Aging Societies

Established Population aging is the most reliably forecast social change in existence, and almost everything said about its consequences is contested. The demography is near-deterministic: the people who will be old in 2060 are alive now, and fertility has already fallen below replacement in more than half the world’s countries. The United Nations’ 2024 projections put the global population peak in the mid-2080s at around 10.3 billion, with the over-65 population passing 2 billion within that window.

Frontier What is contested is everything that follows from it. Whether a rising dependency ratio implies falling living standards, whether immigration substitutes for births, whether automation substitutes for immigration, whether pensions face a solvency problem or a distributional one, whether older electorates starve the young: each has real evidence on both sides, and this brief takes the position that most are not resolved by it.

Established The one thing the evidence does settle is that the headline ratio is the wrong instrument. The old-age dependency ratio counts everyone over a fixed age as dependent and everyone under it as a supporter. It is a demographic construct, not an economic measurement, and the two alternatives that are economic measurements — prospective ratios keyed to remaining life expectancy, and accounts of actual age-specific income and consumption — both show far slower aging. A great deal of alarm rests on an indicator nobody would choose if they were designing one.

Frontier The joint question this brief owns is the one the related briefs each touch and none holds. Long-Term Institutions holds how badly bodies built to act beyond an electoral cycle have fared; Future Labour Markets, that automation is cohort-neutral and incumbent-destructive; Future Taxation Models, that a redesigned tax base rarely survives politics; Human Flourishing, that the wellbeing unit built to arbitrate such trade-offs has not been shown to change a decision. The joint question is the contract itself: who works, who pays, who cares, and whether any lever closes the arithmetic.

Established No single lever closes the arithmetic, all of them together plausibly do, and the binding constraint is distributional rather than demographic. Raising retirement ages is fought hard not because voters misunderstand arithmetic but because healthy life expectancy is distributed far more unequally than life expectancy, so a uniform increase takes a much larger share of remaining healthy years from poorer workers. That is a real objection, and no country has designed around it.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The demographic facts are not in dispute and should be stated precisely. Global total fertility is around 2.2 births per woman and falling; more than half of all countries and areas are below the replacement level of about 2.1. Japan has been above a 28 per cent share aged 65 and over for several years and recorded fewer than 700,000 Japanese births in 2024. South Korea’s fertility fell to 0.72 in 2023, the lowest ever recorded for a national population, then rose in 2024 and 2025. Frontier Whether that reversal is postponed births arriving, a response to very large pronatalist transfers, or noise, nobody knows. It is the most-watched number in the field.

Established The headline dependency ratio overstates aging, and two better measures exist. The prospective approach fixes the old-age threshold not at a birthday but at a remaining life expectancy of about fifteen years, on the argument that what makes a person old is proximity to death rather than distance from birth; on that construction aging in most rich countries proceeds at roughly half the headline rate. National transfer accounts measure what people of each age actually earn and consume rather than assuming it, and find the economic support ratio deteriorating far more slowly. Established The reason is that participation has already absorbed much of the adjustment: employment rates at 55 to 64 have risen substantially across the OECD over three decades, in several countries without anybody legislating it, and projections that hold participation constant miss the largest observed response to aging so far.

Established On pensions, the foundational result is sixty years old and still decides the argument. A pay-as-you-go system returns more than a funded one whenever population growth plus real wage growth exceeds the real interest rate; when it does not, funding wins. Neither is inherently sounder. Aging changes the sign of that comparison, and a country choosing between them is making an empirical bet on growth and interest rates over sixty years, not a moral choice about thrift.

Established The transition between them is the part nobody prices. Moving from pay-as-you-go to funding requires one cohort to finance the pensions of the retired while also saving for its own — paying twice. Chile’s 1981 privatisation is the canonical case: replacement rates came in far below the projections used to sell it, a solidarity pillar was added in 2008, pandemic withdrawals removed a large fraction of balances, and a 2025 reform raised employer contributions and reinstated a collective component. Frontier Reading that as proof funding fails is as unjustified as reading it as proof Chile merely executed badly. The transition cost was real, was borne by the contributing generation, and was not in the original arithmetic.

