Key takeaways
- Wade Pfau ran the 4% rule across 17 developed countries using 109 years of Dimson-Marsh-Staunton data (1900-2008). The worst-case sustainable rate cleared 4% in only 4 of them: Canada 4.42%, Sweden 4.23%, Denmark 4.08%, the United States 4.02%. Japan's was 0.47%.
- Fix the portfolio at 50% stocks and 50% bonds instead of letting the retiree pick the best allocation with hindsight, and a 4% inflation-adjusted withdrawal was not safe in any of the 17 countries. Canada came closest at 3.94%.
- Bengen's 1994 paper never said a portfolio would last forever at 4%. It said that in no historical case did 4% exhaust a 50/50 portfolio before 33 years, and that 4.25% could exhaust one in 28.
- Bengen's own 2025 revision, using 400 quarterly retirement dates from 1926 to 2024 and seven asset classes, raises the worst-case figure to 4.7% - and puts the average across all those start dates at 7.0%.
- None of the headline studies charges fees or taxes. Pfau shows that adding 1.6% stock and 1.2% bond fund fees cut one SAFEMAX from 4.15% to 3.49%, and Morningstar's 2025 base case falls from 3.9% to 3.4% with 1% of expenses.
The short answer: 4% was one country's worst case, not a law of arithmetic
Can you withdraw 4% of your savings each year in retirement, raise it with inflation, and not run out? On US data covering retirements that began between 1926 and the mid-1970s, yes - in every case tested. That test used a portfolio split between US stocks and US government bonds, held in a tax-deferred account, with no fees charged.
Change any of those conditions and the number moves. Change the country and it moves a long way. That's the whole argument of this piece, and the rest of it is evidence.
What Bengen's 1994 paper actually concluded
William Bengen's Determining Withdrawal Rates Using Historical Data ran a portfolio of 50% common stocks and 50% intermediate-term Treasuries through the Ibbotson Stocks, Bonds, Bills and Inflation series, starting a new hypothetical retirement in each year from 1926 onward. He measured what he called portfolio longevity - how many years the money lasted before it was gone.
His numbers were these. At a 3% initial withdrawal, every retiree in the sample got at least 50 years, which was as far as he plotted; the same held up to about 3.5%. At 4%, "in no past case has it caused a portfolio to be exhausted before 33 years, and in most cases it will lead to portfolio lives of 50 years or longer." At 4.25%, a first-year withdrawal "could exhaust a portfolio in as little as 28 years." At 5%, retirees starting in the late 1960s and early 1970s "might have had only 20 years of funds available." At 6%, 31 of the scenario years ran out of money against 20 that didn't. Bengen described that as "less than a 40-percent chance to successfully negotiate retirement."
Two things follow. First, 4% was chosen against a 30-year minimum, not a lifetime guarantee, and it was the floor of the historical distribution rather than a central estimate. Second, Bengen's other conclusion has been almost entirely forgotten. He found that stock allocations below 50% were counterproductive, and advised "a stock allocation as close to 75 percent as possible."
He also stated his tax assumption plainly - "we are assuming that all retirement assets are held in tax-deferred accounts" - and the paper models no management fees of any kind.
The Trinity study measured success rates, and that is the number people quote
Four years later, Philip Cooley, Carl Hubbard and Daniel Walz at Trinity University published Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable in the AAII Journal. They used 1926-1995 data, the S&P 500 for stocks and long-term high-grade corporate bonds for bonds, and reported the share of historical periods a given rate survived.
Their inflation-adjusted table, for a 30-year payout, gives a 4% withdrawal rate a 95% success rate on a 50/50 portfolio, 98% on 75/25, 95% on all-stocks, 71% on 25/75 and 20% on all-bonds. Those are the numbers behind "the 4% rule works 95% of the time."
It's worth knowing how thin that estimate is. Pfau dissected the authors' 2011 update, which extended the data to 2009. The 96% success rate for an inflation-adjusted 4% withdrawal from a 50/50 portfolio comes from 55 possible retirement start years, of which exactly two failed - 1965 and 1966. Fifty-three out of 55: that's 96.4%. A headline probability is resting on two observations.
