A withdrawal rate is not the plan. The order of the years is.
The 4% rule is the most-cited number in retirement planning. Withdraw 4% of the portfolio in the first year of retirement, index that amount for inflation in every year after, and the portfolio is designed to last thirty years. It is also the most misused number in the same planning, because almost nobody quotes it with its failure mode. The rule came from dated historical data, and the data contains a first-decade failure mode that listicles skip. This is the data, start year by start year, with the orders the market actually ran.
The Rule, and the Study That Made It
The rule has a date and an author. William Bengen, a financial planner, published Determining Withdrawal Rates Using Historical Data in the October 1994 issue of the Journal of Financial Planning. He simulated thirty-year retirement periods against U.S. market data back to 1926, with a portfolio of U.S. large-cap stocks and intermediate Treasury bonds, tax-free. Four percent was the highest first-year withdrawal rate that never exhausted the portfolio in any thirty-year window of that dataset, with withdrawals indexed for inflation in each later year. The rule is also called the Rule of 300 — or, turned around, “save 25 times your first year of spending.”
Two years later, three finance professors at Trinity University — Cooley, Hubbard, and Walz — published their own backtest in the AAII Journal, February 1998, against Ibbotson market data from 1925 to 1995, payout periods of fifteen to thirty years, and the same definition of success: assets left at the end. Their finding is the one the industry quotes: for stock-dominated portfolios, withdrawal rates of 3% to 4% continue to produce high portfolio success rates. The original authors updated the analysis with data through 2009 and published it in 2011.
Bengen said the 4% was meant as a worst case. His own example was a retiree who retired in 1968, at a market peak, before the protracted recession and high inflation of the 1970s. In that order, 4% lasted thirty years. He has said the historically average safe rate is closer to 7%.
That is the whole pedigree of the rule: a 1994 paper, a 1998 confirmation, a 2011 update. None of it is a rate. Each of it is a rate tested against an order of years.
Year One: Why the First Decade Decides
A retirement portfolio is drawn down. Every year the plan takes a fixed real amount out of a portfolio that moves with the market. When the market is up, the same dollar withdrawal sells fewer shares, and the remaining shares keep compounding. When the market is down, the same dollar withdrawal sells more shares — and those shares are the ones the later years need to compound back.
That is sequence-of-returns risk. The returns of the early years are not interchangeable with the returns of the late years. The same three returns, in the worst order first, do damage the rest of the sequence must repay; in the best order they do almost none. The damage concentrates in the first decade. The first three years are the test; inside them, the worst three do most of the damage.
This is not a thought experiment. In the 2008 crisis, American workers lost an estimated $2 trillion in retirement savings. At the start of 2008, 43% of 401(k) participants held more than 70% of their account in stocks. The buffer problem was live in real accounts, and the order was the worst of the century.
The Worst Three: Five Retirements, Five Orders
Here is the table the listicles skip. Five start years, each the worst or near-worst of its era, each a thirty-year plan at 4% indexed for inflation. The equity figures are U.S. large-cap price returns as published: the S&P Composite 90, the S&P 500’s predecessor, and the Dow for the older decades; the S&P 500 after 1957.
| Retired into | The order the market actually ran | The 30-year plan at 4% |
|---|---|---|
| 1929, after the crash year | The S&P Composite 90 set its record close of 31.71 in August 1929 and fell to 4.43 by June 1932 — down 86%. The Dow lost 34% in 1930, 53% in 1931, and 23% in 1932. | The window the rule was built to survive: in the 1994 dataset, 4% never exhausted a portfolio in any thirty-year window of the U.S. data back to 1926. |
| 1966, at the peak | The S&P Composite 90 closed at 94.06 on February 9, 1966 — the decade’s high — and the 1966 bear market took the Dow down 19%. The first decade then contains 1973, when the Dow fell 17%, and 1974, when it fell 28%. | Survived in the original studies. Trinity’s 1925–1995 data covers the window; their finding for 3–4% in stock-dominated portfolios is high success rates. |
| 1973, into the double bear market | The Dow lost 17% in 1973 and 28% in 1974, the double bear market, then a decade of 1970s stagnation. The S&P 500’s 1982 low — 102.42 on August 12, 1982 — was the base it tripled from by August 1987. | Survived. The window, 1973–2003, is covered by the original authors’ 2009 data update. |
| 2000, at the tech peak | The S&P 500 fell from 1,552.87 on March 24, 2000 to 768.83 on October 10, 2002 — down 50%. By the end of 2002, market capitalization lost stood at $5 trillion. | In progress — 26 of 30 years observed through 2025. Still solvent; the index now sits about 9 times its 2002 low. |
| 2008, in the crisis | 2008: the S&P 500 fell 38.5%, the worst annual loss on record at the time. The March 2009 trough sat 57% below the October 2007 peak. | In progress — 18 of 30 years observed through 2025. Still solvent; the index now sits about 10.5 times its March 2009 trough. |
Count the complete windows first. The three with full thirty-year data — 1929, 1966, 1973 — all still held assets at year thirty. The other two are still running: 2000 has 26 of 30 years observed through 2025, and 2008 has 18. Zero of the five plans ran out of money. That is the headline the 4% rule deserves.
