Monte Carlo invents a thousand futures. A historical backtest does something blunter: it takes the retirement plan you have and hands it to a retiree who stopped work in 1900, then to one who stopped in 1901, and so on through every year the data covers, and reports how many of them died solvent.
Here is that done to one plan. Dev, from the sequence risk post, retires at 55 with £1,000,000 and draws £40,000 a year in today's money, a 4% rate, for forty years. The historical backtest starts him in each of 124 years and runs every one to 95 on the site's series of real US equity returns. He survives 116 times. He fails eight, and the eight are the whole story. The pound signs are a convention, by the way: the tool never converts anything, so the same figures in dollars or euros give the same results, and the returns behind all of it are American whichever symbol you use.
The eight retirements that failed
They come in four pairs, and only one of the pairs is the crash everyone pictures.
The 1929 retiree is the one who fails first, and by a distance. His first four years lose 10%, 22%, 38% and 6% in real terms, a 59.1% fall in total, and by the end of 1932 the £1,000,000 is £290,030. The 53% year that follows lifts him to £403,746 and no further; he is drawing £40,000 from a pot that now needs to earn nearly 10% just to stand still. The money is gone in 1948, at 75, nine years earlier than any other cohort. Start a year later, in 1930, and the same story ends at 84.
The 1966 retiree never sees a year like 1931. His worst is 1974, at minus 35%, and his first three years actually leave him near where he started, £982,158 at the end of 1968. What he gets instead is sixteen years, 1966 to 1981, in which the market loses 14.4% in real terms while he takes £40,000 out of it every one of them. By the end of 1975 he holds £443,565, by 1985 £236,523, by 1990 £133,418, and the pot is empty at 85. His neighbours who retired in 1968 and 1969 go the same way, at 89 and 85. Nothing dramatic happens to any of them. The plan simply bleeds.
The 2000 retiree is the modern one. Three losing years take 41.8% off, leaving £485,008 at the end of 2002, and 2008 lands on a pot that has never recovered: £321,663 at the end of 2009, £244,872 at the end of 2019, £153,699 at the end of 2022 with seventeen years still to fund. The data ends in 2023, so the tool wraps the sequence round to 1900 to finish his retirement, and he fails at 84 on borrowed returns. It makes no difference. A pot of £153,699 paying out £40,000 a year is finished whatever comes next.
Then the pair nobody names. The retirees of 1906 and 1907 both run out at 90, and the reason is the worst fifteen-year stretch in the whole series, 1906 to 1920, which loses 26.8% in real terms across the panic of 1907, the First World War and the inflation that followed it. That is a deeper hole than 1966 to 1980. It goes unmentioned in most retirement writing because the Trinity study's data begins in 1926 and Bengen's in that same era, and this series, an approximation drawn from the long record Robert Shiller publishes, starts a generation earlier.
Which leaves a count worth holding onto. Eight failures, but only two of them are a crash. The other six are a decade or more of poor real returns with withdrawals running through it, and that is the shape a plan should be tested against, not the one it usually is.
What the other 116 look like
Passing is a low bar. The plan that failed eight times also produced the 1982 retiree, who rode an 18-year run in which the market gained 1,089.8% in real terms and reached 95 with £22,457,124. The retirees of 1921, 1922 and 1933 all finished above £19,000,000.
Across all 124 cohorts the median ending balance at 95 is £4,184,249, more than four times the pot Dev retired on. A hundred of them finish above the £1,000,000 they started with, 83 above £2,000,000 and 56 above £5,000,000. Meanwhile eighteen finish below £500,000, and the 1965 retiree, who passes, does so with £71,092 in hand, having spent most of forty years watching the pot shrink. A pass and a failure can be a single bad year apart.
The fan chart on the tool draws that spread rather than the average, and it is worth reading at 65 as well as 95. Ten years in, a tenth of Dev's cohorts are already under £468,979 and a tenth are above £2,981,114, from the same £1,000,000 and the same £40,000 a year. History does not deliver a retirement. It delivers one of these, and you do not get to pick.
2008 was survivable, 1929 was not
The single most useful cohort to look at is the one that is not on the failure list. The 2008 retiree loses 38% in his first year and ends it with £580,000. Nine years later he is back over £1,000,000, at £1,123,874, and he reaches 95 with £1,807,608, though the last 24 of his years are borrowed from the early 1900s, a point returned to below.
