical foundation of the systemic longevity component. Its existence follows from a standard conditional-variance identity and the explicit assumptions, and this monograph makes no novelty claim for the mathematical decomposition.
For a heterogeneous portfolio with having different benefits and characteristics, the exposure is naturally summarized by the full covariance matrix. A benefit-weighted portfolio concentrated in one cohort may have a larger systematic component than a larger, geographically dispersed portfolio, even if both contain many people. The relevant diversification criterion concerns the behavior of relative to under changing portfolio composition and common-factor loadings.
Time, age, and cohort factors
The mortality process can be described by age , calendar time , and birth cohort . Classical extrapolative models estimate observed mortality and project time-varying factors. In the Lee–Carter specification, a central death rate is represented in a stylized form by
Here describes the age pattern, describes the age-specific response to changes in the mortality index, is the calendar-time index, and is the residual. Identification requires normalization of the factor representation; standard conditions include constraints on the level of and the scale of . Forecast uncertainty arises from the future path of , estimation error, residual structure, and possible change in the age-specific response.
Central death rates and one-year death probabilities are related but distinct quantities. Under a piecewise-constant force of mortality within the year, . A model calibrated to one must apply an appropriate conversion when pricing life-contingent payments in the other. The Poisson log-bilinear treatment by Brouhns, Denuit, and Vermunt embeds mortality projection in a count model and makes exposure-to-risk denominators part of estimation. This is especially relevant when death counts and exposures vary substantially by age.
At retirement ages, the two-factor Cairns–Blake–Dowd specification employs a time-dependent level and age-slope structure for old-age mortality. A common stylized formulation is
The two factors permit future mortality movement to differ across ages, an important feature for pension liabilities whose payment horizon changes with the age composition of beneficiaries. The precise estimated parameters, fit range, and factor dynamics depend on the population and study; the equation is a mathematical illustration of the family, not a calibration of the present research portfolio.
A cohort effect can capture deviations aligned with birth year rather than calendar year alone. Introducing age, period, and cohort components requires care because the three indices are linearly related: . Identification constraints, model selection, and sensitivity to the calibration window are therefore substantive actuarial decisions. Apparent cohort effects can be unstable under revisions to data quality, pandemic-period treatment, or the allocation of trend across factors. The six-model comparison by Cairns and colleagues (2011) is instructive: satisfactory historical fit does not ensure plausible future mortality distributions, and models that meet qualitative criteria can still yield materially different long-horizon forecasts.
Statistical estimation and model validation
Mortality is observed through data-generating and data-reporting systems. Recorded deaths, estimated population exposures, and the definitions of age, calendar period, sex, and territorial coverage influence a mortality estimate. The Human Mortality Database's data principles establish the value of harmonized methods and explicit country-level documentation. They also clarify the limitation of applying general population projections to a selected book of annuitants without adjustment.
Pension and annuity populations may differ from the population at large because of occupation, socioeconomic position, income, medical underwriting, participation conditions, and selection into annuity purchase. A survival model therefore needs both a baseline mortality table and a projected improvement assumption. A technically advanced stochastic model cannot substitute for a defensible baseline population. Nor can a good recent fit erase uncertainty concerning decades of future medical and social change.
The CMI_2025 report page documents a model calibrated on England and Wales mortality through December 31, 2025. Its public methodology FAQ explains that the long-term rate is supplied by the model user rather than discovered as a uniquely determined number in past observations. The November 2026 source inspection also identified a correction notice on the CMI news page concerning certain CMI_2025 working-paper tables. Consequently, this research does not reproduce the affected numerical tables or treat them as independently verified forecasts.
The relevant diagnostics have distinct purposes. In-sample fit assesses the model's representation of observed data. Out-of-sample validation assesses performance on data deliberately withheld from calibration. Residual checks assess unexplained age–time patterns. Predictive interval coverage tests whether realized outcomes fall within stated uncertainty ranges at an appropriate frequency. Sensitivity to calibration windows and improvements after the final observation assesses parameter and extrapolation robustness. Comparison of mortality-level and mortality-improvement assumptions protects against an incoherent combination of base tables and forecast scales.
Nonlinearity, tail behavior, and identification
Life expectancy is a demographic summary; liabilities are weighted functions of survival probabilities. Two mortality paths can have similar remaining life expectancy yet produce different annuity values if the survival difference occurs in payment years with distinct discount factors, benefit indexation, or survivor provisions. Likewise, a change in mortality at early old ages can affect many payment periods, while a change at extreme ages affects fewer beneficiaries but may be important for very late-life guarantees.
The present value of a life annuity is monotone in survival when benefits and discount factors are nonnegative. Under the same cash-flow terms and discount curve, raising survival probabilities at every future payment date cannot lower its expected present value. This mathematical monotonicity does not imply a fixed percentage relationship between life expectancy and financial liability: discounting, the term structure of the mortality change, and the promised benefit pattern govern the elasticity.
Risk measures beyond variance may be relevant to long-duration obligations, including quantiles, expected shortfall, cumulative funding-deficit probability, and conditional survival to advanced ages. Their interpretation depends on the distribution generated by the model and on the valuation horizon. A 99.5% one-year capital assessment asks a different question from a 40-year probability of benefit-payment exhaustion. Both can be legitimate measurements; neither substitutes for the other.
Parameter uncertainty, model-form uncertainty, and future changes in social determinants are not interchangeable with the residual death-count noise fitted inside one mortality model. Reports should therefore disclose the model family, observations and exposure construction, calendar-year cutoff, estimation constraints, choice of long-term improvement, stress definition, discount curve, and how selection differences between populations were handled.
The chapter's central result is structural: the common-factor component of survival creates a liability variance that need not decline per beneficiary as portfolio size rises. It is the mathematically specified joint exposure—not a vague expectation of longer lives—that determines the possibility and magnitude of systematic longevity loss.
Aggregate Annuity Liabilities
Exact cash-flow valuation
Annuity liability valuation begins with the benefit promise and the survival condition for each payment. Let be the nominal amount contractually payable to member at the end of future year , conditional on their survival. Let be the applicable discount factor. For a fixed deterministic discount curve and a defined mortality basis, the expected actuarial present value is
If survival follows a common stochastic factor , the same expression is evaluated by integrating conditional survival probabilities over its distribution. Stochastic interest rates introduce further dependence and may require the joint distribution of demographic factors, discount rates, and relevant financial variables. The equation is therefore a basic contractual valuation identity, with more elaborate models needed for guarantees, dynamic options, and risk-adjusted insurance accounting.
