Safe Withdrawal Rate History: Every 30-Year Window Since 1872

We backtested every real 30-year retirement window in 154 years of US market history. Here's exactly how often a 3%, 4%, 5%, or 6% withdrawal rate actually lasted the full 30 years - no simulation, no resampling, just real history.

If you retired in any year since 1872, held all US stocks, and pulled a fixed 4% of your starting balance every year (adjusted for inflation) for three decades, your money lasted the full 30 years 96.8% of the time. That is not a simulation or an average of made-up scenarios. It is the actual record of every real 30-year window that exists in 154 years of US market history.

This is the question the famous “4% rule” is really asking, and it deserves a real answer instead of a rounded rule of thumb. Below are two complete backtests: one across the longest US stock history we have, and one across a more realistic stock-and-bond blend. Every row is a count of real history, not a model.

What a “window” actually means here

A window is one real retirement, started in one real year. Pick a starting year, set a withdrawal rate, and spend that fixed percentage of your original balance every year for 30 years. Because we work in real (inflation-adjusted) dollars, that means the same purchasing power each year, not a percentage of whatever the balance happens to be at the time. The only question we ask is binary: did the money survive all 30 years without hitting zero?

Then we do that for every possible starting year in the data and count how many windows survived. This is a genuine backtest of undisturbed history - the 1929 crash sits where it really sat, the 1970s stagflation lands in its true order, every crash and recovery kept in sequence. It is not a Monte Carlo simulation that resamples or reshuffles years. (If you want the difference between this kind of every-real-window “cycle” test and a randomized simulation, we cover that separately.) The whole point of keeping history in order is that the sequence of returns matters enormously - a bad decade right at the start does far more damage than the same decade later, which is the single biggest risk a retirement faces.

All US stocks, every window since 1872

Our bundled US stock data (the Shiller series, using total return with dividends reinvested, in real terms) begins in 1872. That gives 154 years of history through 2025, and a 30-year window needs 30 years of runway, so there are 125 distinct starting years to test (1872 through 1996). Here is how each withdrawal rate held up.

Every real 30-year window in an all-US-equity portfolio, 125 starting years from 1872 to 1996, using real total return. A window "survives" if the inflation-adjusted balance never hit zero across the full 30 years. Historical backtest of actual starting points, not a forecast of future results.

Read down that table and the shape of the tradeoff is obvious. At 3.0% and 3.5%, nothing ever failed - every single one of the 125 windows made it, including retirements that started right into the worst downturns on record. Push to 4.0% and you lose exactly four windows: still a 96.8% survival rate, but no longer perfect. From there the failures pile up quickly. At 5.0% roughly one in five windows ran dry, and at 6.0% nearly a third did.

The four failures at 4% are the honest asterisk on the “4% rule.” A 30-year retirement has run out of money at a 4% rate in real history - it is uncommon, but it has happened, and it happened to people who retired just before a long, punishing stretch of markets. The rule is a strong starting point, not a guarantee.

A more realistic mix: 70% stocks, 30% bonds

Almost nobody actually holds 100% stocks through retirement. A 70/30 stock-and-bond split is closer to what a real retiree carries, so it is worth seeing the same backtest for that blend. There is one honest catch: our bundled bond data (the Damodaran series) only reaches back to 1928, not 1872. That is a real data-availability limit, not a choice - we simply do not have trustworthy total-return bond figures before then. So this second table covers 1928 to 2025, which is 98 years and 69 distinct 30-year windows (1928 through 1996). Still a substantial slice of history, just shorter than the all-equity run above.

Every real 30-year window in a 70% US-stock / 30% US-bond portfolio, 69 starting years from 1928 to 1996, using real total return. The window is shorter than the equity table above only because bundled bond data begins in 1928. Historical backtest, not a forecast.

The pattern rhymes with the all-equity table but is not identical. The safe floor is the same - 3.0% and 3.5% never failed across all 69 windows. At 4.0% the blend survived 94.2% of the time, a touch below the all-equity 96.8%, and it also lost exactly four windows. That is a useful thing to sit with: over this shorter, more recent history, adding bonds did not obviously make a 4% retirement safer by this pass/fail measure. Bonds trade some of stocks’ long-run growth for a smoother ride, and over a full 30 years that growth is what refills the account against inflation-adjusted withdrawals. What bonds buy you is a gentler path, not necessarily a higher survival count - a distinction the raw survival rate can hide.

Because the two tables cover different spans (154 years versus 98), do not read them as a clean head-to-head. The all-equity table includes the deep pre-1928 history that the blend simply cannot reach. Use the first for the longest possible record and the second for a portfolio shape closer to a real retiree’s.

What the numbers do and don’t tell you

Three things stand out across both tables. First, 3.5% and below has never failed in either dataset - that is the price of near-certainty, and the price is spending less. Second, 4% is genuinely strong but genuinely imperfect: mid-90s survival, with real failures you can point to. Third, the drop-off above 4.5% is steep, and by 5.5% to 6% you are into territory where a meaningful chunk of real history did not make it.

And the caveat that governs all of it: this is the past, not the future. US markets delivered an unusually good century and a half, and there is no rule saying the next 30 years must match it. A backtest tells you which starting points would have worked, not which ones will. Treat these survival rates as a well-documented record to reason from, not a prediction to bank on.

The strength of a rule like “4%” is that it is roughly right across a lot of history. Its weakness is that “roughly right” still left four real retirements short. The value of looking at every window, instead of one headline number, is seeing exactly where the rule held, where it cracked, and how much breathing room you buy by pulling a little less.

Frequently asked