The 4 Percent Rule Is a Result From One Country and One Century

Name the study, then read what it actually tested

Bengen's 1994 paper set out a protocol, not a promise. The portfolio was US common stocks plus intermediate-term US government bonds, rebalanced once a year, with the equity share between 50 and 75 percent. The data began in 1926. Each hypothetical retiree started in a different calendar year and lived through whatever the market did next for 30 years. The withdrawal was set as a percentage of the pot on day one and then never recalculated: from year two onward the retiree took the same real sum, raised only by inflation. Bengen then reported the highest opening rate that would have survived every start year, including the worst — a floor, not an average.

The Trinity study four years later is a different object. Cooley, Hubbard and Walz used the S&P 500 and long-term high-grade corporate bonds, ran horizons of 15, 20, 25 and 30 years and withdrawal rates from 3 percent upward, tested both inflation-adjusted and flat withdrawals, and published grids of portfolio success rates by allocation and horizon. A success rate in that grid is the share of the historical windows in which the money outlasted the period. It is a frequency in one country's past, not a probability attached to your future, and the two studies do not even use the same bonds. When someone says the rule has a success rate, ask which table, which allocation and which horizon.

The inversion people actually want: the multiple is 1 ÷ w

If you withdraw a fraction w of the pot in year one and you need S to live on, then S = w × P, so P = S ÷ w and the multiple of spending you must accumulate is 1 ÷ w. At w = 4 percent that is 25×; at 3.5 percent, 28.6×; at 3 percent, 33.3×; at 2.5 percent, 40×. Someone spending $30,000 a year therefore needs $750,000 at 4 percent and $1,000,000 at 3 percent — a quarter of a million more, on the same lifestyle, for one point of caution. Notice the shape of the curve: the multiple rises far faster than the rate falls, because 1 ÷ w is a hyperbola, not a line.

There is a second reading of 1 ÷ w that makes the whole debate legible. If the portfolio earned exactly zero after inflation, the pot would last precisely 1 ÷ w years: 25 years at 4 percent, 33.3 at 3 percent, 40 at 2.5 percent. So the multiple is the number of years you are covered before any real growth at all, and the withdrawal rate is a bet on the market covering the gap between that figure and the length of your life. Framed that way, a 4 percent rule is a claim that real returns will stretch 25 years of spending into 30 or more — which is exactly the claim the backtest was testing, in one market, over one stretch of history.

The proof: same average, opposite outcomes

Take thirty annual real returns: five years of −10 percent and twenty-five of +8 percent. Their arithmetic mean is 5.0000 percent and, because the two paths below use exactly the same thirty numbers, their geometric mean is also identical: (0.90^5 × 1.08^25)^(1/30) − 1 = 4.7676 percent. Nothing about the market differs between the two runs. Only the order does. Start with $1,000,000, withdraw $40,000 at the beginning of every year — the 4 percent rule, in real terms, so the withdrawal never changes — and let the remainder earn that year's return.

Path A puts the five bad years first. The pot is down to $443,066 after five years, and from there the 8 percent it earns can never catch up with a withdrawal that has become 9 percent of what is left; it grinds down and is exhausted in year 28, three years short. Path B puts the same five bad years at the end. It reaches $3,690,299 by year 25, absorbs the identical five crashes and still finishes at $2,031,661. Same average, same volatility, same withdrawal — one retiree runs out, the other leaves two million. Run the two sequences to find the largest fixed withdrawal each could have sustained for the full 30 years and the divergence is just as blunt: $38,811 on path A, $80,385 on path B. The safe rate is not a property of the market. It is a property of the market and your birth date together.

Every assumption in the backtest is doing work

Thirty years is not a retirement at 45. It is roughly the horizon of somebody stopping at 65, and the sustainable rate falls as the horizon lengthens, because the 1 ÷ w floor alone runs out: 25 years of covered spending has to stretch to 40 or 45 rather than to 30, and every extra year is another year of withdrawals taken through whatever the market does. Bengen tested 30 years; the Trinity grid stops at 30 as well. Anyone quoting 4 percent for a 45-year retirement is extrapolating beyond the data, and honest sources say so rather than substituting a number of their own.

The fixed real withdrawal is the other assumption worth naming, because nobody actually follows it. The rule as tested requires you to take the same inflation-adjusted sum after a 40 percent crash as after a 40 percent rally, mechanically, for three decades. Real retirees cut discretionary spending in bad years, skip the inflation rise, take a part-time job, or spend more in their sixties than their eighties. Every one of those behaviours changes the arithmetic — usually in the retiree's favour, since flexibility is precisely what stops path A selling into its own decline. Success being defined as not hitting zero is worth naming too: a run that ended with a few thousand left counted exactly the same as one that ended with millions.