The expectation of the sum requires only the marginal survival probabilities for each payment date; the variance and tail quantiles of the total require joint survival information. This distinction is consequential. An actuary can calculate a central best-estimate liability from projected cohort survival without specifying every cross-member covariance, but estimating systematic capital exposure or hedging efficiency requires more of the joint distribution. Mean adequacy and loss-distribution adequacy are separate tests.
A funded pension's economic position also depends on assets. Write for the market-consistent or consistently valued asset portfolio and for benefits and associated obligations. Economic surplus is . The exposure of to a mortality improvement is the asset sensitivity minus liability sensitivity. If assets are ordinary fixed-income securities, they may hedge interest-rate sensitivity without directly offsetting survival-conditional cash flows. Inflation-linked bonds may help with indexation but do not automatically hedge longevity. Mortality-linked assets, reinsurance recoverables, and appropriately designed derivative payoffs can alter the net profile.
Benefit design, spouses, and optionality
For a single-life fixed annuity, each payment is conditional on the annuitant being alive. A joint-and-survivor pension may continue at full or reduced rate after the primary beneficiary's death if a spouse or other eligible beneficiary survives. Its valuation uses joint-life and last-survivor probabilities, which can require a dependency model even without systematic longevity risk. Guaranteed minimum payment periods and death benefits add payments that may occur after the annuitant's death; these modify the direction and term of mortality sensitivity.
An inflation-indexed benefit has rising according to a contractual index or adjustment rule. Its long-duration exposure can be especially sensitive to survival at late dates because both the number of payments and the nominal size of the payment change. Optional early retirement, commutation, spousal elections, and conversion to a life annuity can be linked to participant behavior and financial conditions. The valuation model should distinguish contractual option exercise from purely biological survival.
Mortality selection also changes the distribution of the annuity liabilities. A group purchasing voluntary annuities may differ systematically in expected lifetime from the general population, while an occupational defined-benefit cohort reflects employment and retirement-system selection. The CMI distinction between base and improvement assumptions is especially important here: an improvement overlay applied to the wrong starting mortality table may produce a valuation whose numerical precision conceals population mismatch.
A reproducible mortality and discounting experiment
To isolate the role of systematic longevity, consider a hypothetical homogeneous portfolio of 10,000 individuals, each aged exactly 65 at valuation. Every surviving person receives USD 20,000 at the end of each year for at most 45 years. There are no spouses' benefits, death benefits, annuity purchase expenses, taxes, indexation, lapses, or management costs. The annual effective discount rate is fixed at 3%. At the beginning of the payment year indexed by , each member's age is . The assumed probability of death before the end of that year is
where is an explicitly hypothetical multiplier applied uniformly to all future one-year death probabilities. All values on this 45-year horizon are below one for in the tested set . Survival to payment date is
The deterministic actuarial present value per person is
The baseline is per annuitant, or approximately USD 2.8768 billion for the portfolio. Reducing all modeled annual death probabilities by 20% gives , a gain in liability of USD 21,003.12 per member and USD 210.0312 million for 10,000 members. The relative increase is 7.3009%. Increasing them by 20% yields . These are calculations under a declared stylized mortality curve, not observed experience or an estimate of the effect of an actual intervention.
| Hypothetical mortality multiplier | Annual discount rate | APV per annuitant (USD) | Total APV for 10,000 (USD billions) |
|---|---|---|---|
| 0.8 | 2% | 347,977.33 | 3.4798 |
| 0.8 | 3% | 308,681.33 | 3.0868 |
| 0.8 | 4% | 276,077.74 | 2.7608 |
| 1.0 | 2% | 321,805.57 | 3.2181 |
| 1.0 | 3% | 287,678.21 | 2.8768 |
| 1.0 | 4% | 259,049.25 | 2.5905 |
| 1.2 | 2% | 300,448.62 | 3.0045 |
| 1.2 | 3% | 270,281.87 | 2.7028 |
| 1.2 | 4% | 244,746.09 | 2.4475 |
This table exposes the interaction between long survival and the financial term structure. A reduction in discount rates increases the value of later payments, amplifying the size of longevity-sensitive liabilities. Under the illustrative assumptions, a move from the baseline to increases per-person present value from USD 287,678.21 to USD 347,977.33. These inputs are artificial scenario values. The table makes no claim about covariance between real-world interest rates and mortality improvements.
The hypothetical mortality curve should not be mistaken for an empirical forecast. The annual growth rate of 8.5% in death probability and the 20% multiplicative stresses were selected to make the arithmetic transparent. No conclusion concerning future mortality improvements in a particular country follows from the chosen values. Its legitimate finding is conditional: for the stated benefit and discount schedule, a common downward shift in death probabilities increases the liability by a quantifiable amount.
Common mortality states and portfolio-size sensitivity
A second experiment separates pooled individual uncertainty from common mortality uncertainty. Suppose the multiplier equals 0.8 with probability one half and 1.2 with probability one half. Conditional on that single common multiplier, members' lifetimes are independent and their individual survival follows the schedule above. All discount rates remain fixed at 3%. This construction deliberately assigns a common survival regime to the entire portfolio, so it is a simple diagnostic experiment, not a data-fitted stochastic mortality model.
The conditional mean lifetime benefit is USD 308,681.33 in the favorable-to-annuitants low-mortality state and USD 270,281.87 in the high-mortality state. Consequently, the unconditional mean is USD 289,481.60 per member. It is not identical to the valuation at the midpoint multiplier ; multiplication of yearly survival probabilities makes the mapping from mortality rates to present value nonlinear.
