Updated: August 17, 2026
OPRS may receive compensation when readers open an account through partner links on this page. Our analysis is based on independent research, BBB data, and IRS publications.
The phrase safe withdrawal rate reads like a physical constant. It is closer to a weather forecast. The number a retiree finally sees on a spreadsheet depends on which decades of U.S. data the researcher used, what the equity weight was, whether taxes and fees were modeled, and above all which start years happened to survive the sample. Change any of those inputs and the answer moves by a full percentage point.
This page walks through the historical data landscape retirees run into when they read academic and practitioner literature on the 4 percent rule and its cousins. The goal is not to endorse a number. It is to help a reader tell whether a cited rate comes from a rigorous study or from a chart that skipped the worst starting years.
The three datasets almost every study starts from
Nearly every published U.S. safe-withdrawal study leans on one of three source datasets. Understanding which one a paper uses is the first step in reading it critically.
- Ibbotson SBBI (Stocks, Bonds, Bills, and Inflation), 1926 to present. Now maintained by Morningstar. Annual U.S. total returns on large-cap stocks, small-cap stocks, long government bonds, intermediate government bonds, U.S. Treasury bills, and inflation. This is the dataset William Bengen used in his 1994 paper and the one behind the original Trinity Study.
- Robert Shiller monthly S&P data, 1871 to present. Extended U.S. equity total returns and CPI, downloadable from Shiller’s Yale page. Adds roughly 55 years before SBBI starts, at the price of thinner coverage for bonds.
- Ken French data library. Fama-French factor returns and portfolio deciles for U.S. equities back to 1926 and international coverage after that. Used in studies that need factor tilts or global asset classes rather than a plain S&P proxy.
All three datasets are cited in mainstream finance research, and none of them is wrong. They simply cover different windows and different assets. Any withdrawal-rate paper that does not name its source is either recycling numbers or hiding assumptions.
Bengen 1994: the 4 percent number in its original context
William Bengen was a fee-only financial planner in Southern California when he published Determining Withdrawal Rates Using Historical Data in the October 1994 Journal of Financial Planning. He tested every rolling 30-year window in the SBBI data from 1926 through 1992, using a 50 percent large-cap stock and 50 percent intermediate government bond portfolio, rebalanced annually.
Bengen found that a 4 percent initial withdrawal, then adjusted upward each year by realized CPI, would have survived every 30-year historical window in that sample. He called this the SAFEMAX rate. In the same paper he showed that raising the initial rate to 5 percent broke a handful of the worst windows. The 4 percent figure was, from the beginning, the floor set by the very worst historical starts, not the average or the median.
Two constraints in the original paper are almost always dropped when the 4 percent number is quoted. Bengen assumed no advisor fee, no fund expense ratio, and no taxes on the withdrawals. He also assumed a 30-year horizon, not 40 or 50. Add a 1 percent annual fee and the safe rate drops. Extend the horizon to 40 years and it drops again. The often-quoted 4 percent is a ceiling under Bengen’s exact assumptions, not a universal number.
The Trinity Study: extending Bengen through 1997
In 1998, three finance professors at Trinity University in San Antonio published Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable in the AAII Journal. Philip Cooley, Carl Hubbard, and Daniel Walz extended Bengen’s approach through 1995 data and later updated it through 1997 and beyond. Their paper was picked up by the financial press and became the source most retirees actually read.
The Trinity authors reported success rates rather than a single SAFEMAX. For a 30-year retirement at 4 percent initial withdrawal with CPI adjustment, a 75 percent stock and 25 percent bond portfolio succeeded in about 98 percent of historical windows. Lower equity weights produced fewer successes at high withdrawal rates but held up well at 3 to 4 percent. The Trinity results are broadly consistent with Bengen; they add nuance rather than a new rule.
Later Trinity updates and independent extensions using data through the 2000s and 2010s produced similar bands. The 4 percent floor for a balanced portfolio holds up in the U.S. historical sample. What changes across updates is the exact fringe: whether a start year like 1966 finishes just above zero or just below.
Published SWRs by allocation and horizon: the honest range
Combining Bengen, Trinity, and the follow-up literature, the published historical safe withdrawal rates for a 30-year retirement in the U.S. sample cluster in a band from about 3.5 percent to about 6 percent. The wide end is what happens if the retiree gets a lucky start year. The narrow end is the SAFEMAX-style floor set by the worst starts.
- 50 to 75 percent equity, 30-year horizon: SAFEMAX around 4.0 to 4.2 percent, median historical rate closer to 6 percent. The 4 percent rule sits at the bottom of this band.
