What is a safe withdrawal rate?

What “safe” is supposed to mean

The phrase has a precise, narrow meaning in the research: you pick a percentage of your portfolio in the first year of retirement — say 4% of $1,000,000, or $40,000 — and every year after that you withdraw the same dollar amount adjusted for inflation, regardless of what markets do. A rate is called “safe” if a portfolio following that schedule survived a full retirement (usually 30 years) in every historical period tested. That is a claim about the past, not the future: the rate is whatever the single worst start date in the dataset could support. Most start dates could have supported considerably more; a few came close to the edge. The honest way to read any safe withdrawal rate is therefore as the low end of a historical range, not as a point prediction.

1994: Bengen and the birth of the 4% rule

The number traces to William Bengen's October 1994 paper in the Journal of Financial Planning, “Determining Withdrawal Rates Using Historical Data.” Bengen simulated a retiree holding 50% US large-cap stocks and 50% intermediate-term Treasuries, starting retirement in every year since 1926, withdrawing an inflation-adjusted amount for 30 years. The worst historical start he found was a retiree beginning around October 1968 — straight into the 1969–70 and 1973–74 bear markets with the 1970s inflation shock layered on top. That worst case supported an initial rate of about 4.15%, which was rounded down to 4% in later retellings, and the round number stuck.

Two details are usually dropped from the retelling. First, 4.15% was the floor: typical start years in Bengen's data supported noticeably higher rates, so the historical picture is a band with 4% at the bottom, not a cluster around 4%. Second, Bengen himself never stopped revising the figure. Adding small-cap stocks to the mix lifted his worst-case rate (he called it SAFEMAX) to roughly 4.5% in his mid-2000s work, and in his 2025 book A Richer Retirement he estimates the worst-case rate near 4.7% using a broader multi-asset portfolio. When the author of the 4% rule treats it as a moving estimate, it is reasonable for everyone else to do the same.

1998: the Trinity study reframed it as success rates

Four years later, three professors at Trinity University — Cooley, Hubbard, and Walz — published “Retirement Savings: Choosing a Withdrawal Rate That Is Sustainable” in the February 1998 AAII Journal. Using data from 1926 to 1995 (stocks plus long-term high-grade corporate bonds, a different bond sleeve than Bengen's), they asked a different question: for each combination of withdrawal rate, stock/bond mix, and horizon, what fraction of historical windows ended with money left over? Their headline result is the one most people half-remember: a 4% inflation-adjusted withdrawal from a 50/50 portfolio succeeded in roughly 95% of historical 30-year windows, with stock-heavier mixes doing somewhat better.

The reframing matters. “95% success” is a subtly different claim than “survived the worst case” — it concedes up front that about one historical window in twenty failed. It also introduced the table-of-success-rates format that nearly every retirement calculator since has copied: rates from 3% to 12% across mixes and horizons, so a reader can see how quickly the odds decay as the withdrawal rate climbs.

What both studies leave out

The original studies are careful, but their assumptions are narrower than the way the 4% rule gets used. Five gaps matter most:

• Taxes. Both studies treat the withdrawal as fully spendable. In reality a traditional-IRA withdrawal is ordinary income, and a taxable-account withdrawal owes capital-gains tax on its gain portion. A 4% gross withdrawal can shrink meaningfully once real 2026 brackets are applied — the gap between gross and net-spendable is one of the largest silent errors in retirement math.
• Fees. The backtests assume zero expense ratios and zero advice fees. Every basis point of fee drag comes straight out of the rate the portfolio can actually support.
• The 30-year horizon. The 4% figure is specifically a 30-year result. Early retirees planning 40–50 year horizons face more chances for a bad sequence, and the supported floor drops as the horizon stretches.
• US-only history. The datasets cover the single strongest major equity market of the 20th century. A backtest over US returns bakes in a survivorship assumption that the next 30 years will rhyme with an unusually good past.
• Rigid spending. The mechanical CPI-adjusted schedule is a modeling convenience, not a description of human behavior. Real retirees cut back in crashes and spend more in booms — which is exactly the observation the alternatives below build on.