Established The technically successful answers are automatic adjustment mechanisms, and they work by removing the decision from politics. Sweden’s notional defined-contribution system computes pensions from lifetime contributions and cohort life expectancy, with a balancing mechanism that cuts indexation when assets fall short; it has activated and reduced pensions in payment, the political act no legislature performs voluntarily. Germany has a sustainability factor, Japan a macroeconomic slide, and Denmark and the Netherlands index the pension age to life expectancy, Danish legislation carrying it to 70 by 2040. These are the only designs with demonstrated capacity to absorb a demographic surprise without a crisis.

Frontier Raising the pension age raises employment, and it also raises hardship, and both findings are solid. The best-identified evidence comes from the United Kingdom’s increase in the female state pension age: the employment rate of affected women rose by several percentage points, a large effect by the standards of labour policy, while income poverty among the same women also rose, because the women least able to keep working were the ones who lost the pension. Both numbers belong in any honest account, and which dominates depends on who you are.

Established That distributional fact has a measured magnitude, and it is large. In the United States the gap in life expectancy at age 40 between the top and bottom one per cent of income is roughly fifteen years for men and ten for women, and over the period measured the gains went to the top while the bottom quartile saw none. Healthy life expectancy is distributed more unequally still: in England the gap between most and least deprived areas runs to around nineteen years. A pension age set by average life expectancy allocates retirement years regressively by construction.

Established Health spending under aging is mostly a long-term-care story, not an acute-care story. The finding that acute expenditure is driven by proximity to death rather than age — the red-herring result — has held up across replications: control for time to death and the age gradient in hospital costs largely disappears. Long-term care does not disappear; its costs rise steeply with age independent of mortality timing. Projections that scale total health spending by the over-65 share misallocate the problem.

Established Care labour is the binding physical constraint and it is overwhelmingly unpaid and female. Global unpaid care work runs to roughly sixteen billion hours a day, about three-quarters of it performed by women, with an imputed value near a tenth of world output. In paid long-term care, OECD work estimates millions more workers would be needed by 2040 merely to hold the carer-to-older-person ratio constant. No country has a plan that produces them domestically.

Frontier Pension assets are large and their scale is often mistaken for a solution. Global pension assets run to tens of trillions of dollars, one of the largest pools of long-horizon capital in existence. Established But funded assets are a claim on future output, not a store of it: a large cohort selling to a small cohort faces a price, and the goods consumed by retirees in 2060 must be produced in 2060. Funding changes who holds the claim and how risk is shared. It does not manufacture the output.

Frontier The first open question is whether below-replacement fertility is a problem at all, and there is a serious case that moderate sub-replacement is optimal. Work using national transfer accounts finds that once capital deepening is accounted for — fewer children means more capital per worker — the fertility rate maximising consumption per person in most rich countries lies below replacement, often well below. The same work finds that the optimum is higher, near replacement, for systems that rely heavily on pay-as-you-go transfers. So the answer depends on the institution, not on the biology: sub-replacement fertility is a problem for a specific financing design rather than for a society.

Frontier The second is whether aging depresses growth, and the best cross-country evidence says it has not. Work relating demographic change to output per head across countries since 1990 finds no negative association between aging and growth, and finds that more rapidly aging countries adopt robots faster. The mechanism proposed is straightforward: scarce middle-aged labour raises the return to automating exactly the tasks that scarce labour performs. Frontier The finding is genuinely contested on identification grounds, and it does not establish that automation will arrive in the sector where the labour shortage actually bites.

Frontier The third is whether older electorates starve the young, and the empirical record is mixed enough to refuse an answer. American state-level work found a rising elderly share associated with lower per-child education spending, with a larger effect where old and young differ by race. Swiss cantonal work using comparable methods found no such effect. Both are competently done, and the plausible reconciliation — that it depends on whether the old expect their own descendants to benefit — is a hypothesis, not a finding.

