Retirees who reached retirement with at least $500,000 in non-housing assets still held 88% of that money twenty years later. That's the median household, not an outlier. In England, real financial wealth falls by roughly 31% between ages 70 and 90, while expected remaining life falls by 75%. Money outlives people, and it isn't close.
The finding replicates across countries, datasets and decades. Most retirees don't decumulate. They live on their income, leave the capital alone, and die with most of it intact. Whether that's a mistake or a purchase of insurance is the interesting part.
The American evidence on spend-down
Sudipto Banerjee tracked Health and Retirement Study households from 1992 to 2014 for the Employee Benefit Research Institute. He sorted retirees by the non-housing assets they held immediately before retiring, then followed the median through the next two decades.
Retirees who started with at least $500,000 had a median of $857,450 in their first two years of retirement. After nineteen or twenty years, the median stood at $756,300. That's a fall of 11.8% across two decades.
The middle group spent a little more. Households starting with $200,000 to $500,000 saw their median fall 27.2% by the seventeenth or eighteenth year. The poorest group, holding under $200,000, spent about a quarter of theirs.
Then the number that reframes the whole exercise. About one-third of the entire sample finished the period with more assets than they started with. They didn't decumulate at all. They carried on saving.
Banerjee found the mechanism sitting in the spending data. The median ratio of household spending to household income, across retirees of every age, hovered around one and drifted up only slowly. Retirees were living on the income and leaving the pot alone. Households with pension income showed it most sharply: their median non-housing assets fell just 4% over the first eighteen years of retirement, against 34% for households without a pension.
The distribution matters as much as the median. Among the poorest group, whose median starting balance was only $31,740, 35.1% had less than a fifth of their opening assets left by year seventeen or eighteen. Exactly the same share, 35.1%, had more than 100% of their opening assets left. Both tails are real, and averaging them hides both.
What people actually hold when they die
James Poterba, Steven Venti and David Wise attacked the question from the other end. Instead of asking whether households had saved enough at retirement, they measured what those households held in the last survey wave before death.
Their 2012 NBER paper follows the AHEAD cohort, aged 70 and over when first surveyed in 1993. Median wealth in the wave before death ran between $142,000 and $188,000 for people already living alone in 1993. For people who were part of a couple in 1993 and whose spouse had since died, it ran between $206,000 and $286,000. For those still in a couple when they died, the median sat between $585,000 and $685,000.
Income barely moved either. Total income in the last year observed was about 4% higher, on average, than the same people's income back in 1993. That's at ages where the textbook life-cycle model predicts a long, deliberate wind-down.
An earlier paper by the same three authors found something odder still. Among couples where both members stayed in good health, average financial assets typically rose across the 1992 to 2008 window. Assets only declined where health had deteriorated. Poor health, not age, did the spending.
Britain runs the same experiment
Rowena Crawford's briefing note for the Institute for Fiscal Studies uses the English Longitudinal Study of Ageing, following the same individuals from 2002-03 to 2014-15. Median non-pension wealth among people aged 55 to 64 was £250,000, with 60% of it in owner-occupied housing and 22% in financial assets.
Financial wealth is the liquid part, so it's the fair test of intent. Among people born between 1930 and 1934, who aged from 69 to 81 across the panel, median real net financial wealth fell 14%. Among those born 1925 to 1929, it fell 13%. Among those born 1920 to 1924, who aged from 79 to 91, it fell 1%.
Stack the cohorts and the implied path is a drawdown of at most 17% between ages 70 and 80, and 31% between 70 and 90. Set that against Office for National Statistics projections, which have expected remaining life falling 75% across the same twenty years. Wealth is being consumed at well under half the rate the clock runs.
Crawford's conclusion is blunt. The majority of financial wealth held by currently retired people in England is set to be bequeathed rather than spent. Housing makes it starker: over 40% of owner-occupiers at age 50 are expected to move house before they die, and the equity those moves release mostly isn't consumed either. Our look at housing as a portfolio asset against equity returns covers how large that frozen balance can get.
