Who lives longest in retirement?

Welcome to TRR #21 for 2026. We may be on the home stretch to year-end, but there’s still lots to unpack in the retirement ecosystem and this edition is no exception.

Longevity risk, the risk of outliving one’s actuarial life expectancy, has moved from the fringes of the retirement conversation prior to the Financial System Inquiry and Retirement Income Review, to now be very much at the centre of proceedings.

Despite the increased policy attention, take-up rates for lifetime income stream products have been in the doldrums since the removal of the Age Pension exemption for complying (Tier 3) income streams back in September 2004, with a partial return of the exemption sparking something of a revival over the past 18 months or so.

Yet despite the renewed appetite for the new breed of Innovative Retirement Income Stream (IRIS) products amongst advised and higher-wealth retirees, demand for lifetime income solutions remains subdued, relative to account-based pensions.

A recent research paper helps explain why this is the case. This special edition takes look at the paper’s findings.

Nobel Prize-winning economist William Sharpe is once reputed to have claimed that turning a lifetime of accumulated savings into a steady, reliable income stream in retirement is the “nastiest, hardest problem in finance”, and with good reason.

A first-order problem is that we simply can’t know, before the fact, how long any particular lifespan will be, and so consumption smoothing becomes problematic in the face of an uncertain retirement horizon at the individual or household level. 

Actuaries deal with this uncertainty through the study of population-wide mortality risk, as contained in the Australian Life Tables (ALT), the inverse of which is longevity risk, a risk identified by the retirement income covenant for specific mention.

The ALT, as maintained by the Australian Government Actuary (AGA), focuses primarily on population-wide Australian mortality (longevity) risk based on gender and age (year of birth).

Historically the ALT has had little to say about other sources of longevity variation across the population, but a 2021 AGA-Centre for Population paper, on life expectancy by relative socio-economic advantage, has inspired three researchers to take that paper’s methodology and dive deeper. Much deeper.

In a 2025 research paper titled ‘Towards Fairer Retirement Outcomes: Socio-Economic Mortality Differentials in Australia’, actuarial academics Fei Huang and Andrés Villegas (UNSW) and Francis Hui (ANU) researched mortality differentials across four socio-economic indicators (covariates):

The paper, first presented at the 2025 All Actuaries Summit, draws on ABS Person Level Integrated Data Asset (PLIDA), linking the 2016 Census with social security and demographic records. It covers every Australian aged between 60 and 100 from September 2016 to August 2017, capturing some 57,000 female deaths (corresponding to more than 2.4 million years of exposure) and around 58,000 male deaths (some 2.1 million years of exposure), with crude death rates closely matching ALT 2015–2017.

The authors suggest their approach as being the first comprehensive study of Australian retirement-age mortality across socio-economic subgroups using individual-level linked data for the entire population.

The Huang, Hui and Villegas paper is revelatory insofar as it provides evidence of a level of heterogeneity in Australian mortality risk that was hitherto missing from the longevity debate, bringing into sharper focus the role played by certain socio-economic factors.

Analysis of crude death rates shows each one of the four covariates moving in the expected direction, with gaps between rates that are widest at 60 and narrow sharply with age to be effectively indistinguishable from the ALT rates by age 100.

Having validated the mortality rate differences, the authors then created a regression model to allow for further application of their findings, including examining the extent of variation in period life expectancy among older Australians.

The model also allowed them to assess the financial implications of these differences by calculating annuity rates* and the corresponding lifetime income for each of four distinct socio-economic profiles at age 60, from low advantage to high advantage.  A summary of the results is below:

The modelling results suggest that period life expectancy for an Australian male aged 60 in 2016-17 differed by up to 11.5 years by socio-economic circumstance.  For females, that difference was up to 9.1 years.

Lifetime annuities priced to take these longevity differences into account might have delivered an additional $2,736 per annum to the most disadvantaged males relative to the most advantaged ones, while for females that sum might have been an additional $1,509 per annum.

The authors have now created an Australian Longevity Calculator to allow interested parties to engage with their findings in a visually interactive way. The site also allows the more technically inclined to download open-source data for the modelled period-based mortality probability.