Established The binding bottleneck is care labour, and it resists every substitute proposed for it. Personal care is the archetypal cost-disease sector: output is the time itself, productivity growth is near zero, and wages must still track the rest of the economy or workers leave. Aging raises demand for exactly the service where technology has least purchase.

Frontier The care-robot record is the strongest available evidence against the automation answer, and it is not what advocates expect. Japan has subsidised care robots in nursing homes for over a decade. The best study of adoption found homes introducing robots employed more care workers afterwards, with gains in flexible and part-time staff and lower turnover among regular workers. That is complementarity and better retention. It is a good outcome; it is not labour substitution, and a worker shortage is not solved by a technology that raises headcount.

Established The second bottleneck is that the fiscal adjustment required is small annually and enormous cumulatively. Official long-run projections in the United Kingdom and the United States both show unchanged-policy debt rising to multiples of output over fifty years on health, long-term care and pensions. The annual gap is a few percentage points of GDP. Closing it is not technically hard; it is a fifty-year sequence of decisions no legislature can bind.

Established The third is housing, where the stock is mismatched to the population in both directions. Older households under-occupy family housing they have no financial reason to leave, while the cohort that would use it cannot afford to form households. Japan’s roughly nine million vacant homes are the far end of the same phenomenon: a stock built for a population that no longer exists, in places people no longer live.

Established Everything here depends on measures that are economic rather than demographic, and those measures exist but are not official. Prospective dependency ratios and national transfer accounts are mature research programmes with published country series. Neither is produced by a statistical office as a headline indicator, so neither enters the projections that drive policy: a mandate problem, not a research problem.

Established It depends on healthy life expectancy series disaggregated by socioeconomic position. Pension-age policy cannot be made equitable without knowing how many healthy years each group has left at each candidate age. Several countries publish the aggregate; far fewer publish it by income decile at the age that matters, and none indexes policy to it.

Frontier It depends on the discount rate, which is where intergenerational equity becomes arithmetic. The rate at which future lives are weighted against present ones determines almost every long-horizon conclusion, and the review behind current British guidance treats pure time preference as a value judgement to be declared rather than a parameter to be estimated. Long-Term Institutions holds the record of what happened when that arithmetic met a ministry.

Established The decisive test is a natural experiment already running in Canada, and it runs in both directions. Between 2021 and 2024 Canada raised immigration to the fastest population growth since the 1950s, driven mostly by non-permanent residents, who reached several per cent of the population. Then it reversed: permanent-resident targets cut for 2025 to 2027, a cap on the temporary-resident share, and population growth at approximately zero within a year. Few levers are pushed this hard in one direction and then the other, in a country with quarterly data of this quality.

Frontier What it would settle is the empirical core of the immigration argument. On the expansion side: wages at the bottom of the distribution, rents, per-capita output, age structure. On the contraction side: the sectors that had staffed themselves from the inflow, long-term care above all, and the fiscal balance. Cross-country regressions cannot separate the policy from the conditions that produced it. This one nearly can, and the data already sits in administrative records.

Frontier The second experiment is the one nobody has run: a pension-age increase evaluated by socioeconomic group on health outcomes, not just on employment. Every major pension-age reform is evaluated for its employment effect. Almost none is evaluated for its effect on disability incidence, mortality or healthy years lost, broken down by the occupational and income groups that differ by a decade or more in healthy life expectancy. The data exist in linked administrative records in the Nordic countries and the United Kingdom. Until that evaluation exists, the central distributional objection to the most effective available lever cannot be answered with evidence.

Established A third experiment has already returned a clear result and is widely ignored: automatic balancing works. Sweden’s mechanism has cut pensions in payment when the system’s balance required it, and the system survived politically. That is a demonstration, not a projection, and it is the strongest existing evidence that the pension problem is a governance problem with a known solution rather than an arithmetic one without.

Established The engineering that matters for aging is unglamorous and mostly building-related. Step-free access, grab rails, stair design, lighting and door widths determine whether a person with reduced mobility can remain at home, and retrofitting is cheap relative to residential care. Accessible-design standards in new housing are close to free at construction and expensive at every point after.