The conditional standard deviation of one member's lifetime discounted benefits can be computed from pairwise survival-indicator covariance. The general expression is
Direct summation yields conditional standard deviations of approximately USD 115,719.43 for and USD 114,015.38 for . The mean conditional variance across the two states is approximately . The variance of the conditional mean is . Thus for beneficiaries the exact theoretical aggregate liability variance under this two-state construction is
with dollar units implicit. This equation uses independent life histories conditional on the shared state. It assumes no extra dependence between members and no investment-return uncertainty.
| Number of annuitants | Mean aggregate APV (USD millions) | Liability SD (USD millions) | SD/mean | Variance from common state |
|---|---|---|---|---|
| 100 | 28.9482 | 2.2374 | 7.73% | 73.64% |
| 1,000 | 289.4816 | 19.5403 | 6.75% | 96.54% |
| 10,000 | 2,894.8160 | 192.3406 | 6.64% | 99.64% |
For large , the coefficient of variation approaches , approximately 6.63%, rather than zero. In this example, one common binary demographic regime is sufficient to prevent complete pooling. The particular numerical limit depends on the artificial probabilities and benefits; the mathematical persistence of the common term depends on the law of total variance.
A useful reproducibility check is to calculate each , sum the discounted expected payments, compute the covariance matrix of survival indicators, and substitute the resulting conditional moments into the two-state variance formula. No Monte Carlo simulation is needed for the displayed values, so they are not subject to simulation sampling error.
Reproducibility protocol
The following reference implementation uses annual integer-time survival, deterministic effective annual discounting, and standard floating-point arithmetic. It reproduces the reported expected present values and the variance decomposition to rounding precision. It deliberately has no externally fitted calibration parameters.
from math import sqrt
YEARS = 45
BENEFIT = 20_000
def valuation(multiplier, rate=0.03):
survival = 1.0
s = []
for t in range(1, YEARS + 1):
q = 0.012 * 1.085 ** (t - 1) * multiplier
assert 0 <= q < 1
survival *= 1 - q
s.append(survival)
discount_benefits = [
BENEFIT / (1 + rate) ** t for t in range(1, YEARS + 1)
]
mean = sum(v * p for v, p in zip(discount_benefits, s))
variance = sum(
discount_benefits[t] * discount_benefits[u] *
(s[max(t, u)] - s[t] * s[u])
for t in range(YEARS) for u in range(YEARS)
)
return mean, variance
low = valuation(0.8)
high = valuation(1.2)
mean = (low[0] + high[0]) / 2
within = (low[1] + high[1]) / 2
between = ((low[0] - high[0]) / 2) ** 2
for n in (100, 1000, 10000):
sd = sqrt(n * within + n * n * between)
print(n, n * mean, sd, sd / (n * mean))
Asset–liability matching and valuation interpretation
Retirement liability valuation is frequently accompanied by an investment policy that attempts to finance projected payments. Duration matching, cash-flow matching, inflation-linked investments, and contingent assets answer distinct parts of the problem. A bond portfolio with matching discounted cash-flow timing may protect against certain yield-curve changes yet remain short when additional pensioners survive and continue drawing benefits. A longevity hedge may offset some of that persistence while leaving interest-rate, inflation, sponsor, and counterparty risks.
The OECD pension modeling assumptions provide a useful illustration of how expected returns, charges, and the annuity conversion basis alter projected retirement income. Their assumed real long-run investment return and annuitization factor belong to a comparison methodology, not to a guarantee or a universal expected return for private portfolios. Comparing a liability estimate to assets therefore requires consistent prices, currencies, inflation bases, and risk assumptions.
The aggregate-liability chapter establishes three separate quantities: the expected cost of the promise, the uncertainty about that cost, and the assets or risk-transfer claims available against it. The numerical example quantifies the first two under explicit hypothetical assumptions. Solvency, fairness, and economic desirability require an additional analysis of funding, contract enforcement, regulation, and the distribution of outcomes among stakeholders.
Long-Term Capital Impact
Funding, reserve revisions, and economic surplus
When survival expectations improve, an unchanged life-contingent pension promise typically extends further into the future. The immediate balance-sheet consequence can arise before a single additional payment is made: a revised projection raises the present value of promised benefits. A second consequence occurs over subsequent decades if more beneficiaries than expected remain eligible for payments. The first is a valuation and funding event; the second is a realization of extended cash-flow obligations. Both matter, and their timing may differ.
Let the economic funding ratio be , where is consistently measured assets and is the present value of retirement obligations. If asset value remains unchanged and longevity assumptions increase liability from to , then
For an initial portfolio with , a 7.3009% liability increase like that in the illustrative experiment reduces the funding ratio from 100% to approximately 93.20%, all else equal. The decrease in funding ratio is not 7.3009 percentage points, because the denominator changes. In a real scheme, assets, collateral, reserves, technical provisions, and contribution policies may respond simultaneously, but the simple calculation demonstrates why a revised mortality forecast can create immediate financing pressure.
The longer the liability duration and the greater the weight of survival-contingent payments at advanced ages, the more important the projection horizon becomes. A comparison of one-year cash flows alone can miss exposure accumulated in years 20 through 40. The financial meaning of a projection error must therefore be evaluated using the complete liability cash-flow schedule. It can also vary with the shape of the mortality improvement: an improvement affecting ages 65–75 changes the probability of reaching many later payments; one affecting only ages above 100 has a different cash-flow impact.
The OECD report on mortality assumptions documents the need for funds and annuity insurers to incorporate future mortality improvements into provisions and calls attention to regulatory incentives and good practice. Its analysis is especially relevant because underfunding from mortality optimism or an outdated basis can coexist with apparently successful short-horizon investment performance. Portfolio performance and liability projection quality must be assessed together.
Discount-rate and longevity interaction
An annuity present value is simultaneously exposed to survival probabilities and discount factors. For positive benefits and a flat annual effective discount rate ,
If survival probabilities increase at fixed , increases. If declines at fixed survival and all payments are nonnegative, also increases. The joint stress can therefore materially exceed either isolated movement, although actual economic correlations between discount rates and mortality cannot be inferred from this monotonicity alone.
The illustrative table in Aggregate Annuity Liabilities shows the combined effect of a 20% downward death-probability multiplier and a one-percentage-point lower annual discount rate. For the declared cohort, this raises the value per annuitant from USD 287,678.21 to USD 347,977.33, an increase of approximately 20.96%. The result is a conditional deterministic calculation under two simultaneous changes, not a probability statement about their co-occurrence or a measure of regulatory capital. Asset price responses and portfolio hedges would alter the net surplus consequence.