- 100 percent equity, 30-year horizon: SAFEMAX around 4.0 percent (higher volatility offsets higher expected return), median historical rate around 6.5 to 7 percent. All-stock did not raise the floor.
- 25 percent equity or lower, 30-year horizon: SAFEMAX often drops below 4 percent because bond-heavy portfolios lost real ground during the 1966 to 1982 inflation window.
- 40 or 50-year horizon (early retirement): SAFEMAX drops by roughly 0.5 percentage point per additional decade. A 50-year plan pulls the floor closer to 3.5 percent.
Any published number outside this band deserves scrutiny. A quoted 8 percent safe rate almost always drops fees, drops taxes, or uses a starting decade that never saw a real bear market. A quoted 2 percent almost always adds a large fee or uses international data with worse tail outcomes than the U.S. sample.
Why the worst historical starts drive the safe rate
The 4 percent floor is not set by the average retiree in the historical data. It is set by the four or five worst-timed retirees, each of whom retired just before a decade-long real drawdown. Four start years show up in almost every SWR paper as the binding constraints.
- 1929. The Great Depression start. Equity prices collapsed roughly 80 percent from peak. A retiree here needed a low withdrawal rate and heavy bond exposure to survive.
- 1937. The second leg of the Depression, following the 1937 to 1938 recession. Slightly less severe than 1929 but hit a portfolio that had already been damaged in prior years.
- 1966. The most commonly cited binding case. Stocks were flat in nominal terms for over 15 years and lost roughly half their real value to the 1970s inflation. Bonds also lost real ground. Almost every SAFEMAX study points to a 1966 retiree as the worst case in the post-war U.S. sample.
- 2000. The dot-com peak. A 2000 retiree faced two large equity drawdowns in a decade (2000 to 2002 and 2007 to 2009), then a slow real recovery. Long enough now to appear in updated Trinity-style studies.
Every other start year in the U.S. sample would have supported a higher initial withdrawal than 4 percent. In the median historical window, a retiree could have taken close to 6 percent and still ended the 30 years with a positive balance. The 4 percent rule protects against the four bad starts above, at the cost of underspending in every other window.
The international-data critique: the U.S. is not the world
Wade Pfau, Javier Estrada, and other researchers have run the same withdrawal-rate exercise using international equity and bond data from the Dimson-Marsh-Staunton (DMS) database, which covers 21 countries back to 1900. The results reset expectations sharply.
Estrada found that the U.S. was one of the best-performing markets in the DMS sample. Applied to countries such as Italy, Belgium, Japan, or Germany, a 4 percent initial withdrawal frequently exhausted the portfolio inside 30 years, often within 20. Even relatively good markets such as the U.K., Canada, and Australia produced SAFEMAX rates below 4 percent for the worst domestic starts.
Pfau reached a similar conclusion in a widely cited 2010 paper. Using the same 30-year rolling-window method on 17 non-U.S. developed markets, he found a global median SAFEMAX closer to 3.5 percent, with several countries below 2 percent. His main takeaway is not that the U.S. sample is wrong; it is that the U.S. sample is an outlier, and using it to calibrate a global safe rate assumes American exceptionalism holds for the next 30 years.
Historical SWR by decade of retirement start
The table below groups the SAFEMAX rate for a 30-year U.S. retirement by the decade in which the retirement started, using a 50 percent stock and 50 percent bond portfolio (the Bengen baseline). Figures are typical of the values reported in Bengen 1994 and later extensions in the practitioner literature. They illustrate range, not point estimates.
| Retirement start decade | Typical SAFEMAX (50/50) | What drove the result |
|---|---|---|
| 1920s | ~4.1 percent | 1929 start binds; recovery arrived within the horizon |
| 1930s | ~5.5 percent | Depression already priced in; strong recovery followed |
| 1940s | ~6.5 percent | Post-war expansion, moderate inflation |
| 1950s | ~6.0 percent | Strong equity run; stable bonds |
| 1960s | ~4.0 percent | 1966 start binds; 1970s inflation destroyed bond real return |
| 1970s | ~4.5 percent | Retirees hit stagflation early but rebounded post-1982 |
| 1980s | ~7.0 percent | Bull market start, high initial bond yields |
| 1990s | ~5.5 percent | 2000 dot-com peak damaged mid-window balances |
The sample-size problem: only about three independent 30-year windows
The SBBI data starts in 1926. Through the end of 2026, that is roughly 100 years. Divide by 30 years and the U.S. history contains only about three non-overlapping 30-year retirement windows. Every published SAFEMAX chart shows dozens of rolling windows, but rolling windows share years and are not statistically independent.