Frontier Assistive technology has a deployment problem, not an invention problem. Fall detection, medication management, remote monitoring and lifting aids all exist and work in trials. They fail in deployment for organisational reasons: nobody owns the budget for a device whose savings accrue elsewhere.

Established The other constraint is settlement geometry. Care at scale requires density, and older populations sit disproportionately in low-density places the young have left. Domiciliary care in a depopulating rural area is a travel-time problem before it is a labour problem.

Established This sits directly against Future Labour Markets, whose central finding constrains the automation answer here. That brief establishes that automation’s distributional shape is cohort-neutral and incumbent-destructive — the next cohort is absorbed, the worker holding the job is not. Applied to aging, that predicts automation will not rescue the incumbent care worker or the incumbent older manufacturing worker, which is precisely the population an aging society needs to keep employed.

Established It is adjacent to Long-Term Institutions, which holds the record of bodies built to represent the future. This brief does not re-litigate it; it takes as given that statutory future-generations offices have mostly been short-lived, and asks what follows for a contract that must hold eighty years.

Established It is adjacent to Future Taxation Models on financing — shifting from taxing labour to taxing consumption, property or wealth is the standard recommendation for an economy with fewer workers, and that brief’s durability finding is the reason for caution about it.

Frontier It is adjacent to Longevity Therapies and Future Housing Systems, which own the two variables that would most change the arithmetic — the morbidity gap and the housing stock — and to Human Flourishing, which owns the measurement problem that any attempt to value a year of healthy old age runs into.

Established The institutional requirement that dominates is a mechanism that adjusts without a vote. Every country that has handled a demographic surprise gracefully did so through a rule written in advance — life-expectancy indexation, a balancing mechanism, a sustainability factor — and every country that reserved the adjustment to a future parliament postponed it. That is the clearest institutional finding in the subject.

Frontier Sovereign funds are the other credible commitment device, and their record is better than the offices are. Norway’s fund holds a very large share of national wealth under a fiscal rule limiting annual withdrawals to an estimated real return set below the historical average, respected across governments of both stripes for two decades. What makes it work is the rule and the annual parliamentary accounting around it — and a resource windfall most countries do not have.

Established Public pension funding, where it exists, is a weaker commitment than it looks. American state and local plans hold trillions against liabilities, with an aggregate funded ratio stable but well short of full for more than a decade through a period of strong returns. A plan chronically underfunded during a bull market is one whose contribution rule, not its investment strategy, is the binding variable.

Frontier Automatic enrolment is the one behavioural intervention with a large, replicated effect, and it works for an uncomfortable reason. The Danish evidence is that the overwhelming majority of savers are passive: they do not respond to tax subsidies, which mostly shift where money is held, but they do respond to automatic contributions, which raise wealth. Policy that assumes an optimising saver gets the smaller instrument.

Frontier The intergenerational-equity argument has two strong forms and they are not reconcilable by evidence. The first: current cohorts have written claims on future output — unfunded pensions, health entitlements, public debt, deferred infrastructure, a changed climate — without the consent of those who will pay. The arithmetic is real. The second: generational accounting is not a measurement. It is exquisitely sensitive to the discount rate and assumed productivity growth, treats cohorts as homogeneous when within-cohort variation exceeds between-cohort variation, and ignores the inherited capital, institutions and knowledge the same older cohort transfers. That objection is also real.

Established What can be said with evidence is that the within-cohort variation is very large. A fifteen-year gap in life expectancy at 40 across the income distribution exceeds the gap between consecutive generations on almost any measure. Cohort-framed policy therefore mis-targets: a universal pension-age increase hits a poor sixty-year-old and a wealthy one identically while removing very different fractions of their remaining healthy lives.