With a general discount curve, it is useful to distinguish liability cash-flow duration, liability interest-rate sensitivity, and the mortality-weighted distribution of payments. A long survival tail shifts expected payments farther into the future. For an inflation-adjusted pension, the mortality tail and the indexation rule can reinforce one another. Conversely, the existence of certain death benefits, finite payment guarantees, or survivor arrangements can make the sensitivity less uniform than for a single-life annuity. These are properties of the promises, and they should be projected at a benefit-cell level before aggregating capital exposure.
An asset–liability management program may hold cash-flow-matched bonds, inflation-linked securities, diversifying investments, and longevity transfer contracts. Each has a different relationship with the liability. Interest-rate hedging can reduce one source of surplus volatility while leaving systematic survival changes unhedged. A longevity swap can reduce survival variance while introducing cash-flow and counterparty exposures. Successful capital management depends on the net joint sensitivity rather than the sophistication of any one instrument.
Capital stress and regulatory scope
The European Solvency II regime offers a clearly defined insurer longevity stress. Under Article 138 of Delegated Regulation (EU) 2015/35, the standard-formula longevity capital sub-module uses the loss in basic own funds that would follow an instantaneous permanent 20% reduction in mortality rates used for technical provisions. The shock applies to policies for which falling mortality rates raise relevant technical provisions without the risk margin, subject to the article's conditions. That is a legal formula in a specified regulatory system, rather than a universal economic assessment of future mortality improvement.
The monograph's 20% numerical mortality multiplier was selected partly to illustrate a clear low-mortality perturbation, but its toy output is not a Solvency II capital figure. The experiment values a fixed benefit under a flat discount rate and simplified annual death probabilities. Solvency II valuation depends on applicable risk-free term structures, technical-provision and own-funds definitions, reinsurance recoverables, tax effects, and additional regulatory specifications. A calculated liability increase under a similar percentage mortality stress cannot be presented as the insurer's legally reportable capital requirement.
Capital and funding rules for private pensions differ from insurer solvency frameworks, and the applicable national pension law determines whether a defined-benefit scheme may alter accruals, indexation, contributions, investment strategies, or benefit promises. The UK Pensions Regulator's 2025 guidance explicitly discusses longevity insurance in the context of defined-benefit endgame and risk management. It observes that such insurance can transfer a portion of survival exposure while leaving interest-rate and inflation risks with the scheme. This is an example of jurisdiction-specific supervision, not a globally binding pension rule.
Solvency calculations also depend on the time horizon of the capital measure. A one-year supervisory shock, the lifetime present value of benefits, the ultimate payment horizon, and a probability of cash exhaustion answer distinct questions. The portfolio's mortality process may be uncertain over 40 or 50 years even where next year's expected payment count is comparatively predictable. Effective oversight connects the solvency horizon to the long-run liability while preserving the distinction between the statistical models used in each.
Pension plans, annuity insurers, and public schemes
A private defined-benefit plan is exposed to the present value of promised retirement and survivor benefits. Its financial strength depends on funding assets, sponsor support, funding and recovery rules, and the allocation of risks within the contract. The same demographic path can have different consequences for a closed pensioner-heavy scheme and an open scheme with active members, new contributions, and a long horizon for portfolio turnover. A closed plan may have particularly concentrated longevity exposure because the benefit population evolves without compensating new accruals or contributions.
A life-annuity insurer typically underwrites or assumes the risk of lifetime payments under an insurance contract. Its asset–liability matching, reserves, risk margin, solvency capital, reinsurance, and policyholder protection arrangements influence its resilience. Mortality improvement may also affect the insurer's life-protection book in the opposite direction: fewer deaths can reduce death-benefit payouts for some term-life or whole-life insurance liabilities. Netting such effects requires consistent product classifications, contract cash flows, and recognition of correlations. An annuity insurer is not automatically unhedged in its entire balance sheet simply because it issues annuities.
A public pay-as-you-go pension system links benefits to current and future contributions, demographics, tax capacity, and policy rules. A longer-lived population may increase total expected benefits, but the system's financing response includes workforce developments, migration, labor participation, benefit indexation, tax changes, and statutory retirement-age mechanisms. The 2026 Social Security Trustees cohort projections illustrate how future mortality affects program assumptions across low-, intermediate-, and high-cost variants. Private annuity pricing cannot be transferred directly to a public pension financing analysis without acknowledging the different balance-sheet structure.
For individuals, the key issue may remain income adequacy rather than the solvency of the institution holding their assets. A participant whose employer pension is adequately capitalized may still face uncovered late-life care costs or uncertain household expenditures. A funded retirement portfolio can have a high expected return yet fail a household when withdrawals coincide with adverse market outcomes and a long life. These related exposures require integrated retirement-income planning in research, while the present dossier's primary object remains systematic liability-side longevity.
Governance, risk appetite, and monitoring
A governed longevity program requires a documented benefit inventory, accurate member and beneficiary data, a mortality base selected for the covered population, defensible improvement assumptions, a transparent mapping from mortality factors to liability cash flows, and a strategy for assets and transferred risk. Assumption choice should be attributable and contestable through alternative valuations. Long-term uncertainty should be incorporated into funding and capital discussions without turning a single deterministic path into a probability guarantee.
A defensible reporting sequence includes baseline liability, mortality-only sensitivity, interest-rate-only sensitivity, joint sensitivity, model family comparisons, and residual exposure after transfer. It should also show effects on the actual funding or capital measure used by the institution. The board needs to know whether a hedging contract lowers expected payments, economic tail loss, regulatory capital, or a selected surplus-risk metric. These outputs are related but quantitatively distinct.
The CMI's 2026 research updates underscore the importance of data revisions and technical corrections, including the issuance of corrected versions of some CMI_2025 working-paper tables and ongoing work on base-table and improvement consistency. Governance therefore includes version control of the demographic projection system, a record of changes in source inputs, and a clear decision trail when models are re-estimated. A model's prestige does not protect an institution from the practical consequences of using a superseded table or misinterpreting a statistical output.
Capital management ultimately evaluates whether the institution can continue honoring its promises under financially adverse survival scenarios. Accurate technical provisions, adequate assets, reliable counterparties, and transparent obligations are complementary conditions. The relevant measure of long-term financial strength is the system's ability to sustain the contingent payments, not simply the sophistication of the mortality forecast reported in isolation.
