How the numbers are made

The data

Stocks are the S&P 500 including dividends, 1928–2025, from Aswath Damodaran's dataset at NYU Stern.

Bonds are 75% constant-maturity 10-year US Treasury and 25% three-month T-bills, from the same source. The blend is not decorative: this site models the bond side as Vanguard Total Bond Market, which carries about six years of duration, while a 10-year Treasury alone carries closer to eight. Using the 10-year by itself would overstate the damage rising rates do to the bond sleeve, and that lands squarely on the 1966–1981 cohorts — the exact ones every withdrawal-rate argument turns on.

Inflation is CPI-U, December over December, from FRED. December-to- December rather than annual averages because the return series is measured year-end to year-end, and using an annual average would misalign the deflator by six months in exactly the high-inflation years that matter most.

Life expectancy is the SSA period life table, sexes averaged. The plan runs to your remaining life expectancy at your current age plus five years. That conditional bit matters: US life expectancy at birth is about 78, but a person who has already reached 65 averages about 84, because the at-birth figure is dragged down by deaths they have already survived past.

The simulation

The cone is ten thousand paths drawn by a stationary block bootstrap. Each path walks forward through the historical record one year at a time and occasionally jumps to a random new point, with the jump probability set so runs average eight years.

The blocks are the point. Drawing single years independently — the common shortcut — shuffles 1930, 1931 and 1932 away from each other, and retirements are destroyed by runs of bad years, not by their average. Eight years is long enough to keep a bear market next to the inflation that made it worse.

Alongside that, the app runs every complete historical start year: your actual plan, through the actual sequence of returns from 1928 onward, in order. Only start years with a full horizon of real data are shown. A 40-year plan starting in 2010 would require inventing thirty years, so it isn't offered.

Everything runs in your browser, not on the server. That is partly speed — dragging the allocation slider re-runs ten thousand paths with no round trip — and partly that your balance never has to leave the machine to be simulated.

What "you could be spending" means

The headline figure on your plan is the highest first-year spending this portfolio supports. It is not a rate anyone published, and no rate is assumed anywhere in producing it. A spending level is guessed, run through 4,000 simulated futures of exactly your horizon, and kept if it passes two tests. The highest level that passes is the number you see.

1. No more than 5% of those futures run out of money.
2. In no more than 10% of them does average spending across the whole retirement fall below 70% of where it started.

The second test is the one doing most of the work, and it measures a lifetime average rather than a worst single year. The obvious metric — the deepest dip spending ever takes — is a maximum taken over decades, so it saturates: on a 27-year horizon even a very safe plan dips 30% somewhere, because the inflation rule skips a raise after any losing year. Searching against it returned an absurd 2.3% and made flexible spending look worse than frozen spending. A dip that recovers is not an unaffordable retirement. A decade below plan is.

Why it falls as the horizon lengthens

Nothing is deducted for a longer plan. The horizon is an input to the simulation, not a penalty applied afterwards — the figure drops because a longer retirement has more chances to begin on a bad run of years, and because less of the money's life is spent in the recovery that follows. It is the same mechanism that makes every published safe-withdrawal rate fall with duration.

Which is also why this figure is not the 4% rule and should not be compared to it without checking the horizon. The 4% rule is a single number calibrated to the worst thirty-year stretch in the record. Run this model over thirty years and it lands near 4%. Run it over forty-four and it lands nearer 3.2% — not because it disagrees, but because it was asked a different question.

Holding spending flat is the 4% rule's assumption, not its number. Adopting the assumption does not get you the number.

The part that is a judgment call

Those two thresholds — 5% and 70%-at-the-tenth-percentile — are not facts. They are where this site decided the line falls between a plan worth having and one that only survives on paper, and they set the level of every figure it reports. Loosen them and the number rises; tighten them and you arrive back at the conservatism this whole site argues against, by a different route.

They are deliberately asymmetric. Running out of money gets the tighter bar, because spending less is a worse life while running out is no options at all, at the age you can do least about it.

Where this model is wrong

These are the ones worth knowing about before you act on anything here.

• No taxes. The model spends pre-tax dollars from a single undifferentiated pot. A real plan with a taxable brokerage, a traditional IRA and a Roth has a withdrawal-order problem this does not touch, and the difference is not small.
• No Social Security, pensions, annuities, or part-time work. Every one of those is a floor under your spending, and a floor changes the calculus enormously — it is what makes a deep guardrail cut survivable rather than catastrophic. Excluding them makes this model conservative.
• No house, and no long-term care. Two of the largest items on most real balance sheets, one an asset and one a liability, both absent.
• US-only, and only two asset classes. No international equity, no small-cap or value tilt, no REITs, no TIPS. Broader portfolios have historically supported somewhat higher withdrawal rates, so this omission is also conservative.
• Bonds are all Treasury. Total Bond Market is about a quarter corporate credit. The proxy omits some yield and some crisis correlation.
• Spending is one flat line. Real retirement spending tends to fall in the late 70s and 80s before medical costs push it back up. Modeling it flat overstates late-life spending needs, which — again — is conservative.
• The past is a sample, not a guarantee. A century of US data is roughly three independent 30-year retirements, from the winning economy of that century. It is the best evidence available and it is thinner than it looks.

Note which direction most of those point. This model leaves out several things that would raise your sustainable spending and one large thing — taxes — that would lower it. It is not the neutral machine it looks like, and neither is any other one.

Chart color

The uncertainty cone is drawn in amber rather than red on purpose. Coloring the downside red teaches the reader that the wide part of the cone is a hazard, when it is just the honest width of a future nobody knows yet. Red is reserved here for the handful of places something has genuinely gone wrong. The categorical colors used for the named historical cohorts were checked for color-vision separation and contrast in both light and dark themes rather than picked by eye.