Key takeaways
- In the Trinity study's 1926 to 1995 sample, a 4% inflation-adjusted withdrawal from a 50/50 portfolio survived 95% of 30-year periods. At 5% it was 76%, and at 6% just 51%.
- A 7% fixed withdrawal cleared 91% of 30-year periods on a 75/25 mix. The same 7% indexed to inflation cleared 45%.
- The 96% success rate behind the 4% rule is 53 successes out of 55 retirement start dates, and both failures began in 1965 and 1966.
- Morningstar's 2025 modelling puts the starting safe rate at 3.9% over 30 years and 3.3% over 40, both at a 90% success target, with data as of 30 September 2025.
- In the FCA's 2024/25 data, 259,507 of 578,491 UK plans taking regular withdrawals took 8% or more of the pot, or 44.9%.
The odds on 4%, 5% and 6%, before anything else
You want to know what happens if you take 5% instead of 4%. Here's the historical answer, on a 50/50 portfolio held for 30 years and with withdrawals raised each year for inflation. A 4% start survived 95% of the periods in the record. A 5% start survived 76%. A 6% start survived 51%, which is a coin toss.
Those three numbers come from one table, published by Philip Cooley, Carl Hubbard and Daniel Walz in the AAII Journal in February 1998, and known since as the Trinity study. It ran 200 portfolio success rates: 10 withdrawal rates, five allocations and four payout periods, on stock and bond returns from 1926 to 1995. The whole point of a success rate is that it answers a narrow question. Did the money outlast the period, yes or no.
What that structure hides is the shape of the failure and the size of the survival, and both matter more than the headline percentage. So the table comes first, and then the four things it doesn't tell you.
The table: inflation-adjusted withdrawal rate success rates, 1926 to 1995
Read a row across and you're changing the withdrawal rate. Read a column down and you're changing the horizon and the mix. Each cell is the share of overlapping historical periods in which the portfolio still held something at the end. Withdrawals start at the stated percentage of the opening balance and then move each year with the Consumer Price Index.
| Allocation | Payout period | 3% | 4% | 5% | 6% | 7% | 8% | 9% | 10% |
|---|---|---|---|---|---|---|---|---|---|
| 100% stocks | 15 years | 100 | 100 | 100 | 91 | 79 | 70 | 63 | 55 |
| 20 years | 100 | 100 | 88 | 75 | 63 | 53 | 43 | 33 | |
| 25 years | 100 | 100 | 87 | 70 | 59 | 46 | 35 | 30 | |
| 30 years | 100 | 95 | 85 | 68 | 59 | 41 | 34 | 34 | |
| 75% stocks / 25% bonds | 15 years | 100 | 100 | 100 | 95 | 82 | 68 | 64 | 46 |
| 20 years | 100 | 100 | 90 | 75 | 61 | 51 | 37 | 27 | |
| 25 years | 100 | 100 | 85 | 65 | 50 | 37 | 30 | 22 | |
| 30 years | 100 | 98 | 83 | 68 | 49 | 34 | 22 | 7 | |
| 50% stocks / 50% bonds | 15 years | 100 | 100 | 100 | 93 | 79 | 64 | 50 | 32 |
| 20 years | 100 | 100 | 90 | 75 | 55 | 33 | 22 | 10 | |
| 25 years | 100 | 100 | 80 | 57 | 37 | 20 | 7 | 0 | |
| 30 years | 100 | 95 | 76 | 51 | 17 | 5 | 0 | 0 | |
| 25% stocks / 75% bonds | 15 years | 100 | 100 | 100 | 89 | 70 | 50 | 32 | 18 |
| 20 years | 100 | 100 | 82 | 47 | 31 | 16 | 8 | 4 | |
| 25 years | 100 | 93 | 48 | 24 | 15 | 4 | 2 | 0 | |
| 30 years | 100 | 71 | 27 | 20 | 5 | 0 | 0 | 0 | |
| 100% bonds | 15 years | 100 | 100 | 100 | 71 | 39 | 21 | 18 | 16 |
| 20 years | 100 | 90 | 47 | 20 | 14 | 12 | 10 | 2 | |
| 25 years | 100 | 46 | 17 | 15 | 11 | 2 | 0 | 0 | |
| 30 years | 80 | 20 | 17 | 12 | 0 | 0 | 0 | 0 |
The sample is smaller than it looks. There were 56 overlapping 15-year periods between 1926 and 1995, 51 of 20 years, 46 of 25 and 41 of 30. A 30-year success rate of 95% is 39 periods out of 41, so a single extra bad start date moves it by more than two points.
Two patterns run through the grid. Success falls as the rate rises, steeply between 5% and 7%. And bonds help at low withdrawal rates and hurt at high ones: at 4% over 30 years the all-bond portfolio scored 20, while every mix holding at least half in shares scored 95 or better.
Indexing withdrawals to inflation halves the success rate at 7%
The same authors updated the work in the Journal of Financial Planning in April 2011, using monthly data from January 1926 to December 2009 and monthly rebalancing. That update lets you price the inflation adjustment directly, because it reports both.
A 7% withdrawal held flat in cash terms survived 91% of 30-year periods on a 75/25 portfolio. Raise that same 7% with inflation each year and it survived 45%. Nothing else changed. At 5%, the inflation-adjusted version still cleared 82% of 30-year periods on the same mix.
The other two levers move it less. Push the fixed withdrawal from 7% to 8% on that same 75/25 portfolio and success falls from 91% to 69%. Keep the 7% fixed but flip the mix to 25% shares and 75% bonds, and it falls to 38%. The inflation adjustment costs more than either.