Longevity Risk Transfer
Insurance transactions and retained obligations
Longevity transfer arrangements alter the institution bearing some of the financial consequence of survival improvements. Their legal form determines which liabilities remain with a pension scheme, which are assumed by an insurer or reinsurer, and which remain exposed to counterparty performance. The relevant alternatives include a bulk annuity buy-in, a buy-out, customized longevity insurance or longevity swaps, reinsurance of annuity liabilities, and financial contracts referencing external mortality or survival indices. These mechanisms have different consequences for assets, benefit obligations, collateral, investment flexibility, and member security.
A pension buy-in generally involves purchasing an insurance contract held by the scheme to provide cash flows corresponding to covered benefits. The pension scheme retains obligations to its members, subject to the applicable legal structure, while the insurance contract supplies an asset intended to fund corresponding payments. A buy-out typically involves transferring covered liabilities through insurance arrangements with the effect determined by local law and the completed contractual process. Their treatment of inflation, survivor benefits, administration, and legacy obligations must be examined contract by contract.
The UK Pensions Regulator's guidance on pension options distinguishes longevity insurance from a bulk annuity transaction. It explains that a scheme can insure survival risk for specified members while retaining assets for investment purposes, and that this does not eliminate its interest-rate or inflation risks. The terminology used here follows that regulatory description for the UK context. Similar labels in other markets may be associated with different legal and accounting consequences.
The provider accepting longevity risk must hold capital, manage it through its own asset and reinsurance structure, or transfer components to other risk bearers. A transfer removes exposure from one balance sheet only to the extent that the new counterparty performs the contract. Concentration of longevity exposure among a small number of providers may change where systemic risk is held. Understanding the economic effect therefore requires looking through the transaction to the ultimate obligations and safeguards, rather than treating execution as proof that longevity risk has vanished from the system.
Customized longevity swaps
A customized longevity swap can exchange cash flows linked to the realized benefit obligations of a defined reference population against an agreed fixed or forecast-based leg. A scheme may pay a fixed schedule to an insurer or reinsurer and receive payments matching the realized survival-contingent benefits of covered pensioners, depending on contract terms. When matching is sufficiently exact, the scheme can substantially reduce its longevity exposure for covered members while still retaining assets and other pension risks.
The scientific valuation of such a contract requires the covered lives, demographic data, benefit definitions, death notification procedures, survivor and dependants' treatment, settlement frequency, payment dates, contract duration, termination rules, and the mechanism for resolving data disputes. An arrangement referencing only the initial pension roll may differ from the actual obligations after retirements, commutations, changes in household status, or benefit adjustments. Contract architecture is an integral part of hedge effectiveness.
Suppose is the present value of the liability and is the appropriately valued hedge payoff, with increasing in states in which a payment is needed by the hedger. If the hedged exposure is , the minimum-variance static hedge ratio, assuming finite second moments and , is
The corresponding residual variance is
These are standard least-squares covariance identities. They describe an idealized variance criterion, not the full value or legal enforceability of a real hedge. Hedge price, illiquidity, collateral, counterparty default, capital requirements, stochastic discounting, tail dependence, and mismatch between present-value and realized annual cash flows can all make the economically chosen hedge ratio different from the minimum-variance result.
A customized hedge can align closely with the sponsor's actual pensioner benefits, but data demands, operational complexity, termination provisions, and counterparty concentration become significant. Even an exact-survivor hedge of today's covered pensioners does not necessarily protect future entrants, accrued but deferred benefits, or uncertain benefit-option exercise. A transaction-specific mapping is required before calling it a full longevity hedge.
Index hedges and population basis
Index-linked longevity instruments refer to an external population or published mortality statistic. Their appeal is that a standardized reference could offer broader participation, more transparent measurement, and potentially simpler transfer mechanics than a bespoke contract. Their primary scientific challenge is population basis risk: the survival experience of the liabilities being hedged can differ from that of the reference population, including in the trajectory of future mortality improvement.
The paper Longevity Hedging 101 by Coughlan and colleagues (2011) evaluates basis risk through relationships between insured or assured lives and a wider national reference population. It considers calibration of a static hedge and the consequences of demographic differences. The study's conclusion that effective index hedging can be achievable under appropriate calibration is conditional on those relationships; it is not a universal hedge ratio for any pension plan.
Li and Hardy (2011) compare different two-population mortality models for Canada and the United States. Their case-study estimates of hedge effectiveness differ by risk metric, including 56% on a longevity-value-at-risk basis and 81.61% on a longevity-risk-reduction basis. The figures concern their selected cohorts, index and methodological assumptions. They demonstrate why a single unqualified percentage of hedge effectiveness is inadequate: the risk measure and population connection are part of the reported result.
The decomposition by Cairns and colleagues (2014) identifies population basis risk, recalibration risk, and cohort uncertainty as important factors affecting the correlation between an index hedge and the liabilities it is meant to protect. Its specific case study also examines how the number of individuals affects residual death-count noise, with smaller populations more exposed to that component. The authors' findings support index hedging as a possible component of pension risk management, provided its actual population and parameter basis are analyzed.
An index may track the correct broad age group while still failing to track the liability's cash-flow weights. A pension fund can have materially different proportions of high-benefit survivors, male and female members, affluent pensioners, disability retirees, or late-retirement members compared with the national index. If members with larger pensions also exhibit different longevity trajectories, even a precise prediction of the count of survivors need not deliver precise protection of monetary benefits. This distinction is sometimes called value basis risk rather than merely demographic basis risk, but the assessment here rests on the actual weighted cash flows.
Pricing, collateral, counterparty risk, and residual risk
The financial price of longevity transfer may include expected payout value, uncertainty compensation, capital costs, expense allowances, counterparty credit charges, and transaction-specific terms. In a deep liquid market, observable prices might help identify market compensation for systematic longevity. In a thin or customized market, a model's price of risk can be particularly dependent on assumptions, available risk-bearing capital, collateral rules, and negotiations.
The Cairns–Blake–Dowd (2006) study investigates longevity bonds and explicitly considers parameter uncertainty alongside stochastic mortality. The chapter by Coughlan (2012) surveys longevity indices and capital-market transfer solutions, including transactions known publicly at the time. These research works establish the analytical instruments and historical development; they do not supply a current tradable price for the hypothetical contract in this monograph.
