Key takeaways
- Across 69 rolling 30-year retirements from 1928 to 2025, a 4% fixed percentage withdrawal from a 60/40 portfolio never ran the money out, while the inflation-adjusted 4% rule failed in 4 of them.
- The price was income. In the median window the percentage rule's worst year paid 76% of year one's income in real terms, and in the window starting 1966 it paid 41%.
- Real income dropped below 90% of its first-year level at some point in 47 of the 69 windows, and below 70% in 29 of them.
- The percentage rule also spent more in total: median lifetime real income of 1.51 times the starting balance, against 1.20 times for the inflation-adjusted rule.
- Morningstar put the 2025 base-case starting safe withdrawal rate for level real spending at 3.9%, rising to 5.7% once spending is allowed to move with the portfolio.
A fixed percentage withdrawal can't run out of money, and that isn't the same as safe
You've got a pot and you want an income from it. One rule says take 4% of what you started with and give yourself an inflation rise every year. The other says take 4% of whatever the pot is worth today, every year, and let the number move. Which one survives?
The second one, always. That isn't a close call and it isn't really a finding about markets. It's arithmetic: if you only ever remove a fraction of what's left, there's always something left. Morningstar's 2025 retirement income research puts it flatly, saying a 4% withdrawal of the portfolio balance per year "entirely solves the problem of running out of money, but it does so at the expense of the retiree's standard of living being buffeted by changes in the value of the portfolio."
So the real question isn't survival. It's what the buffeting costs. Run both rules through every 30-year window in the US return record since 1928 and the inflation-adjusted rule failed 4 times out of 69. The percentage rule failed none. But the percentage rule's income fell to a median worst year of 76% of what it paid in year one, and in the worst window it fell to 41%. You don't avoid the risk. You move it from your balance to your bank statement.
The two rules differ in one place: what the percentage is a percentage of
Both rules start identically. On a £500,000 pot at 4%, both pay £20,000 in year one. Then they diverge.
The inflation-adjusted rule, which Vanguard's 2021 UK research note calls the "pound-plus-inflation" rule, fixes the amount in real terms. Vanguard describes it as selecting "an initial amount that would be withdrawn annually from the portfolio" which then "would increase each year by inflation". Your £20,000 becomes £20,400 after 2% inflation, regardless of what the portfolio did. This is the rule William Bengen described in 1994, quoted in a later academic paper as: "Assuming a minimum requirement of 30 years of portfolio longevity, a first-year withdrawal of 4 percent, followed by inflation-adjusted withdrawals in subsequent years, should be safe."
The percentage of portfolio withdrawal fixes the rate instead. Vanguard's description: the investor "chooses a specific fixed percentage of the portfolio to withdraw each year", and "while this strategy ensures that the portfolio is never depleted, it can lead to big swings in annual withdrawals". Their example is blunt. Starting from a £1,000,000 portfolio and a £50,000 withdrawal at 5%, a 20% market fall in year one leaves £760,000, and 5% of that is £38,000. That's a 24% pay cut in a year the market fell 20%, because the withdrawal came out before the fall.
Notice what the second rule does not do. It carries no inflation adjustment at all. Morningstar is explicit that the method "does not incorporate any type of inflation adjustment", and that any rise in the withdrawal "is completely contingent on the portfolio value increasing". In 2022 UK CPI ran at 9.1%, the highest annual rate in the Office for National Statistics series that starts in 1989. A pound-plus-inflation retiree was owed a 9.1% rise that year. A percentage retiree was owed whatever the portfolio had.
What the inflation-adjusted rule did across 69 historical retirements
Here's the survival record. The portfolio is 60% S&P 500 and 40% ten-year US Treasuries, rebalanced annually, using annual total returns and US CPI from 1928 to 2025 as compiled in Aswath Damodaran's NYU Stern dataset. Withdrawals come out at the start of each year. A window counts as a failure if the portfolio couldn't fund the scheduled withdrawal before the horizon ended.
| Starting rate | 20 years | 25 years | 30 years | 35 years | 40 years |
|---|---|---|---|---|---|
| 3.0% | 100% | 100% | 100% | 100% | 100% |
| 3.5% | 100% | 100% | 100% | 100% | 98% |
| 4.0% | 100% | 100% | 94% | 91% | 85% |
| 4.5% | 100% | 92% | 84% | 73% | 68% |
| 5.0% | 94% | 81% | 72% | 62% | 54% |
| 6.0% | 77% | 62% | 52% | 47% | 37% |
Windows available: 79 at 20 years, 74 at 25, 69 at 30, 64 at 35, 59 at 40. The 94% at 4% over 30 years is 65 windows out of 69, and the 4 that failed all started in the same narrow band: 1965, 1966, 1968 and 1969. They ran dry in years 28, 26, 28 and 28. That clustering is the whole story of the rule's weakness, and it has a name you'll meet in any decumulation discussion: sequence risk, the plain fact that two retirements with the same average return end in different places depending on when the bad years land.