Wealthier households do draw down faster, but not much faster. Crawford estimates each additional £10,000 of financial wealth is associated with a 1 percentage point greater decline over six years. Even at the top of the distribution, drawdown runs slower than the decline in remaining life expectancy.
Spending falls even when nothing forces it to
David Blanchett measured what retirees spend rather than what they hold. Writing in the Journal of Financial Planning in 2014, he found real household spending falling roughly 1% a year through retirement. The fitted decline from age 60 to age 90 was −0.96% a year, with a t statistic of −4.31.
The path isn't a straight line. Blanchett described a "retirement spending smile": real spending falls fastest through the middle years, less steeply early on when people still travel, and less steeply again at the end when medical costs arrive. Total real spending still falls at every stage he measured.
That finding sits awkwardly beside the standard planning model, which holds real spending flat for thirty years and asks whether the portfolio survives. Real retirees spend less each year in real terms. The model is solving a harder problem than the one they're living.
Mechanism one: the tail risks are genuinely large
Longevity risk is not a rounding error. Among 65-year-olds in the United States, about one in five will die before 75 and another one in five will live to at least 90. Davidoff, Brown and Diamond put similar numbers on a younger cohort: roughly 16% of 62-year-old men die before 70, and another 16% reach 90 or beyond.
Health costs stack on top. Mariacristina De Nardi, Eric French and John Bailey Jones estimated out-of-pocket medical spending from the AHEAD data and found it rising very steeply with age. Average expenditure for a woman in poor health climbs from about $1,200 at age 70 to about $19,000 at age 100. Their structural model reproduces slow decumulation without any bequest motive at all — medical expense risk alone explains why the richest elderly singles run their assets down so slowly.
Long-term care is the fat tail. Lockwood cites evidence that over half of 50-year-olds will spend time in a nursing home before they die, at an average cost then estimated at about $84,000 a year in the United States. Neither the US nor the UK covers that through universal social insurance. A retiree self-insuring a risk of that size has to hold a large balance, and most of the time will never claim against it.
Mechanism two: bequests behave like a luxury good
Lee Lockwood's 2018 paper in the American Economic Review is the sharpest work on why the insurance doesn't get bought instead. About 5% of American retirees buy a life annuity. About 10% buy long-term care insurance. Standard life-cycle models, which assume people care only about their own consumption, cannot fit that.
Models where bequests are a luxury good fit it well. The logic runs backwards from intuition. A bequest motive is usually assumed to raise demand for products that protect an estate. Lockwood shows the opposite for late-life risks: if leaving money is itself valuable, the opportunity cost of holding precautionary capital collapses. Unspent wealth isn't wasted, so there's little reason to pay an insurer to guarantee it away.
That's why the two puzzles are one puzzle. Low annuity take-up, low long-term care cover and slow decumulation all fall out of the same preference. Annuity loads make it worse — Lockwood notes US long-term care insurance carries loads averaging 18% of premiums — but loads alone were never enough to explain the gap.
Mechanism three: the safety margin is built into the advice
The withdrawal-rate literature has a design feature that reads as a warning. William Bengen's 1994 paper set the 4% starting rate by asking what had never failed across the historical record. His own summary of the result: a 4% first-year withdrawal, then inflation-adjusted, had never exhausted a portfolio before 33 years, and in most cases produced portfolio lives of 50 years or longer.
Read that as a distribution rather than a rule. A rate calibrated to the worst case will, in the median case, leave a very large balance after thirty years. The rule was engineered so failure is rare. Dying rich is the arithmetic consequence, not an accident. Our review of safe withdrawal rates tested across 17 countries shows how much that safe rate moves once you leave the US historical record.
Fixed-rate rules also ignore feedback. Spending that responds to portfolio value supports a higher opening rate, which is the whole point of the guardrail approach and the price of a 5.2% start. And the reason a fixed rate needs so much padding in the first place is sequence risk, where identical average returns produce opposite outcomes. None of that removes the tail. It does explain why a retiree following the conventional rule ends up with a balance most heirs would call substantial.
