How much does sequence-of-returns risk matter in retirement?
Enormously, once withdrawals begin — the order of returns decides the outcome more than the average return does. We replayed all 819 real 30-year start months in the 1926-08 to 2024-09 monthly record through the QuantCalc planner’s own engine (its Historical Sequences mode) with the identical $1,000,000 portfolio, 60/40 allocation and $40,000 (4%) a year withdrawal. The plan ran out of money in 27.16% of the start months whose first decade was in the worst 10% (22 of 81) and in 0.00% of the best 10%, against 2.69% of all 819 (22 of 819). Median balances after 30 years were $108,028 versus $4,225,894. The retirement that began in October 1964, the worst first decade in the record, was down to $65,049 after 30 years; the same months in reverse order left $1,274,051. Spending rises with the inflation that actually happened each month (the BLS consumer price index), as the planner replays it — see inflation.
Key numbers — each row links to the section that carries the full table. Updated 2026-10-15: every figure now comes from the planner’s engine (what changed).
| Key number | Value | Where on this page |
|---|---|---|
| Failure rate, worst 10% of first decades (1st-decade real CAGR -1.95%) | 27.16% | Full results |
| Failure rate, best 10% of first decades (1st-decade real CAGR 10.99%) | 0.00% | Full results |
| Failure rate, all 819 start months | 2.69% | Full results |
| Median balance after 30 years, worst vs best 10% | $108,028 vs $4,225,894 | Full results |
| Worst first decade (October 1964): forward vs the same months reversed | $65,049 left vs $1,274,051 | Same months, two retirements |
| Failure rate on a constant 2% inflation assumption instead of realized inflation (worst 10% / all) | 17.28% / 1.71% | Inflation |
| Simulated retirements | 819 real start months, 30-year horizon | Methodology |
“Sequence-of-returns risk” is one of the most repeated phrases in retirement planning and one of the least quantified. The idea is simple: if you are withdrawing money, a bad run of markets in your first years does lasting damage that a good run later can’t undo — because you sold assets cheap to fund spending and there’s less left to recover. We measured it with the plan held fixed: identical $1,000,000 portfolios, identical 4%-of-initial withdrawal, identical 60/40 allocation. The only difference is the month the retirement began. Every figure below comes from the QuantCalc planner’s own engine: run this plan before tax in the planner’s Historical Sequences mode and it reports 97.3% success, the complement of the 2.69% failure rate below.
Headline finding: The identical plan — $1,000,000, 60/40, $40,000 a year raised with inflation for 30 years — ran out of money in 22 of the 81 start months whose first decade was in the worst 10% (27.2%) and left a median of $108,028. In the best 10% it never ran out and left a median of $4,225,894. Across all 819 start months it failed 22 times, and every one of those failures began between February 1965 and February 1969: the risk is not spread across history, it is concentrated in the starts that met a crash first.
The same months, two completely different retirements
The cleanest way to see sequence risk is to remove averages from the argument entirely. The worst first decade in the record, after inflation, belongs to the retirement that began in October 1964 (-3.3% a year real over its first ten years). We ran its 360 months two ways through the same arithmetic the engine uses: forward, and reversed (the returns reversed, the month-by-month inflation kept in its order). Both orders contain the identical monthly returns, so they have the identical average: 4.4% a year arithmetic and 3.8% geometric, after inflation.
Forward, the bad decade comes first and the $40,000-a-year plan was down to $65,049 after 30 years. Reversed, the same bad months arrive at the end, after the portfolio has grown, and the plan finishes with $1,274,051. Same returns, same average, same withdrawals. For comparison, the best first decade in the record, the retirement that began in August 1982, finished with $6,962,035.
Average return is the wrong number for a retiree. While you are accumulating, order doesn’t matter — a dollar is a dollar no matter when it grows. The moment you start withdrawing, order becomes everything: bad early years force you to sell more shares to fund the same spending, permanently shrinking the base.
Headline chart: failure rate by what your first decade looked like
30-year failure rate of an identical 4% plan, by first-decade market luck
All 819 real 30-year start months (August 1926 to October 1994, the planner’s Historical Sequences mode), ranked by first-decade real portfolio return and split into tenths. Same $1,000,000 start, same $40,000 a year. Bars show the share of each group that ran out of money within 30 years.
The pooled failure rate of 2.7% is real, but it is a blend of two nearly disjoint worlds. All of the failures sit among the unlucky starters; the “average” retiree barely exists.