The grid above also shows the rule has nothing to do with the number 4. At 3.5% nothing failed over 35 years. At 5% over 30 years, better than a quarter of windows failed. Our longer piece on withdrawal rate success rates works through the horizon and allocation grid in more detail.
What the percentage of portfolio rule charged for its perfect survival record
Run the same 69 windows with the percentage rule and the survival column is 100% at every rate, so it tells you nothing. The informative column is income. For each window we track what the withdrawal bought in year-one money, then record the worst year.
| Rate of balance | Windows survived | Median worst year, % of year one income | Worst window | Median lifetime real income | Median real end balance |
|---|---|---|---|---|---|
| 3.0% | 69 of 69 | 85% | 48% | 1.33x | 1.95x |
| 3.5% | 69 of 69 | 81% | 44% | 1.43x | 1.67x |
| 4.0% | 69 of 69 | 76% | 41% | 1.51x | 1.43x |
| 4.5% | 69 of 69 | 71% | 38% | 1.58x | 1.22x |
| 5.0% | 69 of 69 | 64% | 35% | 1.64x | 1.05x |
| 6.0% | 69 of 69 | 55% | 29% | 1.71x | 0.76x |
Lifetime real income and end balance are multiples of the starting pot. The chart above plots the third column: as the rate rises, the deepest income dip gets deeper, from 85% of year one at a 3% rate to 55% at 6%. That curve is the trade the rule makes.
Put the distribution next to it and the retirement income volatility stops being abstract. At a 4% rate, real income dipped below 90% of its first-year level in 47 of 69 windows, below 80% in 38, below 70% in 29, below 60% in 21 and below 50% in 13. A pay cut of a tenth or more wasn't the unlucky case. It happened in roughly two windows in three.
The 1966 retiree is where the two rules part company
Take the single hardest start date in the record and watch both rules work.
A 1966 retiree on the inflation-adjusted rule at 4% got exactly the income they signed up for, every year, until year 26. Then the money stopped. Total real income drawn: 1.02 times the starting pot, against the 1.20 the plan promised over 30 years, and nothing left for the last 5 years.
The same retiree on the percentage rule never ran out. But their income ground down for 17 years, bottoming at 41% of the first year's spending power in year 17, and it was still only 79% by year 30. They drew 0.82 times the pot in total and finished holding 0.97 times the pot in real terms.
Both outcomes are bad, and they're bad in ways that aren't interchangeable. One retiree had full income for 25 years and then nothing. The other kept a cheque arriving for all 30 years, having cut their standard of living by more than half in the middle of it. Which of those is the worse retirement isn't an arithmetic question, and no backtest answers it. What the arithmetic does say is that the 1966 percentage retiree ended with 1.79 times their pot across spending plus what was left, and the inflation-adjusted retiree ended with 1.02.
The rule that never fails also spent more, which is the part that surprises people
The percentage rule is usually filed as the cautious option, the one you take if you're frightened. The lifetime income column says otherwise. Over the 69 windows at a 4% rate, the percentage rule's median lifetime real income was 1.51 times the starting pot. The inflation-adjusted rule's was 1.20.
There's no mystery in the gap. A surviving inflation-adjusted plan is capped at 30 years of 4% by construction, which is 1.20 and not a penny more, however well markets do. Upside goes to the estate instead: that rule's median real end balance was 1.39 times the starting pot. The percentage rule has no cap. When the portfolio compounds, the withdrawal compounds with it, so good decades show up as income rather than inheritance.
So the two rules aren't ranked. They answer different questions. One buys a predictable income and pays for it with a tail risk of running dry plus a large expected bequest. The other buys certainty of never running dry and pays for it in income volatility, and gets handed more total spending for the trouble.
The strongest case against level real spending is waste, not failure
The sharpest academic attack on the inflation-adjusted rule doesn't argue it fails too often. It argues it wastes money even when it works. In a 2008 paper, "The 4% Rule, At What Price?", Jason Scott, William Sharpe and John Watson note that the rule finances "a constant, non-volatile spending plan using a risky, volatile investment strategy", so "retirees accumulate unspent surpluses when markets outperform and face spending shortfalls when markets underperform". Their estimate is that "a typical rule allocates 10%-20% of a retiree's initial wealth to surpluses and an additional 2%-4% to overpayments".
The 1.39 times median end balance in our own run is that criticism showing up in the data. A rule that ends the median retirement with more real money than it started with, while capping lifetime spending at 1.20, isn't failing. It's over-saving on the retiree's behalf, which is a cost even though it never appears as one.
The counter-case runs the other way too, and Vanguard made it by omission. Its UK research dropped the percentage of portfolio approach from the main simulation entirely, "because of the extreme swings in annual spending that the strategy can sometimes produce during market shocks". A rule with a 100% survival rate was still judged unusable for the comparison. That's a fair verdict on a 41% worst year.