There is a sixth, quieter limitation: the classic studies are valuation-blind. They hand the same 4% to a retiree starting at extreme equity valuations and one starting after a crash. Forward-looking estimates exist largely to fix that.

What current research estimates: roughly 3.3% to 4.7%

Morningstar's annual State of Retirement Income report is the most widely cited forward-looking estimate. Instead of replaying history, it runs Monte Carlo simulations seeded with current valuations and bond yields, and asks what starting rate leaves money after 30 years with 90% probability. Because the inputs move, the answer moves: 3.3% in the 2021 edition, 3.8% in 2022, 4.0% in 2023, 3.7% in 2024, and 3.9% in the December 2025 edition covering people retiring in 2026 (asOf 2025-12, for a 30-year horizon with 30–50% in equities). The same report notes that flexible strategies — delaying Social Security, guardrails-style adjustments, TIPS ladders — can support meaningfully higher starting rates than the fixed approach.

Put the strands together and the honest summary is a band, not a number. Forward-looking Monte Carlo at a 90% success bar currently lands around 3.7–3.9%; the classic historical worst-case method gives 4.0–4.15% on Bengen's original two-asset portfolio; and Bengen's own updated multi-asset worst case reaches 4.5–4.7%. Which end of the band applies to you depends on assumptions you should get to see: the horizon, the asset mix, the success threshold, and whether taxes and fees have been subtracted yet.

The alternatives: trading steadiness for a higher start

Guardrails (Guyton–Klinger)

Jonathan Guyton and William Klinger's March 2006 Journal of Financial Planning paper, “Decision Rules and Maximum Initial Withdrawal Rates,” attacks the rigidity directly. Their most-used rules form a 20/10 guardrail: if markets fall far enough that your current withdrawal has drifted to more than 20% above the initial rate, you cut the withdrawal by 10% (capital-preservation rule); if it drifts more than 20% below, you raise it by 10% (prosperity rule). Because the plan promises to cut spending in bad sequences, it can start higher: the paper reports initial rates of 5.2–5.6% at its 99% confidence bar for portfolios with at least 65% stocks.

The cost is variability. In a 1966-style or 2000-style start, the rules fire repeatedly, and mid-retirement spending can end up well below where a fixed-rate plan would have held it. A higher starting rate is not free money — it is a trade: more income up front in exchange for accepting cuts when the sequence turns against you. Before adopting it, you should see the income floor across historical sequences, not just the starting rate.

VPW (variable percentage withdrawal)

The Bogleheads community's VPW method drops the fixed dollar path entirely. Each year you withdraw a percentage of the current balance, with the percentage taken from a PMT-style amortization over your remaining horizon — so it rises as you age, and the schedule deliberately spends the portfolio down rather than preserving a bequest. By construction the money cannot run out early; what varies instead is your income, which floats up and down with the portfolio. In a deep bear market, a VPW retiree's annual income drops roughly in proportion to the portfolio, which is survivable for flexible spending but dangerous for fixed obligations like rent or insurance premiums.

Three strategies, one trade-off

How to use any of these numbers

Three habits keep the research honest when you apply it to your own plan. First, insist on a range — a worst start year, a median, and a strong start year — because a single success percentage hides how different the paths inside it are. Second, convert gross to net-spendable: the studies' withdrawal is pre-tax, and your account mix (traditional, Roth, taxable) decides how much of each dollar you keep. Third, subtract your actual fees before comparing your rate to any published figure.

Our safe withdrawal rate calculator runs all three strategies on this page side by side — fixed real, Guyton–Klinger 20/10, and VPW — across every rolling start year since 1928, with an explicit fee input and a bracket-aware gross-to-net layer using 2026 IRS tables, and reports each result as a low / median / high band with the worst start year called out. For the mirror-image question — given a spending level, how long does the money hold out — the how long will my money last tool reports a depletion-age range rather than a single date.