3. The withdrawal rate, and why no fixed figure is safe

Risk 8: Horizon, contributions and withdrawals

If the order of returns decides the outcome, the obvious question is how much can be taken out without running out of money. The most quoted answer in finance is a number, 4%, and this module explains where it came from, what its small print contains and why calling it safe is an abuse of language.

Before going on, it is worth spelling it out: this is education, not financial advice. There is no figure recommended for anyone here, only an analysis of why the figures in circulation mean less than they appear to.

Where the number came from

In 1994, an American financial adviser called William Bengen published a piece of work that did something nobody had done seriously: instead of using average returns, he took the real history of the US market from 1926 onwards and tested, year by year, what would have happened to a retiree who withdrew a percentage of the initial capital and uprated it with inflation each year. He looked for the percentage that would have survived thirty years even starting on the worst possible date, and he came out with something just above 4%, with a portfolio holding between a half and three quarters in equities.

The worst possible date turned out to be the mid-sixties, the case from the previous module: a flat stock market for sixteen years plus inflation that ate the rest. The 4% is neither an average nor a prudent figure picked by eye. It is the result for the worst imaginable client of a single market over a single century.

Four years later, three professors at Trinity University broadened the approach by calculating success rates: what percentage of all historical periods would have survived with each combination of withdrawal rate, asset mix and duration. That is where the tables with figures like 95% success came from, and they are still in circulation. The Trinity study popularized the rule, though the original work was Bengen's.

The four things the number takes for granted

This is where it pays to be unpleasant to the figure:

• One specific period and one specific country. The United States in the twentieth century was the winning market, and using it as your sample is an elegant form of survivorship bias. The long-run data of Dimson, Marsh and Staunton, which reconstructs more than twenty markets from 1900 onwards, shows considerably lower real returns in most of them. When Wade Pfau repeated Bengen's exercise across those countries, in most of them the rate that would have survived was below 4%, and in several of them far below. A Japanese, French or Italian retiree who had applied the rule on the wrong dates would have been ruined.
• Zero costs. The original work does not deduct fees. The arithmetic is brutal: if your portfolio carries 1.2% a year between fund, adviser and custody, the total extraction is not 4% but 5.2%, 30% more drain. That extra point does not come out of the return of a good year; it comes out of the same account your food comes out of.
• Zero taxes. A Spanish retiree who sells fund units pays tax on the gain, and that payment appears in no American table. If you need 20,000 euros net, you have to sell more than 20,000.
• Exactly thirty years. If retirement lasts forty, the sustainable figure falls appreciably, because the portfolio has to withstand ten more withdrawal cycles. And life expectancy at 65 in Spain is among the highest in the world.

Why a success rate is not a probability

There is a deeper criticism and it is one of logic, not of data. A 95% success rate is computed over overlapping historical sequences, not over independent futures. The thirty years starting in 1965 and those starting in 1966 share twenty-nine: they are not two observations, they are almost the same one. The number of genuinely distinct histories a century of data contains is small, and on that basis a figure with two decimal places gets built.

Moreover, success defined this way is binary and does not distinguish magnitudes. Finishing with one euro counts as a success, and running out of money in year twenty-eight counts as a failure, just as running out in year ten does. A retiree does not live an average: he lives one path.

And above all there is the least realistic behavioral assumption of the lot: that the retiree would carry on withdrawing his inflation-indexed amount, unmoved, while watching his wealth evaporate. Nobody does that. Flexible rules (cutting spending after a bad year, skipping the inflation uprating, setting floors and ceilings) change the result completely, and that flexibility is worth more than any decimal place in the starting rate.

A real case: retiring in Japan in 1989

The example that disarms any excessive confidence in the American series is Japan. The Nikkei 225 index closed December 1989 at around 38,900 points, in what was then the largest stock market in the world by capitalization, with an economy every textbook presented as the model to copy.

It did not return to that level until February 2024, thirty-four years later. A Japanese retiree from 1989 who had applied the 4% rule to domestic equities would have exhausted his portfolio long before the index recovered its starting level, and he would have done it selling units for decades in a falling market, exactly the mechanism of the previous module taken to the extreme.

The lesson is not that Japan was an odd case. It is that the 4% describes what happened to a fortunate country and not a property of markets, and that whoever repeats it as though it were a physical constant is confusing a sample with a law.