The second number is the one that keeps it honest. A rolling thirty-year analysis of 1926–2014, using the S&P 500 and five-year U.S. Treasuries with inflation-indexed withdrawals, found the worst window sustained only 3.5% — the best, 10%. The original 1994 and 1998 studies ended in the mid-1990s and found 4% never failed in their data; the table ending in 2014 carries the data through two further bear markets and puts the edge of the data at 3.5%, not 4%. Both are true, both are dated, and the assumption sets differ — the later table uses five-year Treasuries where the original used intermediate bonds. That is not a contradiction to resolve. It is the spread, and it lives in the order.
The 2026 Starting Point
A rate is only as good as the starting point it is applied to. Here is the dated set, as of September 2026:
- Inflation: the consumer price index was up 3.4% over the twelve months through August 2026, per the Bureau of Labor Statistics. Core was up 2.4%; energy was up 16.3% on the oil shock of the U.S.–Iran war.
- Rates: the 10-year Treasury yield stood at 5.01% on September 16, 2026, a high not seen since 2007. The two-year was at 4.74%. The Federal Reserve hiked to a 3.75–4.0% range on that date, its first hike in three years.
- The market: the S&P 500 closed at 7,022.95 on April 15, 2026, a record high. The index’s long-run record since 1926 is about 9.8% a year including dividends, about 6% after inflation.
Read that as an order, not a rate. The rule was calibrated on U.S. data that ended in the mid-1990s. A 2026 start sits at a record high, with inflation above the rule’s 3% planning assumption and the 10-year at 5%. The dated data says what that implies. Four percent survived most thirty-year windows in 1926–2014; the worst window sustained 3.5%; the same rule tested against other developed markets was safe in only 4 of 14 countries in Wade Pfau’s replication, and his 2020 estimate for the then-current regime was 2.4%; and a 2010 Journal of Financial Planning analysis found that realistic assumptions about the chance of an emergency withdrawal each year cut the sustainable rate from 4% to about 3%.
So the honest 2026 adjustment is not a new magic number. It is a spread with a buffer in the early years: plan the first withdrawal at the worst-window figure, 3.5%, hold the first one to three years of spending in a bonds-and-cash ladder, and treat 4% as the ceiling of the data rather than its center. For a horizon longer than thirty years — retirement at 40, 50, or 55 instead of 65 — the dated studies point to rates closer to 3%, not 4%.
The Buffer: What the Downturn Sells
The buffer is the piece of the plan the rule’s number hides. A bonds-and-cash ladder sized to the first several years of spending is what the downturn sells instead of the portfolio. It is sized to the withdrawal, not to the rate.
The arithmetic, dated to the 2008–09 sequence. A retiree starts January 1, 2008, with a $1,000,000 portfolio and a plan to withdraw $40,000 in year one — 4% — indexed at a 3% planning assumption.
- No buffer, the plan fully in stocks. 2008 takes the portfolio down 38.5%, to about $615,000. The first $40,000 is sold into that decline. 2009 returns 23.5%, and after the second withdrawal, about $41,200, the portfolio stands at about $669,000. Two years in, the plan has sold its shares at the bottom of the worst decade of the century.
- With a one-year buffer, $40,000 of bonds and cash held alongside the portfolio. Year one’s spending comes from the buffer, so the stock sleeve falls 38.5% but is not sold. The 2009 recovery lifts the sleeve 23.5%, and the second withdrawal is sold from a sleeve above its year-end-2008 level. Two years in, the portfolio stands at about $718,000.
Same rate, same market, same order. The spread between the two plans after two years is roughly $49,000, and it came entirely from not selling low. That is what the buffer buys.
The Spread: What This Means for Your Number
Your number — how much you need to have saved — is the job of the retirement number for 2026, and the mechanics are in the retirement calculator. Take that number. The question this table adds is whether it survives the first decade, and the answer depends on the order, not the rate.
The interface between the number and the drawdown is a multiple. At 4%, it is 25 times your first year of spending. At the worst window in the 1926–2014 data, 3.5%, it is about 29 times. A retiree at 65 with a thirty-year horizon plans at 25x if they trust the center of the data and at 29x if they plan against its edge. Retire at 55, and the horizon stretches past thirty years, and the dated studies point to the 3% end of that range — 33 times — rather than the 4% end.
That is the stress test, the whole of it. The 4% rule survives every dated start year the data holds — and the worst window the data holds sustains only 3.5%. The spread between the plan that survives the sequence and the one that does not is not a small number. It is the difference between selling into 1931 and holding through it, between a 2008 withdrawal taken at the trough and one taken from the recovery. A rate is what you pick. The order is what you survive.