Frontier On immigration the strongest version of each position deserves stating, because both are usually caricatured. For expansion: it raises the working-age share immediately and with certainty, which no other lever does; migrants in most rich countries have a net fiscal position close to neutral or mildly positive; and the sectors with the sharpest shortages, long-term care above all, are staffed by migrants in almost every receiving country. Against scale: the arithmetic to hold a support ratio constant through migration alone is unachievable, migrants themselves age, absorption costs in housing and services fall on the places with least capacity, and political sustainability is part of a policy’s evaluation rather than a separate matter. Frontier Both positions are consistent with the evidence; the disagreement is over the weight placed on adjustment costs and on consent, and this brief does not resolve it.

Frontier The care-labour question carries the sharpest ethical edge and the least attention. Recruiting care workers from lower-income countries pulls trained labour out of health systems that are also aging, and several source countries have restricted outflows for that reason. The receiving country gets carers; the sender loses them and gains remittances. Whether that is fair is contested; that it is the de facto plan in most rich countries is not.

Frontier The civilizational question is whether a shrinking population is a decline or a transition. The pessimistic case rests on scale effects: fewer people means fewer researchers, smaller markets, thinner networks and slower idea generation, and endogenous-growth theory supports the argument. The optimistic case rests on the capital-deepening arithmetic that makes sub-replacement fertility look optimal, plus the observation that the countries furthest along the transition have not seen falling output per head.

Established The transition is a one-way ratchet on the timescale of policy. Even an immediate return to replacement fertility would leave the working-age population falling for decades, because the cohort that would have the children is already small. Every response operates on the adjustment, not the underlying shape.

Frontier The under-examined civilizational risk is institutional, not economic. Institutions built during rapid growth encode growth in their financing: pension schemes, municipal infrastructure funded by expanding rate bases, universities sized for larger cohorts, defence establishments recruiting from a shrinking pool. The failure mode is not a crisis but a long sequence of small insolvencies in bodies that cannot shrink gracefully.

Speculative The most plausible long-run equilibrium is a society in which the third of life after fifty is institutionally reorganised rather than financed. That means phased exit rather than a cliff, portable entitlements, retraining that enrols older workers, and a pension age varying with contributory history rather than birthday. Every component exists somewhere; nowhere are they assembled.

These horizons track when the levers stop being contested and start being scheduled, not when the demography changes — the demography is already fixed.

• 10 yr: Frontier Life-expectancy indexation of pension ages becomes the default in Northern Europe and is fought to a standstill elsewhere; the Canadian immigration reversal produces its first clean evaluations; long-term-care workforce shortfalls become the binding constraint that pension arithmetic currently is; the South Korean fertility reversal is either confirmed as a trend or resolves into a tempo effect.
• 25 yr: Speculative Automatic balancing mechanisms spread because the alternative is repeated crisis legislation; healthy-life-expectancy-adjusted or contribution-history-based retirement ages appear in at least one national system; the housing stock begins to be reallocated by price and vacancy rather than by policy.
• 50 yr: Speculative Most of the world is past its population peak and the institutional question shifts from financing aging to managing shrinkage — municipal consolidation, university closure, infrastructure decommissioning — for which almost no country has a legal procedure.
• 100 / 250+ yr: Handwave Claims that radical life extension dissolves the problem, or that artificial wombs or automation make the working-age population irrelevant, work by assertion. Each requires a technology that does not exist, and each would create a distributional problem at least as hard as the one it replaces.