This paper by Fei, Hui and Villegas goes some considerable way toward explaining the “annuity puzzle” in Australia, which at present sees the market penetration of lifetime annuity and related income stream products sit somewhere between 2% and 4% of the overall retirement income market (which for APRA-regulated funds is rapidly approaching $600 billion).

The "annuity puzzle" has attracted many explanations including bequest motives, a preference for liquidity, reluctance to hand over a lump sum and the safety net of the Age Pension. This paper puts hard numbers on one of the less comfortable explanations: for many Australian retirees, a uniformly priced annuity may simply represent poor value.

Consider the paper's most disadvantaged male profile. Priced on his own mortality, $100,000 at 65 would buy $8,521 a year. Priced on reference annuitant rates, he would receive $6,381, or 25% less. Even on population-wide ALT rates, this individual gives up about 18%. Declining that product is not a behavioural bias; it is a reasonably rational response to being asked to subsidise someone else’s longevity risk.

The mirror image is the most advantaged male, for whom ALT-based pricing would overpay by around 20%. High-advantage annuitant longevity sits well above the population average, and is precisely why lifetime annuities may appear to not represent ‘good value’ to more middling cohorts. Adverse selection and the annuity puzzle are, in part, the same problem viewed from different angles.

Heterogeneity in life expectancy matters because the recently finalised Best Practice Principles, which we examined in our February special edition, put lifetime income offerings squarely in the frame for funds. Principle 5 treats a lifetime income product as a core component of a quality solution, and Principle 9 expects at least one trustee-designed solution to include a lifetime income component.

As we noted then, it is now difficult for a trustee to credibly claim best practice "without at least evaluating longevity risk solutions". This paper sharpens what that evaluation should involve. A sizable lifetime income allocation in a RIC cohort of top income decile professionals in a fund full of knowledge workers may be as valid as no such allocation for members of a fund concentrated in lower-income occupations and more disadvantaged postcodes. It’s horses for courses and Know Your Member.

It also gives new substance to Principle 4's call for at least three cohorts reflecting a fund's actual membership. Most cohorting today leans on age, gender and balance. It is increasingly apparent that home ownership, marital status, personal income and postcode (as a proxy for relative social disadvantage) also carry material longevity signals, and should not be summarily dismissed as irrelevant.

In our view, longevity-aware RIC cohorting, incorporating relationship status, homeowning status and socio-economic circumstance, is the requisite progression for funds who wish to go beyond age, gender and balance as segmentation variables. And indeed we know of, and applaud, funds that are already on this journey.

The Retirement Reporting Framework is where this becomes a public issue. APRA's revised draft reporting standards, released in September, will collect lifetime income take-up, drawdown behaviour and balance utilisation from late 2027, segmented by age, sex and balance. No socio-economic attribute is proposed as it currently stands.

Without that context, comparisons across funds will need to be done judiciously. Low lifetime income take-up at a fund serving a disadvantaged membership may reflect sound guidance rather than a material shortcoming. Average balance at death will partly reflect how early members die, not just how ‘well’ they drew down. Trustees should be ready to explain their numbers through a longevity lens before others interpret those numbers for them. Context matters when it comes to engaging members with lifetime retirement income propositions.

The findings don’t necessarily suggest that product providers should move to socio-economic pricing with immediate effect. The UK's experience with postcode-based and enhanced annuities shows it can work, but it raises questions of fairness, privacy and complexity that the Australian industry has barely begun to address. The paper's own limitations also counsel care. Its data are from 2016–17, an update using the 2021 Census is underway, and health information, the most obvious missing variable, is not yet included. The authors readily acknowledge that they are closer to the start of a journey in understanding Australian longevity risk heterogeneity than the end.

The practical takeaway is nonetheless clear. The ALT is not a neutral assumption for any given fund's membership, annuitant tables are even less so. If the annuity puzzle has partly been a pricing puzzle all along, solving it starts with knowing who your members really are by cohort, and how long they are likely to live, given their socio-economic circumstance.

The key lies in understanding the distribution of possible longevity outcomes, not merely a measure of central tendency.

* Annuity rates were calculated based on a $100,000 investment by an individual aged 65 in 2016, with improvement factors from ALT 2015-2017, using the 125-year projection scenario. Further, the interest rate was assumed to be 3% p.a., and that the product was devoid of additional features such as indexation, guarantee periods or investment-linked returns.