One country, one century

Both studies are built on United States data from 1926 onward. That matters because the twentieth-century US equity market is not a neutral sample. Dimson, Marsh and Staunton have assembled long-run return series for more than twenty markets going back to 1900, published annually in the Global Investment Returns Yearbook, and their central finding is exactly this: outcomes across countries diverge enormously, and the United States sits among the strongest performers rather than in the middle. Several markets in that dataset were closed, expropriated or wiped out at some point in the century, so the countries a naive study can even collect data on are already the survivors.

The honest consequence has two halves, and it is worth stating both. First, direction: a market with a lower long-run real return than the one the backtest used produces a lower sustainable withdrawal rate on the same portfolio and the same horizon, because the withdrawal is being financed out of less growth — that follows from the arithmetic and needs no new data. Second, magnitude: we are not going to put a number on the safe rate for France, Germany, Spain, Portugal or Italy here, because doing that properly means running the same rolling-window exercise on each country's own long-run series, and that series is not reproduced in this article. Anyone quoting you a country-specific safe rate should be able to name the dataset, the currency, the inflation index and the window length. If they cannot, they are quoting the American number with a different flag on it.

Using the number without believing it

The division survives everything above. Spending ÷ w still gives the capital, and 1 ÷ w still gives the multiple, whatever w you decide is defensible. Treat w as the input you are uncertain about and read the range rather than the point: at $30,000 of spending, the honest answer is somewhere between $750,000 and $1,200,000, and knowing that band is more useful than a false precision at 25×. Anything that reduces the spending the portfolio has to cover — a state pension, an occupational pension, a paid-off home — reduces the required pot by the same multiple, which is usually the largest single lever available.

The second lever is the one path A exposes. A withdrawal that flexes downward after a losing year, or a cash buffer of a year or two that lets you avoid selling into a fall, attacks sequence risk directly, which is the risk the headline rate cannot price. Neither is a recommendation for your own circumstances, and neither turns a backtest into a forecast. Regulators across Europe require past-performance material to carry that warning for a reason: a rule derived from history describes what did happen once, in one place, and says nothing about what will happen to you.

Frequently asked questions

Is the 4 percent rule a probability that my money will last?

No. Bengen reported a worst case: the highest opening rate that survived every 30-year window in the US record from 1926 onward. Trinity reported frequencies: the share of historical windows in which each rate and allocation lasted the full period. A frequency in one country's past is not a probability about your future, because your retirement is a single draw from a distribution nobody has observed. Treat both as descriptions of what happened, not forecasts.

The two paths have the same average return. Why do they end so differently?

Because a withdrawal is taken in currency, not in percent. When the pot has fallen to $443,066, the fixed $40,000 is 9 percent of it, and units sold at that price are gone before the recovery arrives. When the same crash lands on a pot of $3,690,299, the identical $40,000 is barely 1 percent. Averages are indifferent to order; a portfolio you are drawing from is not. This is why two people with the same fund and the same rule can retire five years apart and get opposite results.

What rate should someone retiring at 45 use instead?

Neither study answers that, and this article will not invent a figure. What can be said from the arithmetic alone: the 1 ÷ w floor means 4 percent covers 25 years of spending before any real growth, so a 45- or 50-year horizon leans much harder on returns than a 30-year one, and the sustainable rate must be lower. How much lower depends on the market, the asset mix and the flexibility of the spending, which is exactly where a calculator stops and regulated advice starts.

Does the rule transfer to a non-US portfolio?

Not automatically. The direction is clear from the arithmetic — a lower long-run real return supports a lower withdrawal — and Dimson, Marsh and Staunton document how widely long-run outcomes diverge across markets since 1900, with the United States among the strongest. The magnitude is another matter: a defensible national figure requires re-running the rolling-window test on that country's own return and inflation series, in its own currency. This article does not do that, so it does not quote one.

If sequence risk is the real danger, what actually reduces it?

Anything that stops you selling assets into a fall. In the model above, path A dies because a fixed withdrawal keeps taking 9 percent out of a shrinking pot; a withdrawal that flexed downward, or a cash reserve covering a year or two of spending, breaks that loop. Lower fees help too, since they come off the top exactly like a withdrawal. None of these is a recommendation for your situation and none removes the risk — they change how a bad opening decade propagates.