• Depends on This brief depends on results held by other briefs on this map. Long-Term Institutions supplies the record of bodies built to act beyond an electoral cycle, the form every intergenerational-equity proposal here would take. Future Labour Markets supplies the distributional shape of automation, which constrains the automation answer to the care shortage. Future Taxation Models bounds the shift from taxing labour to taxing consumption or wealth. Longevity Therapies holds the only variable that changes the shape of the problem rather than its financing.
• Requires (not on this map) Five constraints sit outside what research delivers. The first is a pension age indexed to healthy life expectancy by occupational group rather than to average life expectancy, the only design that answers the central equity objection to the most effective lever. The second is a domestically reproducible long-term-care workforce — pay, training and status enough to staff the sector without permanent recruitment from countries that are also aging. The third is completed-fertility evidence separating timing from quantum, without which every pronatalist evaluation measures postponement. The fourth is an automatic balancing mechanism that survives its first benefit cut somewhere other than Sweden, the test of whether that design travels. The fifth is a market or tax treatment that releases under-occupied housing, the stock mismatch no demographic policy touches.
• Enables Getting the measurement right enables the rest: prospective and economic dependency ratios published as official statistics would reframe every fiscal projection, and healthy life expectancy by income decile at the candidate pension age would turn the central distributional argument into a design constraint. A system that adjusts by rule enables long-horizon fiscal planning; a domestically reproducible care workforce enables aging in place at a cost that does not depend on a migration policy surviving an election.
• Adjacent Adjacent to Human Flourishing, which owns the valuation problem for a year of healthy old age; to Future Housing Systems, the stock mismatch; to Cellular Rejuvenation, the biology that would compress morbidity; to Wealth Distribution Systems, the within-cohort variation that dominates the between-cohort one; and to Future Civil Services, whose workforce is among the most rapidly aging in the economy.

Established “The old-age dependency ratio shows how many workers support each retiree.” It does not. It divides people over a fixed age by people between two fixed ages, counts students and the unemployed as supporters and working seventy-year-olds as dependants, and takes no account of what anybody earns or consumes. Both economic measures — prospective ratios and national transfer accounts — show aging proceeding substantially more slowly. The headline ratio is not a bad estimate of the economic burden. It is not an estimate of it.

Frontier “Immigration can solve population aging.” The arithmetic was done a quarter-century ago and has not been refuted: holding a support ratio constant through migration alone requires inflows on a scale no receiving country has sustained, because migrants age and the ratio must be held rather than moved once. That finding is routinely cited as proving immigration does not help, the opposite of what it says. Immigration reliably raises the working-age share and buys time; it cannot stabilise the ratio.

Frontier “Automation will fill the labour gap.” The macro evidence that aging induces automation is real, and the absence of a negative growth association is a serious finding. The problem is sectoral: automation displaces routine tasks in tradeable goods, while the shortage is in personal care, where the best study of care-robot adoption found employment of care workers rising. Both hold at once, and together they say the automation response arrives in the wrong sector.

Established “Funded pensions are safe and pay-as-you-go pensions are a Ponzi scheme.” The comparison has a known answer: pay-as-you-go outperforms when population plus wage growth exceeds the real interest rate, and funding outperforms when it does not. Beyond that both are claims on future output. A funded scheme does not store 2060 groceries; it stores a claim that must be sold to whoever works in 2060. The Ponzi framing is rhetoric attached to a genuine empirical question about growth and interest rates.

Frontier “Aging will bankrupt health systems.” Acute spending tracks proximity to death far more than age; control for time to death and most of the age gradient in hospital costs disappears. What aging raises steeply is long-term care — a different budget, workforce and politics. Projections that scale total health spending with the over-65 share get the size roughly right for the wrong reason and misdirect the response.

Frontier “The old vote themselves benefits at the expense of the young.” The best-known American evidence finds lower per-child education spending where the elderly share is higher; comparable Swiss work finds no such effect. Turnout gradients by age are real and large, but the inference from them to policy outcomes is exactly what the conflicting evidence fails to support. The claim is plausible, widely believed, and not established.

Speculative “Longevity medicine will dissolve the problem.” It would — if it compressed morbidity rather than extending frailty. The measured gap between healthy life expectancy and life expectancy has widened slowly over the period in which lifespans rose, the opposite of compression. A therapy that extends life without extending health makes the long-term-care problem worse.

Handwave “Pronatalist policy can restore replacement fertility.” No country has done it. Measured effects of transfers, leave and childcare are real, small, and frequently concentrated in the timing of births rather than the number, while the programmes cost around a percentage point of output. Hungary’s large programme and South Korea’s decades of spending are read in opposite directions by advocates and critics. Claims that a sufficiently generous package returns a rich country to 2.1 extrapolate far outside anything observed.