In 1994 a financial adviser named William Bengen asked a question nobody had answered properly: given real market history rather than average returns, what withdrawal rate would have survived the worst starting year on record?
Out of that question came the most quoted number in retirement planning. The 4% rule says you can withdraw 4% of your portfolio in the first year of retirement, increase that amount with inflation each year afterwards, and be reasonably confident the money outlasts you. It is also the most quoted number out of context. It is not a law, it was never intended as one, and the man behind it has spent decades saying so.
The worst year on record
Bengen tested every thirty-year retirement window in US data, including retirements that began just before the 1929 crash and in the stagflation of the 1970s. The worst case tolerated about 4.15%. He rounded down. Then in 1998 the Trinity Study examined the same territory with a portfolio survival framing and produced the success-rate tables that made the number famous.
Both were doing something genuinely valuable: replacing "assume 7% a year" with "here is what actually happened to people who retired at the worst possible moment." That is the strength of the rule, and it's worth holding on to. It is built on sequences that really occurred, not on a smooth average that never has.
The small print
But the 4% figure arrived wearing conditions, and every one of them changes the answer.
It assumes a thirty-year retirement, not forty-five; retire at 45 and the money must work for half as long again. It assumes a US stock and bond portfolio, roughly 50/50 to 75/25, held over the century in which the US produced the best equity returns of any major market, so survivorship bias is doing quiet work in the result. It assumes no fees, and a 1% platform-and-fund charge comes straight off the withdrawal rate; our investment fee calculator shows what that compounds to. It assumes you take the inflation-adjusted amount with perfect rigidity every single year, including the year the market falls 40%, which no real retiree does. It assumes no tax, so whatever your jurisdiction takes comes afterwards. And it defines success as "did not hit zero": a portfolio that ends year thirty with £4,000 counts as a success in the tables. Few people staring at that balance would call it one.
Strip the conditions away and you are left with a number that sounds like a guarantee. It never was one. It was the worst historical case for a specific portfolio over a specific length of time, nothing more.
Where the floor gives way
The big one, for anyone reading a FIRE site, is a retirement much longer than thirty years. Extending the horizon lowers the sustainable rate, because there are more years for a bad sequence to appear and less time to recover from it. Most analysis of forty-plus-year retirements lands between 3% and 3.5%, and at £30,000 of spending that is the difference between needing £750,000 and needing £1,000,000. A quarter of a million pounds hangs on a percentage point people quote as if it were fixed.
The second failure is a bad first decade. Two retirees with identical average returns can end up in completely different places depending on the order those returns arrive in. Losses early, while the pot is at its largest and you are selling into them, do damage that later gains cannot undo. This is sequence-of-returns risk, the single largest danger in drawdown, and the retirement crash test models exactly this.
Then there are the quieter two. Fees left out of the arithmetic: if the research assumed no costs and you pay 1%, you have already spent a quarter of your withdrawal rate before you start. And lower expected returns than the historical record: if the next thirty years are less generous than the American twentieth century (a mainstream view, given starting valuations), the rate that survived history is not the rate that survives the future.
The author disagrees with you
Whichever direction you lean, Bengen probably thinks you're too confident. He has repeatedly said 4% was a floor derived from the worst case, not a target, and that with wider diversification and some flexibility a higher rate is often defensible; he has published work suggesting nearer 4.7% under different portfolio assumptions.
Sit with that, because it cuts both ways. The number is not sacred in either direction. It is one output of one method on one dataset, and treating it as a constant of nature is the error, whether you round it up or down.
The rigidity in the original test is exactly what makes 4% conservative. Real people do not raise their spending by inflation in the middle of a crash; they postpone the kitchen, holiday closer to home, and wait. Strategies that build that flexibility in explicitly do better than a fixed rate. Taking a fixed percentage of the current balance means the pot can never run dry, though income falls in bad years, sometimes sharply. Guardrails start higher, then cut if the rate drifts above a ceiling or raise it if it falls below a floor, which buys most of the upside of flexibility with far less volatility in income. Our withdrawal strategy comparator runs these side by side, and the trade is always the same: steadier spending versus a smaller chance of running out. You cannot maximise both.
One pool of money that isn't
There's an omission in the rule that no withdrawal-rate debate fixes. It treats the portfolio as a single pool, and real retirements are not. If a large share sits in a pension you cannot reach for another seven years, a 4% draw on the total is not a draw you can actually make. The arithmetic is fine and the plan still fails, which is a different failure from running out of money and needs its own calculation.
The same applies to tax wrappers. Drawing 4% from an ISA and 4% from a pension produce different spendable incomes, so the rule tells you what to withdraw rather than what you get to spend.
An anchor, not a promise
So use 4% the way its own history suggests. Start with 25× spending to get an order of magnitude; what a FIRE number is covers that step if it's new. Lower the rate if your retirement is long, your fees are high, or your spending is inflexible. Stress the plan against real sequences rather than averages, which is what the historical backtest is for. And decide in advance what you will cut if the first five years go badly, because deciding during a crash is how people end up selling at the bottom.
The rule's genuine contribution was never the number. It was the method: stop assuming an average, and check what happened to the people who retired at the worst possible time.