The Trinity authors were explicit about what they excluded: "The study did not adjust for taxes or transaction costs."
Pfau ran the same rule in 17 countries, and 13 of them failed it
The correction to a US-derived rule is to run it somewhere else. Pfau's 2010 paper in the Journal of Financial Planning, An International Perspective on Safe Withdrawal Rates: The Demise of the 4 Percent Rule?, did exactly that. It used the Dimson-Marsh-Staunton dataset of stocks, bonds, bills and inflation for 17 developed markets across the 109 years from 1900 to 2008. With a 30-year retirement and data ending in 2008, retirements start between 1900 and 1979 - 80 start dates per country, 1,360 retirement episodes in total.
He then handed each retiree an advantage no real person has. In every country and every start year, the model picks the fixed allocation across that country's stocks, bonds and bills - one of 5,151 combinations. The choice is whichever would have produced the highest sustainable withdrawal rate over the following 30 years. Perfect hindsight, deliberately, so that no one could accuse the study of rigging the allocation to make 4% fail. No fees are charged and no tax is collected.
The measure reported is Bengen's SAFEMAX: the highest inflation-adjusted withdrawal rate that survived the single worst retirement start year in that country's record. The chart above plots it for all 17.
| Country | SAFEMAX, 30 years (%) | Worst start year | Failure rate at a 4% withdrawal |
|---|---|---|---|
| Canada | 4.42 | 1969 | 0.0% |
| Sweden | 4.23 | 1914 | 0.0% |
| Denmark | 4.08 | 1937 | 0.0% |
| United States | 4.02 | 1969 | 0.0% |
| South Africa | 3.84 | 1937 | 1.3% |
| United Kingdom | 3.77 | 1900 | 3.8% |
| Australia | 3.68 | 1970 | 2.5% |
| Switzerland | 3.59 | 1962 | 5.0% |
| Netherlands | 3.36 | 1941 | 2.5% |
| Ireland | 3.28 | 1911 | 25.0% |
| Norway | 3.13 | 1915 | 32.5% |
| Spain | 2.56 | 1957 | 36.3% |
| Italy | 1.56 | 1944 | 62.5% |
| Belgium | 1.46 | 1911 | 40.0% |
| France | 1.25 | 1943 | 42.5% |
| Germany | 1.14 | 1914 | 25.0% |
| Japan | 0.47 | 1940 | 37.5% |
Four countries out of 17 cleared 4%, and the United States ranked fourth of the four. Six came in under 3%. The spread between the top of the table and the bottom: that's a factor of nine.
The most important line in the paper is the one that removes the hindsight. Pfau fixed the allocation at 50% stocks and 50% bonds - the split Bengen and the Trinity authors both used. On the SAFEMAX test, a 4% withdrawal rate wasn't safe in any of the 17 countries. Canada's fell to 3.94%, with the United States and Denmark tied at 3.66%. His own summary: "who but the wealthiest could possibly save enough to live comfortably from the global SAFEMAX withdrawal rate of 0.47 percent?"
The US number is high because the American 20th century was unusually good
Pfau's summary statistics explain the ranking. On DMS data from 1900 to 2008, US real equity returns compounded at 6.01% a year. Only three of the other 16 countries beat that - Australia at 7.26%, Sweden at 7.23%, South Africa at 7.07%. Only four had lower equity volatility than the US figure of 20.43%. Australia is the only country in the set with both a higher return and a lower standard deviation. On the bond side, only three countries beat the US real return of 2.12%, and only two countries had lower average inflation than the American 2.98%.
Dimson, Marsh and Staunton made the wider point in 2004, and Pfau quotes it. US stock market capitalisation grew from about 22% of the world total in 1900 to 54% in 2003, an outcome that "would have been difficult to predict in 1900 and cannot be extrapolated into the future." Over 1900-2002 the US real compounded equity return was 6.3% against 5.4% for their developed-country index. Choosing US data to write retirement rules is itself a choice made after seeing which market won.
