Does 4% behave the same over 30, 40 and 50 years?
We tested three withdrawal rates and three US stock–bond mixes against the same 1928–2025 data. Here are the results, assumptions and every historical window.
The longer horizon changed the answer
With 60% stocks, 40% Treasuries, a 4% initial withdrawal and a 0.2% annual fee, this model funded every expense in 65 of 69 thirty-year windows. Over fifty years, it funded 33 of 49 windows.
The denominator changes because fewer complete long windows fit in the data. These are overlapping historical observations, not independent trials or forecasts. They show why a single “4%” label is not enough to describe a retirement plan.
Lowering the initial rate to 3.5% raised the fifty-year 60/40 result to 44 of 49 windows. Raising the stock allocation to 80% at a 4% rate raised it to 39 of 49. Neither change removes uncertainty, and more stocks also changes the path and volatility.
The complete comparison
All rows start with 1,000,000, no further saving or pension, and constant purchasing-power spending. A 4% initial rate is 40,000 per year, paid monthly. “Funded” means every planned withdrawal was met.
| Horizon | Stocks / bonds (%) | Initial rate | Windows funded | Observed fraction |
|---|---|---|---|---|
| 30 years | 60 / 40 | 3% | 69 / 69 | 100.0% |
| 30 years | 60 / 40 | 3.5% | 69 / 69 | 100.0% |
| 30 years | 60 / 40 | 4% | 65 / 69 | 94.2% |
| 30 years | 80 / 20 | 3% | 69 / 69 | 100.0% |
| 30 years | 80 / 20 | 3.5% | 69 / 69 | 100.0% |
| 30 years | 80 / 20 | 4% | 65 / 69 | 94.2% |
| 30 years | 100 / 0 | 3% | 69 / 69 | 100.0% |
| 30 years | 100 / 0 | 3.5% | 68 / 69 | 98.6% |
| 30 years | 100 / 0 | 4% | 64 / 69 | 92.8% |
| 40 years | 60 / 40 | 3% | 59 / 59 | 100.0% |
| 40 years | 60 / 40 | 3.5% | 57 / 59 | 96.6% |
| 40 years | 60 / 40 | 4% | 50 / 59 | 84.7% |
| 40 years | 80 / 20 | 3% | 59 / 59 | 100.0% |
| 40 years | 80 / 20 | 3.5% | 59 / 59 | 100.0% |
| 40 years | 80 / 20 | 4% | 53 / 59 | 89.8% |
| 40 years | 100 / 0 | 3% | 59 / 59 | 100.0% |
| 40 years | 100 / 0 | 3.5% | 58 / 59 | 98.3% |
| 40 years | 100 / 0 | 4% | 52 / 59 | 88.1% |
| 50 years | 60 / 40 | 3% | 49 / 49 | 100.0% |
| 50 years | 60 / 40 | 3.5% | 44 / 49 | 89.8% |
| 50 years | 60 / 40 | 4% | 33 / 49 | 67.3% |
| 50 years | 80 / 20 | 3% | 49 / 49 | 100.0% |
| 50 years | 80 / 20 | 3.5% | 47 / 49 | 95.9% |
| 50 years | 80 / 20 | 4% | 39 / 49 | 79.6% |
| 50 years | 100 / 0 | 3% | 49 / 49 | 100.0% |
| 50 years | 100 / 0 | 3.5% | 46 / 49 | 93.9% |
| 50 years | 100 / 0 | 4% | 42 / 49 | 85.7% |
Exactly what we tested
- US stock total returns, 10-year Treasury total returns and US inflation from the NYU dataset linked below.
- Each possible start year with a complete 30-, 40- or 50-year period ending no later than 2025.
- Asset weights reset at the beginning of each retirement year. Withdrawals are proportional across holdings.
- Annual asset returns are converted to equivalent constant monthly returns within that year. Each withdrawal is made before that month's return.
- A 0.2% annual portfolio fee is applied multiplicatively. Taxes and exchange-rate movements are excluded.
- The final withdrawal counts as funded when the required amount is available, including exactly zero remaining afterward.
Annual data cannot reproduce the actual order of monthly market moves. These results therefore differ from tests using observed monthly prices, beginning-of-year lump-sum withdrawals, different bond series or other rebalancing rules. They are results of FourPercent model 2.0.0, not a replication of the Trinity study or Bengen's original paper.
What the table does not say
A 100% entry means every available historical window in this dataset funded the modeled spending. It does not establish a safe rate for the next fifty years. The sample is one successful national market, long windows overlap heavily, future tax rules are unknown and the tested investor never changes behavior.
Increasing stocks did not improve every row. At 3.5% over fifty years, 80/20 funded 47 windows while 100% stocks funded 46. The comparison is specific to this dataset and withdrawal timing; it does not establish an optimal allocation.
Before translating any result to your life, account for pension timing, tax, spending flexibility and large irregular costs. The calculator lets you test those cash flows explicitly.
A pension bridge changes the cash flow
Consider a hypothetical 40-year-old with 300,000 invested, saving 1,500 monthly in today's money, retiring at 55 and spending 3,000 monthly through age 95. Assume a 7% nominal return, 2.5% inflation, 0.2% fee and no savings growth above inflation.
The following is a separate fixed-return example. It is not another historical success test. All amounts are in today's purchasing power.
| Scenario | At retirement | At age 95 | Fixed-return outcome |
|---|---|---|---|
| No pension | 926,871 | 1,119,949 | All spending funded |
| Pension 1800 from 67 | 926,871 | 2,254,882 | All spending funded |
| Pension plus 50000 expense at 60 | 926,871 | 2,045,164 | All spending funded |
The portfolio at retirement is identical because the changes occur later. Pension begins twelve years after retirement, so the portfolio must bridge those years first. The one-time expense reduces both capital and subsequent compound growth. A simple annual-spending multiple cannot show that timing.
Download and reproduce
Published by FourPercent. Analysis generated with our public calculation engine, with automated checks for withdrawal timing, inflation, fees, cash flows and exact depletion. This is original model analysis; it has not had an independent professional review.
- Every historical window and outcome (CSV)
- Full assumptions, comparisons and worked examples (JSON)
- Underlying annual market and inflation data (CSV)
- Calculation engine, version 2.0.0 (JavaScript)
- Reproduction script for Node.js
To reproduce locally, save the engine beside the reproduction script. Create a data folder beside them and save us-stock-returns.json there. Run node reproduce-research.cjs. It writes the comparison JSON and window CSV into that folder.
Underlying source: Aswath Damodaran, NYU Stern — historical returns on stocks, bonds and bills, retrieved 27 September 2026. Read all model conventions or report a correction.