The practical consequence is that confidence intervals on any historical safe rate are far wider than the point estimate suggests. A study reporting a 4.15 percent SAFEMAX cannot honestly claim two-decimal precision from three effective observations. The right way to read these numbers is as a range: somewhere in the neighborhood of 3.5 to 6 percent for the U.S. sample, with the exact figure depending on how a small number of start years happened to unfold.
Adding international data helps at the margin because it multiplies the effective sample, but only if the retiree is willing to accept that other countries’ returns are relevant to their U.S. plan. Most retirees are not, and researchers disagree on whether they should be.
Why forward-looking SWRs tend to be lower than historical
Wade Pfau, David Blanchett, and other researchers have argued for the last decade that the historical 4 percent should probably be adjusted downward for a retiree starting today. Their argument does not depend on predicting a bear market. It depends on two starting-condition variables: bond yields and equity valuations.
Bengen’s SBBI window covered decades where intermediate government bonds yielded 4 to 8 percent nominal. Bond yields in August 2026 sit closer to 4 to 5 percent. That carries more expected return than the 2010s but less than the 1980s starting yields that produced the top-of-band SWRs.
The same is true of equity valuations. A 30-year plan starting from a Shiller CAPE above 30 has historically produced lower average returns than one starting from a CAPE below 15.
Blanchett, Pfau, and Michael Finke published a 2013 paper titled Low Bond Yields and Safe Portfolio Withdrawal Rates. They argued that a 4 percent rate could not be considered fully safe when starting bond yields were near zero. The critique carries less force in a higher-yield environment. It still suggests that a retiree starting today should not assume the historical median. A rate in the 3.5 to 4 percent range remains the mainstream cautious anchor.
How the safe-rate framework is used and abused
The 4 percent rule and its Trinity variants were designed as a stress test, not a spending policy. Bengen himself has said in later interviews that the safe rate was meant to answer the narrow question of what withdrawal a retiree could commit to today and never have to cut. It was not meant to be the retiree’s actual withdrawal in every year.
Two common abuses show up in advisor materials and consumer articles.
- Presenting a single number as a law. A brochure that says the safe withdrawal rate is 4 percent, without naming the dataset, the horizon, the allocation, or the fee assumption, is oversimplifying. The number rests on assumptions the reader deserves to see.
- Inflating the rate by dropping constraints. A retirement calculator quietly assuming no fees, no taxes, a 25-year horizon, and end-of-year rebalancing can produce a safe rate 1 to 2 percentage points higher than the same setup with realistic frictions. The higher number sells more advisor engagements; it also produces more retirees running out of money.
Advisors who use the framework well tend to combine it with a rules-based flex mechanism: a Guyton-Klinger style guardrail, a floor-and-ceiling policy, or a periodic Monte Carlo re-check. The 4 percent number then serves as an anchor for the initial-year draw, not a promise for every year of the plan.
What the SEC and the primary literature actually say
The U.S. Securities and Exchange Commission’s Office of Investor Education has published a public investor bulletin explicitly on the 4 percent rule. It flags several of the same caveats covered above: the rule is one study’s result under specific assumptions, is sensitive to the starting decade, and does not account for advisor fees or taxes. It is a rare case of a federal regulator publishing a plain-language read of an academic result.
A retiree who wants to check any published SWR number against the primary sources can start with three documents. Bengen’s original 1994 paper is freely available in PDF from the retailinvestor.org archive. The Trinity Study reprint sits on the AAII site. The SEC investor bulletin ties the two together in plain English. Reading the three before accepting a broker’s quoted safe rate is a low-cost sanity check.
Sources cited
- U.S. Securities and Exchange Commission, Office of Investor Education and Advocacy, Investor Bulletin: The 4% Rule, an updated plain-language explainer of the origin, assumptions, and limitations of the 4 percent withdrawal rule for individual investors.
- William P. Bengen, Determining Withdrawal Rates Using Historical Data, Journal of Financial Planning, October 1994, the original SBBI-based paper that introduced the 4 percent SAFEMAX rate for a 30-year U.S. retirement.
- Philip L. Cooley, Carl M. Hubbard and Daniel T. Walz, Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable, AAII Journal reprint of the Trinity Study, presenting the success-rate tables that extended Bengen’s method across allocations and horizons.