A newer study reaches the same place from a different direction. Anarkulova, Cederburg, O'Doherty and Sias, publishing in the Journal of Pension Economics and Finance in 2025, built a dataset of asset-class returns across 38 developed countries. They simulated a 65-year-old couple facing both market risk and mortality risk. Their finding: a couple willing to accept a 5% chance of running out of money could withdraw 2.31% a year. That's a different question from Bengen's - it prices longevity risk as well as returns - but it points the same way.
The case against the critics is stronger than the critics usually allow
A piece that stopped here would be as lazy as one that treats 4% as a law. There are four serious objections, and they're good.
4% was always a worst case, and worst cases are rare. This is the big one. Bengen's own updated research, set out in his 2025 book A Richer Retirement and summarised in an AAII interview, ran 400 quarterly retirement start dates from January 1926 to the end of 2024 across seven asset classes - adding US micro-cap, mid-cap and international stocks and Treasury bills to his original mix. The worst-case rate rises to 4.7%, and the worst start date is October 1968. But the average SAFEMAX across all 400 start dates is 7.0%, and Bengen's own estimate is that a retiree buying in at the April 2009 market bottom could have supported 8.0%. A retiree who took 4% was, in the typical historical case, taking a little over half of what the market would have supported.
Morningstar's 2025 research puts the same distribution on US history. For rolling 30-year windows starting between 1927 and mid-1995, the starting safe rate on a 50/50 portfolio ranged from 3.9% for the late-1968 retiree to 10.5% for the mid-1982 retiree. On an all-equity portfolio the range is wider still - 3.4% for the unlucky 1929 retiree, 18% for the July 1932 one.
The international worst cases are mostly wars. Look at the failure years in the table: Germany 1914, Belgium 1911, France 1943, Italy 1944, Japan 1940. Pfau concedes the point himself - "some of the worst outcomes were connected with World Wars I and II, and we can hope that such devastating wars will never happen again." Strip the two world wars out and the international spread narrows toward the 3.3%-4.4% band occupied by the top nine countries.
The method has a bias of its own. Pfau's countries hold only their own domestic assets. A Belgian retiree in 1911 with a globally diversified portfolio is a different experiment, and a better-behaved one. Pfau has also shown that rolling historical simulations over-weight the middle of the sample. In a 30-year study of 1926-2009, each year from 1955 to 1980 appears in 30 of the simulated retirements while 2009 appears in one. Real bond returns over 1955-1980 averaged -1.4% against 5.2% for the rest of the period. The method quietly loads the dice against bonds.
Nobody spends like this. The 4% rule finances a rigid, inflation-linked spending plan out of a volatile portfolio. Scott, Sharpe and Watson priced that mismatch. A typical rule allocates 10%-20% of a retiree's initial wealth to surpluses that are never spent, plus another 2%-4% to overpayments for the spending pattern it buys. Real retirees cut back after a crash, which changes the arithmetic substantially - that's a separate question, and a separate piece.
Every headline number here is before fees and before tax
This is the least discussed and most consequential gap. Bengen assumed tax-deferred accounts and modelled no fund charges. Trinity "did not adjust for taxes or transaction costs." Pfau charges nothing, noting his figures suit "withdrawals from a Roth IRA" and must otherwise be read as pre-tax. Morningstar states that its research "doesn't incorporate the impact of expenses or taxes."
Both Pfau and Morningstar quantify the omission. Pfau takes Bengen's 4.15% SAFEMAX for a 50% large-cap and 50% intermediate-government-bond portfolio, applies mutual fund charges of 1.6% on stocks and 1.2% on bonds. The SAFEMAX then falls by 0.66 percentage points to 3.49%. Morningstar assumes 1% in annual expenses on a 60/40 portfolio drawn from a tax-deferred account and its 3.9% base case drops to 3.4%.
