How this backtest works
Most retirement calculators assume one flat average return for every year of the plan. That average can be exactly right and still miss the thing that actually breaks retirements: the order those returns arrive in. A portfolio that loses 30% in year one of retirement, while withdrawals are also draining it, ends up in a very different place than one that loses 30% in year twenty-five, even if both retirements average the same long-run return. This is sequence-of-returns risk, and averages can't see it.
This calculator addresses that directly by replaying real history. Every phase before retirement — accumulation, career breaks, one-time lump sums like a home down payment or inheritance — runs as a simple deterministic straight-line projection, producing a single ending balance labeled the projection. That balance then becomes the starting point for the terminal retirement phase, which is tested against every historical starting year with enough subsequent data to cover the full drawdown length: 1928, 1929, 1930, and so on through the most recent completed year. Each of those replays uses that specific year's actual sequential stock and bond returns and actual inflation. The percentage of those replays that never ran out of money is the historical success rate.
Why only the terminal phase is backtested
Sequence-of-returns risk is sharpest exactly when you're withdrawing money and exposed to markets at the same time — which is what retirement drawdown is. During accumulation, you're adding money each year, so a bad early year is far less damaging and tends to average out over decades; a deterministic projection is a reasonable simplification there. Backtesting a non-terminal drawdown — say, an early retirement followed by a return to work — would require threading one continuous historical path across phases with different contribution and withdrawal rules at each step, a meaningfully more complex problem than testing one locked terminal phase. That's a possible future extension, not needed for the common case this calculator targets: any number of upstream phases followed by one final, permanent retirement.
The data behind the replay
Annual U.S. stock returns (S&P 500 including dividends) and 10-year Treasury bond returns come from Aswath Damodaran's historical returns dataset at NYU Stern, spanning 1928 to the most recently completed year. Annual inflation is the Bureau of Labor Statistics' CPI-U annual average. Your withdrawal is entered in today's dollars; since upstream growth is nominal, it is first grown at your assumed upstream inflation rate to express it in dollars at retirement start, then within each cycle it is inflation-adjusted using that cycle's own actual historical inflation path, not a flat assumed rate — so a cycle starting in 1973 sees withdrawals grow the way they actually would have during that decade's inflation.
Reading the chart and the percentiles
The single amber line is the deterministic upstream projection. Once the terminal phase begins, thin lines fan out — one per historical starting year, each tracing that cycle's actual balance path. The bold overlay lines mark the 10th, 50th (median), and 90th percentile ending balances across all cycles. A wide gap between the 10th and 90th percentile is the visual signature of sequence-of-returns risk: the same withdrawal plan can end in wildly different places purely because of which years the retirement happened to start in.
Why the results are shown in today's dollars
The simulation itself runs entirely in nominal terms — nominal stock and bond returns, with your withdrawal rising each year by that cycle's own actual CPI. That is the correct way to run it, and it is how the success rate is determined. But displaying the results in nominal dollars would be misleading, for a reason specific to historical-cycle backtesting.
Each cycle ends in a different calendar year, having lived through a different inflation history. Over a 35-year drawdown, cumulative inflation ranges from about 1.7× for a cycle starting in 1928 to 5.5× for one starting in 1966 — a threefold difference in what a dollar means at the finish line. Sorting nominal ending balances across those cycles and reading off a median is sorting quantities measured in different units.
It also produces false rankings. Take two cycles that both survived: a 1928 start ends at roughly $6.8M nominal, a 1941 start at $9.2M. The nominal figures say 1941 was the better retirement. Converted to constant purchasing power, 1928 ends around $1.57M and 1941 around $0.96M — 1928 was better by more than half again. Roughly three quarters of all cycles change rank when you deflate them.
So every ending balance here is deflated by its own cycle's actual CPI and then back through your assumed upstream inflation rate, putting it in the same dollars you typed the withdrawal in. The success rate is unaffected — whether a portfolio hits zero is a nominal-versus-nominal question inside a single cycle, and the answer is identical either way. This also matches how FIRECalc and cFIREsim report: both present results in constant dollars rather than end-year nominal.
The one figure on this page that remains nominal is the upstream projection — the balance entering retirement — because it is a single deterministic path with no cross-cycle comparison problem, and it is the number you would actually see on a statement. Its today's-dollars equivalent is stated directly beneath it so the handoff between the two sections is explicit.
Frequently asked questions
What is historical-cycle backtesting?
It replays the actual sequence of historical stock and bond returns starting in each historical year, rather than assuming one average return, to see whether your withdrawals would have depleted the portfolio. It's the same method behind the Trinity study and the "4% rule."
Why does the sequence of returns matter more than the average?
A bad market year early in retirement, while you're withdrawing, permanently shrinks the base left to recover. The same bad year late in retirement barely matters. Averages can't distinguish between these two very different outcomes.
Where does the historical data come from?
Stock and bond returns are from Aswath Damodaran's dataset at NYU Stern (1928-present); inflation is the BLS CPI-U annual average. Both are public data series.
Why is only the retirement phase backtested?
Sequence risk is sharpest when withdrawing and exposed to markets simultaneously — specifically retirement. Earlier accumulation phases stay deterministic since a bad early accumulation year is far less damaging and tends to average out.
What counts as a successful historical cycle?
The portfolio balance never hits zero across the entire drawdown length using that year's actual returns and inflation. The historical success rate is the percentage of all testable years that succeeded.
What do the 10th/50th/90th percentile balances mean?
Sorting every historical cycle's ending balance, the 10th percentile is a pessimistic outcome, the 50th (median) is typical, and the 90th is an optimistic one. The spread between them matters more than any single number. All three are in today's dollars.
Are the results in today's dollars or future dollars?
Today's dollars — the same purchasing power as the withdrawal you entered. The simulation runs in nominal terms, then each cycle's ending balance is deflated by that cycle's own actual CPI and by your assumed upstream inflation. This is necessary rather than cosmetic: cycles end in different years with 1.7× to 5.5× cumulative inflation, so nominal results aren't comparable across cycles. The success rate is identical either way. The upstream projection is the one figure left nominal, with its today's-dollars equivalent shown beneath it.
How are withdrawals inflation-adjusted?
Your entered withdrawal is in today's dollars. Because upstream growth is nominal, it is first grown at your assumed upstream inflation rate to dollars at retirement start (that becomes the year-one withdrawal), then each later year within a cycle increases it by that specific cycle's actual historical inflation, not a flat assumed rate.
What are stock allocation and fee drag?
Stock allocation splits the retirement portfolio between stocks and bonds for blending each year's historical return. Fee drag is an annual percentage — fund expenses and advisory fees — subtracted from that blended return every year.
Can I model a down payment, career break, or inheritance?
Yes — add any number of growth phases and lump-sum events in any order before retirement. These upstream phases are deterministic straight-line, not historically backtested.
Why is the retirement phase locked to the end?
Backtesting a non-terminal drawdown (e.g. early retirement then back to work) requires threading a continuous historical path across differing phase rules — significantly more complex, and deferred as a future extension rather than part of this scope.
Is this free?
Yes, entirely free with no signup, like every calculator on this site.
What are this backtest's limitations?
Only U.S. market history (1928-present) is used; only 98 overlapping starting years exist, so long drawdowns have fewer independent cycles; taxes, Social Security, RMDs, and changing spending needs aren't modeled; and past returns never guarantee future ones.