The failures cluster, which is the part a single percentage can't show you. Of the eight 50/50 portfolios that failed at a 7% fixed withdrawal, four started in the late 1920s and early to mid-1930s, and the average one lasted 289 months, about 24 years. Of the 18 that failed at 5% with inflation adjustments, all but one spanned the high-inflation years of the late 1970s. Failure isn't randomly distributed across the sample. It's concentrated in two regimes.
The horizon changes the answer as much as the rate does
Move from 30 years to 20 on a 50/50 portfolio at 6%, and the success rate goes from 51 to 75. Move to 15 years and it's 93. That's the same portfolio and the same spending, judged over a different length of life.
Morningstar's The State of Retirement Income: 2025 reaches the same shape from the opposite direction, using forward-looking return assumptions rather than history. Its base case is a 3.9% starting rate over 30 years, defined as the highest rate at which 900 of 1,000 simulated paths still hold a positive balance. Stretch the horizon to 40 years and that falls to 3.3%.
Which row applies to you is a demographic question, not a market one. The ONS puts cohort life expectancy at 65 in the UK in 2023 at a further 19.8 years for men and 22.5 for women. Those are averages, and an average is not a planning horizon. A 30-year row isn't the typical case for a 65-year-old, and it isn't meant to be: it's the tail.
A 96% success rate is 53 retirement dates out of 55
The strongest objection to reading these tables as odds comes from Wade Pfau, who took the most-quoted cell apart in a 2011 working paper. The 4% inflation-adjusted rate on a 50/50 portfolio over 30 years shows a 96% success rate. That is 53 successes out of 55 possible retirement start dates, because a 30-year window needs a start date no later than 1980. The two failures were 1965 and 1966. Neighbouring years, one economic episode.
The 5% cell decomposes the same way. Of the 55 start dates in the 1926 to 2009 data, 17 could not have sustained 5%, which leaves 38 that could and a 69% success rate. These are counts of a few dozen overlapping histories, not draws from a distribution.
Pfau's second criticism is structural. In overlapping windows, the middle of the record gets counted far more often than the edges. Every year from 1955 to 1980 appears in 30 of the simulated retirements, while 2009 appears in one. So the sample is not a fair draw of market history, and it over-weights a stretch when the arithmetic mean real return on long-term corporate bonds was -1.4% a year, against 5.2% for the years outside it.
That bias has a direction. Historical simulation gives a bonds-only portfolio a 35% success rate, which reads as an argument for equities. Run Monte Carlo simulations on the same underlying data and success rates above 90% are reachable with only 20% in shares. Two methods, one dataset, opposite implications for allocation.
Success is binary, so it hides the difference between $9 and $4,128
A portfolio that ends with one dollar counts as a success, and the tables treat it identically to one that ends with 10 times what it started with. The 2011 update publishes the median ending values, and they're worth putting beside the success rates.
Take a 6% inflation-adjusted withdrawal over 30 years. The median 50/50 portfolio ended with $9 for every $1,000 it started with. The median all-equity portfolio ended with $4,128. Both count as survivors in the grid above.
William Bengen's original 1994 paper measured the other thing entirely. He plotted portfolio longevity in years, and found a 3% initial withdrawal never left a portfolio short of 50 years. A success rate compresses that into a yes. If what you care about is a bequest, or the size of the buffer at 80, the success rate is the wrong summary statistic and the ending values are the right one.
What UK drawdown pots actually take out
Set the tables beside behaviour. In the FCA's retirement income market data for April 2024 to March 2025, 578,491 UK pension plans made regular partial withdrawals, counting firms that ran at least 750 such plans. Of those, 259,507 took 8% or more of the pot during the year, which is 44.9%. In 2018/19 the same share was 39.6%.
Two things stop that being a direct comparison with the table. The FCA measures the percentage of the pot withdrawn in one year, not an initial rate held for 30. And the rate falls sharply with pot size: among plans of £250,000 and above, 14.2% withdrew at 8% or more. Small pots are often being deliberately run down, sometimes alongside a defined benefit pension or a State Pension that isn't in the number at all.
What survives the caveats is a distribution problem. On the 1926 to 1995 record, an 8% inflation-adjusted withdrawal from a 50/50 portfolio cleared 5% of 30-year periods and 64% of 15-year ones. If a pot is doing a job that lasts, the horizon row you land in decides whether that rate was ever in the range the history supports. The arithmetic behind the pot itself is in our piece on the 25x rule, and the mechanics of cutting spending when a bad sequence arrives are in withdrawal guardrails.
What would move these numbers
Four things, and none of them is a market forecast.
The sample end date. The 1998 table stops in 1995 and the update stops in 2009, so a 30-year window can't yet include anyone who retired after 1980. The 1995 to 2025 stretch enters the record only as those windows close, one start date a year.
Costs. Neither study charges a fee: the Trinity paper states plainly that it did not adjust for taxes or transaction costs, and Morningstar's 2025 work says the same about expenses and taxes. Every basis point of platform and fund cost comes out of the same return the withdrawal is taken from.
Return assumptions. Morningstar's 3.9% is a forward-looking figure. Feed long-term historical returns into the identical model instead and the 50/50 answer rises to 4.4%. That moves the pot a given spend implies from 25.6 times to 22.7 times.
Geography. These are US shares, US bonds and US inflation, and the US was a good draw: its average real stock return of 6.6% a year was beaten by only three of the 16 other developed markets Pfau compared. Across 1900 to 2008 he found a 4% rule that failed in 62.5% of historical 30-year cases in Italy and 42.5% in France, which is covered in our piece on the safe withdrawal rate across 17 countries. Cooley, Hubbard and Walz drew their own planning line at a 75% success rate, on the reasoning that it worked three times in four. Whether one country's twentieth century is the right sample to compute that from is the question the table can't answer about itself.