Put that next to the country table. The gap between the United States at 4.02% and the Netherlands at 3.36% is 0.66 percentage points - the same size as Pfau's fee haircut. For the nine countries at the top of the table, what a retiree paid in charges mattered about as much as which country they retired in. The long-run cost of active management isn't a separate topic from this one.
Where the current estimates sit
Morningstar's State of Retirement Income: 2025, published on 3 December 2025, is the closest thing to an annually refreshed number. Its base case is a 3.9% starting rate for a new retiree seeking level inflation-adjusted spending over 30 years, at a 90% probability of finishing with money left. That's not a backtest: it comes from 1,000 Monte Carlo paths built on forward-looking return and inflation assumptions, with expected inflation of 2.46%.
Its recent history reads 3.3% in 2021, 3.8% in 2022, 4.0% in 2023, 3.7% in 2024 and 3.9% in 2025. That's a number that moves with valuations and yields rather than a constant. The highest rate in the 2025 model came from portfolios holding between 30% and 50% in equities, which is a long way from Bengen's 75%. Two things explain the difference: Morningstar is forecasting rather than backtesting, and today's bond yields do more work than they did in the samples Bengen and Trinity used. A 30-year TIPS ladder priced on 30 September 2025 supported 4.5%. Flexible spending rules in the same report reach 5.7%, but that's a different mechanism. The 5.7% belongs to Morningstar's constant-percentage and endowment methods, which reset spending off the portfolio balance each year. Its Guyton-Klinger guardrails row starts at 5.2%. Both sit in the same Exhibit 15, and our piece on what flexible withdrawal rules cost in spending volatility works through the trade.
What this evidence cannot tell you
All of it is backtesting or simulation built on backtests. Pfau's 1,360 retirement episodes come from 17 country histories, and the 30-year windows overlap heavily, so they're nowhere near 1,360 independent observations. Trinity's 30-year table rests on 41 overlapping windows drawn from 70 years of one country's data.
Sequence risk is why the average case isn't the planning case. Two portfolios can earn the same average return over 30 years and end in completely different places depending on when the bad years land. A withdrawal taken during a crash sells more shares to raise the same income. That's why the 1968 and 1969 retirees anchor almost every one of these studies while the 1982 retiree could have taken 10.5%.
The 30-year horizon is a convention, not a fact about anyone. Pfau's perfect-foresight assumption inflates his SAFEMAX figures relative to anything achievable; his 50/50 results, which don't use it, are the more realistic comparison. His returns are domestic-currency, domestic-asset returns, so they don't describe an investor who held foreign equities alongside domestic ones. And Morningstar's 3.9% inherits the accuracy of Morningstar's own capital markets assumptions, which are a forecast like anyone else's.
What would change the conclusion
If US asset returns mean-revert toward the international average. That's the scenario Pfau wrote the paper to illustrate. His US SAFEMAX of 4.02% sits at the top of the international distribution because US returns did. A US century that looked like the international median would produce a SAFEMAX in the low threes.
If the war-driven failures are treated as unrepeatable. Drop Germany, France, Italy, Belgium and Japan and the 12 countries that remain run from 2.56% in Spain to 4.42% in Canada. That's still a wide spread, and eight of the 12 sit below 4%. But it's an argument about a point or so of withdrawal rate rather than about whether the rule collapses.
If spending is allowed to flex. Every number above assumes a retiree keeps withdrawing the same real amount through a 50% market decline, which few people do. Relax that and the safe rate rises materially in every study that has tested it.
If the horizon isn't 30 years. A 30-year rule tells you almost nothing about a 45-year retirement, and Bengen's charts show the damage from a bad start reaching back 20 years or more.
The thing worth watching isn't the headline percentage but the two inputs underneath it. Those are what your portfolio actually costs you each year, and how far your current withdrawals sit from the rate you planned on. Both are things a tracker such as LedgerTouch can show continuously; neither is something the research can settle for you.