Both camps have converged on hybrids rather than either pure rule. Morningstar's 2025 report found the base-case starting safe withdrawal rate for level real spending was 3.9%, and that flexible methods lifted it as far as 5.7%. Its own constant-percentage test wasn't even the pure rule: it applied "a floor so that the spending in a given year doesn't drop below 90% of the initial withdrawal amount". The withdrawal guardrails approach is the same instinct written as a formal ruleset.
UK drawdown investors are mostly withdrawing at rates neither rule supports
There's a gap between this literature and what UK pension savers do. The FCA publishes the rate at which drawdown plans are actually being emptied, measured as the percentage of the pot withdrawn during the year, which is the percentage rule's own yardstick.
In 2024/25, of 578,491 plans taking regular withdrawals, 259,507 were drawing at 8% of the pot or more. That's 44.9% of them. Another 68,132 sat in the 6% to 7.99% band. Plans drawing under 4% a year, the zone where both rules in this piece are comfortable, numbered 156,463, or 27.0% of the total.
Read the survival grid at 6% and you can see what that implies for the inflation-adjusted rule: 52% of 30-year windows, and 37% of 40-year ones. The percentage rule at 8% doesn't run out, because it can't. Put that rate through the same 69 windows and the median retirement's final year pays 41% of what year one paid in real terms, with the last year below the first in all 69 of them. Neither rule makes an 8% draw sustainable. It's worth saying that many of those plans aren't trying to be: some belong to people with other income, some are deliberately short-dated, and the FCA figures don't tell you which.
A Monte Carlo built on the same returns is harsher than the history
Rolling windows have a weakness. There are only 69 of them at 30 years and they overlap heavily, so the 1960s cohorts appear again and again and the record is thinner than 98 years of data suggests. A Monte Carlo built by drawing calendar years at random from the same 1928 to 2025 series, 10,000 trials, breaks that overlap.
It's less kind. The inflation-adjusted 4% rule survived 90.8% of trials rather than 94% of windows. The percentage rule's median worst year came in at 79% of year one income, close to the historical 76%, but the fifth percentile fell to 37%. Drawing years independently destroys the mean reversion and the inflation persistence embedded in the real sequence, which is exactly why it produces worse tails. It isn't a better answer than the history. It's a second answer, and the honest reading is that the fixed percentage withdrawal's income dip sits somewhere around a quarter in the middle of the distribution and around two-thirds in the tail.
What this evidence can't tell you
Start with the returns. This is a US series: the S&P 500 and ten-year Treasuries, in dollars, with US CPI. A UK investor holds neither, pays UK inflation and faces currency exposure on the equity sleeve. Our piece on the safe withdrawal rate outside the US shows how much that substitution matters, and it's a lot. The 1928 to 2025 window is also one country's history. Nothing in the method says the next century of it, or any other market, produces the same grid.
The run charges no fees, no taxes and no trading costs, and all three come straight out of the return the portfolio earns. Every survival figure in the grid is therefore better than the same investor would have seen in practice. Nor does it model a real spending pattern: it assumes you need exactly the same real amount at 95 as at 65, which Employee Benefit Research Institute work cited in the same Morningstar report contradicts, finding inflation-adjusted household spending "has historically fallen by 19% from age 65 to 75, 34% from age 65 to 85, and 52% from age 65 to 95".
Both rules here are also pure, and almost nobody runs a pure rule. A real retiree cuts back in a bad year, has a state pension underneath the portfolio, and reacts to a 41% income fall long before year 17. Morningstar's own test of the constant percentage method put a 90% floor under spending for that reason. The pure version is the right thing to measure, because it isolates the mechanism, and it's the wrong thing to mistake for a forecast of anyone's retirement.
And a backtest is not a forecast. A set of 69 overlapping windows from one country's history is a sample, not a distribution, which is the reason the Monte Carlo cross-check is in this piece at all.
What would change the conclusion
If the retiree has guaranteed income underneath the portfolio, the income volatility that makes the percentage rule hard to live with mostly stops mattering. A 41% fall in a portfolio drawing that covers a quarter of spending is a different event from a 41% fall in all of it. Morningstar makes this point about its own volatility numbers, noting that for nearly all retirees portfolio withdrawals are only part of the household's cash flow.
If real returns come in materially below the 1928 to 2025 record, the inflation-adjusted rule's failure count rises and the percentage rule's income dip deepens, but not symmetrically. The percentage rule degrades smoothly, because it never promised a number. The inflation-adjusted rule degrades by cliff, because it did.
If the horizon isn't 30 years, the grid moves more than the rate does. At 4% the inflation-adjusted rule survived every 25-year window and 85% of 40-year ones. Someone retiring at 55 is reading a different row from someone retiring at 70, and the row matters more than the choice of rule.
The number worth watching isn't the withdrawal rate. It's the gap between what the portfolio is paying this year and what it paid in year one, in real terms, because that single figure is what the percentage rule hides and the inflation-adjusted rule can't show you until it's too late. LedgerTouch tracks withdrawals against a portfolio's real value; a spreadsheet with a CPI column does the same job.