Set that beside the crash test, which drops the same plan by 35% on day one and then grows it at a flat 5% with no rebound, and reports the money gone at 90. The two tools disagree about the same event because one of them keeps the recovery and the other deliberately throws it away. What killed the 1929 and 1966 retirees was not the size of the fall but the length of time the market stayed down, and the backtest is the only tool on this site that can show the difference, because it is the only one that knows what happened next.
The rate is the dial, and it is not linear
Change nothing but the withdrawal rate, keeping the £1,000,000 pot and adjusting the spending to match, and the failures behave like this. At 3% and 3.25%, every one of the 124 cohorts survives. At 3.5% one fails, and it is 1929. At 3.75% two fail, 1929 and 2000. At 4% it is eight. At 4.25%, twelve; at 4.5%, twenty-three; at 5%, thirty-two, a plan that fails one retirement in four.
Read the middle of that ladder carefully, because it is where most plans sit. Going from 3.5% to 4% costs seven cohorts. Going from 4% to 4.5% costs fifteen. The steps get dearer the higher you climb, so the difference between £37,500 and £40,000 a year on this pot is not a rounding error. It is the whole 1960s, and 1906, 1907 and 1930 besides.
Horizon moves the same list. Retire at 65 with the identical figures and only five cohorts fail; at 50, ten do. Whichever age, the names barely change: 1929, 1930, 1966, 1969 and 2000 are on every list. The bad years are fixed. What your rate and your age decide is how many of them you can afford to meet.
Where the past runs out
How far to trust any of this depends on the data, which is thinner than 124 cohorts makes it sound.
The series is 124 annual real returns for US large-cap equities, and nothing else: no bonds, no cash, no UK data, no fees or tax. It averages 8.2% a year, compounds at 6.34%, and 39 of its 124 years were losses, the longest losing run being the four from 1929 to 1932. That 6.34% is also why the backtest is so much kinder to Dev than the Monte Carlo simulator, which scores him at 61%: the simulator uses the 5% he typed, the backtest uses the American century, and what an 85% success rate actually means sets out that gap in full.
The windows overlap. A 1966 retirement and a 1967 retirement share 39 of their 40 years, so 124 cohorts are nearer three independent lifetimes than 124 trials, and a 93.5% pass rate is not 116 separate pieces of evidence.
And 39 of the 124 cohorts, every start year after 1984, run out of real data before the retirement ends and borrow the early 1900s to finish. The 2000 cohort above dies on 1904's returns. In his case the outcome was settled long before the splice, but a plan that starts in 2010 is being scored partly on the First World War, and it is worth knowing which cohorts are which before reading the later ones as history.
Spending that moves
Every failure above assumes Dev takes £40,000 in real terms every year regardless. The standalone backtest runs fixed spending only; the strategy switch is on the FIRE calculator, where the backtest runs beside Monte Carlo. Pick the simplified guardrail rule there, which trims the draw by a tenth when the rate drifts 20% above target and raises it when it drifts 20% below, and all 124 cohorts survive; the tenth-percentile ending balance rises from £261,851 to £1,127,357. The percentage-of-pot rule does the same, at £1,040,247. The median falls under both, from £4,184,249 to £2,737,784 and £2,370,743, because both rules spend the good years rather than banking them.
That is the trade every flexible rule offers, and the backtest shows only half of it. It counts survivals, not the income the 1966 retiree would have been living on in 1980 after a decade of cuts, which is the number that decides whether the rule is bearable. The withdrawal strategies tool reports that side.
Reading your own backtest
Run your plan through the backtest, and go past the gauge to the toughest-starts line beneath it, which names the six cohorts that failed earliest. If they cluster in 1929 and 1930, your plan is exposed to a crash it could not recover from, and the crash test is the next thing to run. If they cluster in the 1960s, the exposure is a long grind of poor real returns, and the fixes are a lower rate or spending that flexes. Then find your own plan on the rate ladder above, and ask whether the cohorts between your rate and the next quarter-point down are ones you would be willing to be. The how-to guide covers the controls; this is what the output looks like when you read all of it.