The counter-argument: insurance you bought and never claimed
The strongest objection to the underspending story is that it isn't underspending. The IFS makes the case against its own headline in the same document. Holding wealth, Crawford argues, provides insurance against unexpected costs that is valuable in expectation even if the risk never materialises and the money is never used. Judging the decision by the outcome is hindsight.
That objection has real force. Nobody knows their date of death in advance. A retiree who spends down to a thin balance at 75 and then needs four years of residential care has made an unrecoverable error. A retiree who dies at 92 with £300,000 unspent has made a recoverable one — the money still exists, it just went to heirs.
Structural work supports the same reading. De Nardi, French and Jones fit observed saving behaviour using medical expense risk and no bequest motive whatsoever. If a model with pure self-interest and realistic health costs reproduces the data, calling the behaviour irrational is a stretch.
The critics of the underspending frame also point at heterogeneity, and they're right to. Poterba, Venti and Wise report that 12.1% of the original single-person households in their sample were below the income poverty threshold with no financial assets at all. About 23% were below twice the poverty line with no financial assets. Among that single-household pathway, 52% had annuity income under $20,000 and financial assets under $10,000. Averages describing "retirees" cover people whose problem is the exact opposite.
Where the objection weakens is on the frequency of the shock. Crawford's team examined the ELSA end-of-life interviews. Only a small minority paid out-of-pocket medical costs. At most 7% spent more than six months in a nursing or residential home, and not all of those were paying for care. Average funeral costs were small next to average wealth. The catastrophic end-of-life expense is real, but it's rarer than the balances held against it imply.
Where the UK drawdown data cut the other way
Britain's defined contribution market doesn't look cautious at first glance. Under the FCA's retirement income data for April 2024 to March 2025, 961,575 pension plans were accessed for the first time, and £70,876m was withdrawn — up 35.9% on the prior year. Drawdown policies sold reached 349,992 against 88,430 annuities, so drawdown outsold annuities by nearly four to one.
Across all pot sizes, 259,546 of 578,491 plans taking regular withdrawals were drawing 8% a year or more. That's 44.9%, and it looks like the reverse of underspending.
Split by pot size and the picture inverts. Among pots of £250,000 and above, 34,357 plans withdrew under 2% and 41,413 withdrew 2% to 3.99%. Together that's 52.3% of large pots drawing less than 4% a year. Only 20,490, or 14.2%, drew 8% or more. Among pots under £10,000, by contrast, 83.5% drew 8% or more — which is what emptying a trivial balance looks like, not a spending plan.
Two behaviours are being averaged together. Small pots get cashed out. Large pots get preserved. The high headline rate is a composition effect.
What would change the conclusion
Three things would move it. First, the datasets are old. The HRS work runs to 2014, the ELSA panel to 2014-15, and Blanchett's spending series to 2009. Cohorts retiring now hold defined contribution pots rather than defined benefit income, and Banerjee's own result — 4% drawdown for pensioners against 34% for everyone else — predicts faster spending as guaranteed income disappears. The FCA large-pot figures don't show that yet, but the sample is young.
Second, care costs could rise enough to justify the balances. If the share of retirees facing a long, self-funded care episode moves well above the 7% ELSA found, precautionary holdings that look excessive today become adequate. Policy decides this more than markets do.
Third, and most likely to matter, the bequest evidence could turn out to be measurement rather than motive. Lockwood's finding rests on fitting a model to observed choices. It doesn't prove people intend to leave money. Banerjee flagged the same gap: what share of actual bequests are planned rather than accidental remains an open question. If most of it is accidental, the underspending reading survives. If it's deliberate, most of what looks like a mistake is just a preference nobody asked about.
What the evidence supports today is narrow and reasonably firm. Retirees spend far less than the withdrawal-rate literature says they safely could, they spend less in real terms each year, and most of the financial wealth held at retirement is still there at death. Whether that's caution, generosity or arithmetic is not settled by the balance sheets alone.