A pension plan paying fixed premiums and receiving survivor-dependent benefits replaces some mortality uncertainty with performance exposure to a counterparty. Collateral arrangements, posting frequency, investment of collateral, default closeout, replacement cost, and the possibility of a sharp revaluation can all affect the net risk. If premiums increase following a contractual reset or if a hedge requires substantial margin when asset values decline, liquidity needs may arise even when the final lifetime hedge payoff is favorable.
A derivative referencing an external mortality index adds data-publication and index-governance risk. Index rules may be revised, reported deaths may be delayed, population definitions may change, and settlement dates may be later than actual pension payments. Index-linked insurance can also be exposed to regional or demographic divergence between the covered portfolio and the reference group. The hedge basis should be evaluated under both central forecasts and adverse joint mortality scenarios, rather than only through the historical Pearson correlation between two mortality series.
A realistic evaluation includes at least four dimensions: expected economic benefit, tail-risk reduction, capital and accounting effect, and implementation resilience. A contract can improve one and worsen another. For example, an expensive hedge may substantially lower surplus variance but increase expected costs; a low-cost index hedge may have attractive average correlations but leave unacceptable basis risk in the financially most adverse mortality states. The economically appropriate solution follows from the institution's obligations and risk tolerance.
Contract-design and international considerations
A sound transfer transaction should establish the identity and integrity of the covered population, the exact definition of survival, the actuarial conversion of status into payments, the applicable currency and benefit escalation terms, and the permitted treatment of corrected administrative records. The contract should allocate responsibility for errors in birth dates, beneficiary identity, spousal status, and reported deaths. It should also state how the parties handle future changes to a reference index and whether the insurer or reinsurer can terminate, recapture, or commute the obligation.
International comparability of longevity transfer remains constrained by mortality data coverage, regulatory recognition, accounting rules, capital regimes, and legal doctrines governing pension promises and reinsurance claims. The United Kingdom's development of longevity insurance is relevant evidence of a market mechanism; an equivalent contractual form elsewhere may encounter different data-protection, security, or pension-benefit requirements. The scientific work of Pitacco and coauthors (2009) places these transfer problems within the broader valuation of pensions and annuity liabilities under uncertain mortality, which remains the common actuarial foundation.
The chapter's conclusion is a net-risk interpretation: longevity transfer is effective to the extent that a legally enforceable and financially performable payoff offsets the actual survival-contingent obligations. Its success is measured after accounting for residual population basis, benefit mismatch, counterparty risk, cost, and the other risks that remain on the pension or insurance balance sheet.
Scenario Robustness
Demographic states and diagnostic experiments
A longevity risk report is robust when its conclusions remain interpretable under reasonable changes in demographic and financial assumptions, and when material changes in those assumptions produce visible, explained changes in its outputs. Robustness is a property of the analysis and its reporting process. It should be demonstrated through transparent sensitivity tests, competing model specifications, and explicit limitations rather than asserted from the use of a single recognized mortality model.
The simplest diagnostic varies one component at a time. A mortality-only stress tests a change in projected survival while holding benefit definitions and financial discount factors fixed. A discount-only stress tests the financial present value of a given expected payment schedule. An inflation or pension-escalation stress changes future benefit amounts. A joint stress changes several relevant processes together. The results should be compared on a common definition of liability, assets, and reporting date.
The illustrative model in the preceding chapter includes a complete grid of mortality multipliers and discount rates. In that controlled experiment, the per-member annuity present value ranges from USD 244,746.09 under high death probabilities and 4% annual discounting to USD 347,977.33 under low death probabilities and 2% discounting. The range is not a confidence interval or a forecast. It is an output of explicitly selected parameters. A well-governed interpretation records which input caused each difference and avoids describing the endpoints as likely or unlikely without a probabilistic model.
More demanding scenarios alter the shape of improvement rather than scaling the whole death-probability curve. Examples include improvement concentrated between ages 65 and 80; improvement concentrated above age 85; a persistent positive cohort effect; a temporary pandemic-related mortality disturbance followed by a different long-term trend; and delayed mortality improvements after a decade of stagnation. Even if two scenarios produce similar period life expectancy in one year, they may shift annuity cash flows differently because the affected ages and payment dates differ.
The OECD 2025 demographic assessment offers a factual backdrop: national and sex-specific longevity gains are varied, and recent improvement rates have changed over time. The CMI_2025 model documentation provides an example of updating mortality projections with new observations. These sources motivate scenarios but do not determine their weights for a specific insured population. A scenario grid remains a designed analytical experiment unless it has been fitted and validated probabilistically.
Alternative model families and structural uncertainty
A fitted mortality model simultaneously specifies how historical information is summarized and how its parameters evolve into the future. A Lee–Carter age–period model, an older-age two-factor model, an age–period–cohort model, a mortality-improvement projection framework, and a multi-population model can have differing factor loadings and long-term extrapolations. The appropriate comparison is between projected survival and liability distributions under a common portfolio, not solely between regression fit statistics.
The peer-reviewed comparison of six mortality models by Cairns and coauthors (2011) directly addresses the gap between historical fit and plausible density forecasts. Their study examines whether model forecasts meet qualitative criteria concerning demographic plausibility, uncertainty, and stability relative to the calibration sample. Its results support a disciplined form of model comparison: a model can fit past mortality yet behave implausibly when extrapolated to advanced ages or distant decades.
Model risk also arises from the chosen relationship among different populations. A fund's members may improve differently from national mortality, and two countries may share long-term trends while retaining different short-term deviations. The Li–Hardy basis-risk study demonstrates the practical importance of joint modeling of portfolio and reference populations in longevity hedges. A model that reproduces the national trend but misses relative portfolio improvement can mismeasure both direct liabilities and the benefits of an index hedge.
Parameter uncertainty should be separated from model specification uncertainty. Confidence intervals around a fitted mortality trend describe uncertainty conditional on the chosen functional form and assumed dynamics. They generally do not encompass unmodeled improvements in medical technologies, changes in population composition, future changes in mortality reporting, or new social patterns. A set of model families, together with explicit structural scenarios, may reveal a larger and more consequential range of outcomes than the error band of a single model. This does not automatically turn a scenario range into a probabilistic credible interval.
Mortality shocks, cohort persistence, and data revisions
A transitory mortality shock and a sustained longevity improvement affect annuity liabilities differently. Excess mortality in one year may reduce survival counts, while a long-term change in mortality improvement affects many future years. A pandemic can also influence future mortality through delayed healthcare, behavioral changes, long-term morbidity, and statistical selection of survivors. The direction of the long-term effect cannot be inferred mechanically from the sign of one year's deaths.
