Retirement Drawdown

Spending stays constant in purchasing power. Returns are real: adjusted for inflation. For example, a nominal 8% return with 3% inflation is (1.08 / 1.03) − 1 = 4.85% real. Inflation is already accounted for; no separate inflation assumption is applied.

Withdrawals happen at the beginning of each whole year, followed by that year's return. The first year that cannot fund the full planned withdrawal is a failure; any available balance funds a partial withdrawal. A funded final withdrawal counts as success even if the ending balance is zero. Zero spending is fundable with zero assets. Taxes, fees, pensions, and contributions are excluded.

The order of returns matters when spending: early losses can force you to sell more assets and leave less to recover. Constant return shows one smooth assumption and hides this sequence risk.

Historical markets tests every full contiguous window in the checked-in S&P 500 total-return series, including dividends and pre-1957 large-cap backfill. Nominal stock returns are adjusted by that year's observed CPI inflation. Windows overlap and are not independent. This is one US stock benchmark, not a diversified portfolio or a prediction. No years are wrapped or padded.

Monte Carlo uses independent annual lognormal gross real returns. Expected return means the arithmetic mean of simple annual returns; volatility is their standard deviation. The fitted log variance is ln(1 + (SD / (1 + mean))²), and log mean is ln(1 + mean) minus half the log variance. Losses cannot exceed 100%; no individual return is clipped. Finite seeded pseudorandom uniforms feed Box–Muller normal shocks. This simplified independent-year model omits changing regimes and may understate market risks.

Scenario A and B use the same historical windows or the same trial/year normal shocks. Edits keep the seed; Reshuffle explicitly changes it. At zero volatility, Monte Carlo matches the constant-return calculation at the same return.

Success fractions describe these tested historical windows or simulated trials, conditional on your assumptions. They are not guaranteed probabilities of future success. All percentiles include unsuccessful paths. The chart's median and 10th–90th percentile band are pointwise summaries, not individual market paths. Values between percentile observations use linear interpolation.

Historical dataset: damodaran-1928-2025-v1, 98 annual observations, 1928–2025. Source: Aswath Damodaran, NYU Stern, historical US returns workbook, retrieved November 4, 2026. Uses Returns by year stock total returns and observed inflation, cross-checked against the workbook's Inflation Rate sheet (CPIAUCNS, all urban consumers, not seasonally adjusted). Real return = (1 + nominal) / (1 + inflation) − 1.

NYU source and methodology ↗ · CPI series ↗

Model annual-lognormal-v1. Bounds: 1–100 years (historical up to 98), 100–2,000 trials, assets and spending $0–$1 billion; constant returns −100%–100%; Monte Carlo means −50%–50% and volatility 0%–50%. This is an illustrative model, not a forecast or financial advice.