The CMI_2025 public FAQ explains its blending of recent experience and user-supplied long-term improvement assumptions. It also documents recent England and Wales mortality relative to the pre-pandemic baseline. Those figures describe a particular national data series, not a causal estimate of future improvement for every pension cohort. In demographic modeling, a decision to give full weight, reduced weight, or a separate overlay to shock years can materially change the starting trajectory.
The CMI's working-paper news record reported a correction to specified tables in Working Paper 211, alongside a later consultation concerning CMI_2026. This matters for the integrity of source use: the present manuscript relies on publicly verifiable explanatory material about the model and does not reproduce affected technical table values as verified findings. The report distinguishes the current CMI_2025 release from the CMI_2026 consultation, which was a proposal at the November 14, 2026 research cutoff.
Cohort persistence is particularly important for closed retirement portfolios. If a common improvement continues through many years, the annuity value may change in response to a cumulative adjustment to the entire survival path. A temporary one-year increase in death probabilities cannot be interpreted as cancelling a previously established long-run improvement trend without a demographic model justifying that conclusion. Capital governance should track both observed mortality deviations and how those observations change future projected improvement.
Backtesting, sensitivity tests, and transparent reporting
Backtesting a mortality model requires a clear division between calibration and evaluation periods. A forecast should be judged against subsequent observed data that were not used to fit its parameters. Evaluation needs to distinguish death counts, death rates, survival probabilities, and financial liability errors. An accurate prediction of population death rates can still translate poorly to a pension liability if age and benefit weights are misspecified.
A useful validation design includes: age-by-age residual plots; out-of-sample point forecast errors; predictive interval coverage; parameter sensitivity; calibration-window stability; demographic subpopulation comparisons; model ranking under a prespecified scoring rule; checks for mortality probabilities within valid ranges; and reconciliation of age, sex, birth cohort, and calendar-year definitions. The financial layer then compares the present value and risk distribution generated by each model for the same contractual cash flows and discount assumptions.
Data provenance should be preserved from original death counts and exposures through adjustment, estimation, projection, benefit valuation, and reported capital metrics. The Human Mortality Database overview demonstrates an institutional approach that distinguishes source data and harmonized calculated mortality rates. The practical lesson for a pension portfolio is to document the starting data, any transformations applied, and the population to which the mortality basis is intended to refer.
A source-to-claim audit also needs to test time specificity. An OECD publication from 2014 can document the state of risk-management discussion and the methods then analyzed; it is not an authority for a purported current country rate in 2026. A 2025 OECD report can support its stated life-expectancy estimates and their definition, but cannot by itself provide an empirical calibration of the hypothetical USD annuity portfolio. A 2026 U.S. Trustees projection is a public-program scenario and should be attributed as such rather than described as observed life expectancy for a private insurer.
Integrated findings and limitations
The mathematical and institutional evidence supports a coherent set of conclusions. Longevity risk has an idiosyncratic component, which can be reduced by pooling, and a systematic component arising from common survival factors, which can remain material in the average liability of a very large portfolio. Liability sensitivity depends on the entire trajectory of payment eligibility, the benefit schedule, and the discount curve. Capital exposure depends on what assets and risk-transfer claims offset those liabilities. Hedging effectiveness depends on the covered population and value weights, the index or contract definition, counterparty performance, and the risk metric being optimized.
These conclusions apply across several retirement-financing institutions, but their numerical implementation must remain contract- and jurisdiction-specific. Private pension schemes, annuity insurers, public pension programs, and individually held retirement assets differ in their obligations and financial flexibility. The methods share probability and present-value foundations, while the balance-sheet results depend on the relevant legal and financial arrangements.
The limitations of the present research are explicit. There is no new proprietary mortality forecasting model, actuarial pricing system, or financial doctrine. The two-state and mortality-multiplier experiments are invented diagnostic inputs with exact arithmetic, not fitted to a country or insurer. Full private pension liability datasets were not accessed. Several academic sources were examined through publisher bibliographic records, abstracts, or repositories where full texts were unavailable. Country comparisons use the definitions stated by their originating institutions. External refereeing has not been commissioned because it requires a separate decision or mandatory formal requirement. These boundaries shape the claims and are part of the completed internal scholarly record.
The methodological contribution of this monograph is an integrated, reproducible interpretation connecting conditional survival dependence, actuarial liability valuation, funding sensitivity, transfer design, and evidence-specific source attribution. It identifies which claims follow from mathematical identities, which are consequences of declared hypotheses, which belong to a particular regulatory system, and which reflect published empirical research. That distinction is sufficient for a defensible scientific synthesis without claiming the discovery of a new theory.
Conclusion
Longevity risk in retirement portfolios is a consequence of the uncertain duration of life-contingent financial obligations and the dependence among the lives to which those obligations attach. Its systematic component survives the pooling of idiosyncratic lifetime uncertainty because a shared mortality trajectory changes the expected benefits owed to many participants simultaneously. The conditional-variance decomposition formalizes this result: the part of portfolio uncertainty associated with individual outcomes decreases relative to portfolio size under suitable assumptions, while the common-state component may persist. The identity is familiar in probability theory; its scientific significance here lies in the interpretation of actual retirement liabilities and their exposure to joint demographic change.
The valuation of those liabilities requires a cohort-based survival path, the precise form of the benefit promise, appropriate discount factors, and the population characteristics of the pensioners or annuitants. A one-year mortality table and a single average life-expectancy figure cannot describe the complete cash-flow distribution of a multi-decade portfolio. The Lee–Carter and Cairns–Blake–Dowd traditions, multi-population basis-risk research, and the current use of improvement frameworks all demonstrate the importance of age, period, and cohort structure. Selection, data quality, calibration windows, and model-form uncertainty determine how confidently a forecast may be carried into distant payment years.
The hypothetical calculation establishes the magnitude and persistence of the mechanism under declared assumptions. For 10,000 age-65 annuitants with fixed annual payments of USD 20,000 for at most 45 years and a flat 3% discount rate, the stipulated baseline yields an expected liability of about USD 2.8768 billion. Multiplying all annual death probabilities by 0.8 raises the liability to about USD 3.0868 billion, an increase of 7.3009%. In an artificial two-state common mortality distribution, the total portfolio's liability standard deviation continues to reflect shared mortality uncertainty as the number of beneficiaries increases. These figures are reproducible examples; they are not country-specific estimates or insurer reserve recommendations.
Capital consequences depend on the assets and contractual protection available against the promises. A more favorable survival path for annuitants can generate an immediate liability revaluation, a subsequent stream of longer payments, or both. Funding ratios, economic surplus, regulatory own funds, and liquidity needs may respond differently. Solvency II's specified longevity stress is one jurisdictional instrument for insurer capital measurement and cannot be inferred from a simple actuarial present-value experiment. Pension and public retirement rules must likewise be interpreted in their legal and institutional setting.
Risk transfer can change where systematic longevity exposure resides. A customized hedge may provide close cash-flow matching for a selected pensioner population. An index instrument can reduce exposure more broadly, while introducing basis risk arising from differences between the reference population and the actual benefit-weighted portfolio. The empirical and methodological literature demonstrates that the effectiveness of such a hedge depends on its cohort, model, recalibration, and risk metric. Counterparty performance, contract settlement, collateral, and retained investment and inflation risk remain part of the net assessment.
The final analytical conclusion is that longevity is a long-duration balance-sheet relation among survival, contingent promises, financial resources, and risk bearers. Sound retirement portfolio management evaluates the entire relation. Its scientific foundation consists of transparent mortality and benefit data, coherent stochastic and actuarial assumptions, appropriately chosen risk measures, reproducible sensitivity calculations, enforceable transfer arrangements, and governance capable of revising assumptions when the evidence changes.
This Research Dossier contributes an integrated scientific synthesis, detailed derivations, an exact numerical scenario calculation, a cross-jurisdictional comparison, and source-specific editorial verification. It introduces no new authorial theory, proprietary model, or doctrinal system. Its original contribution is the organization and independent interpretation of established results and their consequences for retirement portfolio obligations.
References
Brouhns, N., Denuit, M., & Vermunt, J. K. (2002). A Poisson log-bilinear regression approach to the construction of projected lifetables. Insurance: Mathematics and Economics, 31(3), 373–393. https://research.tilburguniversity.edu/en/publications/a-poisson-log-bilinear-regression-approach-to-the-construction-of/.
Cairns, A. J. G., Blake, D., & Dowd, K. (2006). A two-factor model for stochastic mortality with parameter uncertainty: Theory and calibration. The Journal of Risk and Insurance, 73(4), 687–718. https://doi.org/10.1111/j.1539-6975.2006.00195.x.
Cairns, A. J. G., Blake, D., Dowd, K., Coughlan, G. D., Epstein, D., & Khalaf-Allah, M. (2011). Mortality density forecasts: An analysis of six stochastic mortality models. Insurance: Mathematics and Economics, 48(3), 355–367. Research bibliographic and abstract record.
Cairns, A. J. G., Dowd, K., Blake, D., & Coughlan, G. D. (2014). Longevity hedge effectiveness: A decomposition. Quantitative Finance, 14(2), 217–235. https://doi.org/10.1080/14697688.2012.748986.
Continuous Mortality Investigation. (2026a). CMI Working Paper 211: CMI Mortality Projections Model, CMI_2025. Institute and Faculty of Actuaries. CMI Working Paper 211.
Continuous Mortality Investigation. (2026b). Mortality improvements and CMI_2025: Frequently asked questions. Institute and Faculty of Actuaries. CMI_2025 frequently asked questions.
Continuous Mortality Investigation. (2026c). CMI news: Mortality projections, technical corrections, and consultation announcements. Institute and Faculty of Actuaries. CMI news and corrections. Accessed November 14, 2026.
Coughlan, G. D. (2012). Longevity risk transfer: Indices and capital market solutions. In P. Barrieu & L. Albertini (Eds.), The Handbook of Insurance-Linked Securities (Chapter 21). Wiley. https://doi.org/10.1002/9781119206545.ch21.
Coughlan, G. D., Khalaf-Allah, M., Ye, Y., Kumar, S., Cairns, A. J. G., Blake, D., & Dowd, K. (2011). Longevity hedging 101. North American Actuarial Journal, 15(2), 150–176. Research and DOI record.
European Commission. (2015). Commission Delegated Regulation (EU) 2015/35 supplementing Directive 2009/138/EC, Article 138: Longevity risk sub-module (as presented in the EIOPA Solvency II Single Rulebook). Article 138 official rulebook.
Human Mortality Database. (n.d.). Overview: Data sources, scope, and principles. University of California, Berkeley, and Max Planck Institute for Demographic Research. Human Mortality Database overview. Accessed November 14, 2026.
Lee, R. D., & Carter, L. R. (1992). Modeling and forecasting U.S. mortality. Journal of the American Statistical Association, 87(419), 659–671. https://doi.org/10.1080/01621459.1992.10475265.
Li, J., & Hardy, M. (2011). Measuring basis risk in longevity hedges. North American Actuarial Journal, 15(2), 177–200. Research and DOI record.
Organisation for Economic Co-operation and Development. (2014). Mortality Assumptions and Longevity Risk: Implications for Pension Funds and Annuity Providers. OECD Publishing, Paris. https://doi.org/10.1787/9789264222748-en.
Organisation for Economic Co-operation and Development. (2025a). Life expectancy. In Pensions at a Glance 2025: OECD and G20 Indicators. OECD Publishing, Paris. OECD 2025 life-expectancy chapter.
Organisation for Economic Co-operation and Development. (2025b). Methodology and assumptions. In Pensions at a Glance 2025: OECD and G20 Indicators. OECD Publishing, Paris. OECD 2025 methodology chapter.
Pitacco, E., Denuit, M., Haberman, S., & Olivieri, A. (2009). Modelling Longevity Dynamics for Pensions and Annuity Business. Oxford University Press. https://doi.org/10.1093/oso/9780199547272.001.0001.
The Pensions Regulator. (2025). New models and options in defined benefit pension schemes. Guidance for trustees and employers, published June 3, 2025. The Pensions Regulator's guidance.
U.S. Social Security Administration, Office of the Chief Actuary. (2026). Cohort life expectancy: Table V.A5 in the 2026 OASDI Trustees Report. 2026 Trustees cohort mortality table.