Full results: the three groups
The worst and best tenths are 81 start months each. “1st-decade real CAGR” is the portfolio’s annualized return over the first ten years after the inflation that actually happened. Balances are in today’s dollars, deflated by each start month’s own inflation.
| Group | 1st-decade real CAGR | Failure rate | Ran out | Median after 30 yrs | 10th pct | 90th pct |
|---|---|---|---|---|---|---|
| Worst 10% of first decades | -1.95% | 27.16% | 22 of 81 | $108,028 | $0 | $835,884 |
| All 819 start months | 5.34% | 2.69% | 22 of 819 | $1,497,942 | $353,033 | $4,125,434 |
| Best 10% of first decades | 10.99% | 0.00% | 0 of 81 | $4,225,894 | $3,006,454 | $5,964,055 |
The worst tenth is made of start months in 1964, 1965, 1966, 1967, 1968, 1969, 1970, 1971, 1972, 1973 — first decades that ran through the inflation and bear markets of the 1970s. Three things stand out:
- The first decade decides the outcome. A first decade near -1.9% a year real is enough to push the failure rate to 27.2%; in the tenth that averaged 11.0%, no start month ran out.
- The unlucky end of the good starts beats the lucky end of the bad starts. The 10th percentile of the best tenth ($3,006,454) is close to the 90th percentile of the worst tenth ($835,884).
- The failures are one era. All 22 began between February 1965 and February 1969.
A second lens: the planner’s bootstrap model
The planner’s other history-based model resamples blocks of monthly returns instead of replaying them in order. It runs on a shorter panel, July 1990 to May 2026, because it also needs international and real-estate returns, which do not reach back to 1926. On 50,000 of its paths (25 per engine run, seeds 1 to 2000) the same plan shows the same shape, at lower levels because that window has no 1929 and no 1970s:
| Group | 1st-decade real CAGR | Failure rate | Ran out | Median after 30 yrs | 10th pct | 90th pct |
|---|---|---|---|---|---|---|
| Worst 10% of first decades | 1.47% | 4.12% | 206 of 5,000 | $1,115,020 | $235,088 | $2,565,544 |
| All 50,000 paths | 6.96% | 0.42% | 208 of 50,000 | $3,557,367 | $1,221,370 | $8,017,394 |
| Best 10% of first decades | 12.22% | 0.00% | 0 of 5,000 | $7,410,520 | $3,988,885 | $13,197,597 |
CC-BY-4.0 — free for any use including republication and journalism, with attribution to QuantCalc Research.
Inflation: what actually happened, not an assumption
The planner replays each historical month’s returns in order and, with them, that month’s inflation: the withdrawal rises month by month with the BLS consumer price index (CPI-U, U.S. city average), and every balance on this page is deflated by the same index (methodology). That matters most in the two eras that dominate this study: from January 1973 to January 1982 consumer prices rose 9.2% a year, while from January 1929 to January 1933 they fell 6.8% a year (January CPI in the Shiller dataset). Inflation from January 1926 to January 2023 averaged 2.9% a year.
On a constant 2% assumption instead (the planner’s convention for this mode before 2026-10-15, still available through the API as historicalInflation: "fixed"), the same 819 start months fail 1.71% of the time (14 of them) and the worst tenth 17.28% (14 of 81), and the worst first decade moves to September 1929: a flat rate makes the 1970s starts look easier and the Depression starts harder than they were. On realized inflation the hardest starts are the late 1960s, as in our cohort study, which finds 1966 the hardest year to retire into on a 60/40 portfolio.
What this does and doesn’t say
This is not an argument that the 4% rule is broken, nor that retirement is a coin flip you can’t influence. It measures how much of the outcome is decided by something you don’t control — the start month — when the plan is held rigid.
- A rigid plan is the worst case. Spending stays at $40,000 a year in today’s dollars and is never cut. Retirees who trim spending in down years, hold a cash buffer or use a bond tent blunt sequence risk; this study measures the exposure those defenses insure against.
- History is one path, overlapped 819 ways. Thirty-year windows that start a month apart share almost all their months, so the 819 start months are a handful of independent histories, not 819 draws. The planner says the same about its Historical Sequences mode.
- Realized history ran ahead of today’s forecasts. US stocks compounded about 10.2% a year nominal across this panel, well above current capital-market forecasts (methodology), so these failure rates are lower than a forecast-based run of the same plan. Our forward-looking 4% study covers that side.
Methodology
Engine: the QuantCalc planner’s C engine, POST /api/simulate with returnModel: "historical" (study engine sequence-risk-engine-2.0.0, run 2026-10-15). The API returns 30 sample paths per run, so the study code replays the engine’s historical model for each of the 819 start months and proves the replay equal to the engine before using it: the success rate, every reported percentile, the mean and every year of every sample path, across six run sizes (7,272 comparisons, zero differences). The study gate re-runs that proof.
Data: the planner’s August 1926–September 2024 monthly panel: US equity (market return including the risk-free rate) and one-month Treasury bills from the Ken French Data Library, and a par-bond total return built from Shiller’s long-term yield. Details in the planner’s methodology.
Plan: $1,000,000; 60% US stocks and 40% bonds, rebalanced monthly; $40,000 a year withdrawn monthly from the first month and raised every month with the start month’s own realized CPI-U; no other income; 30 years. A start month “fails” if the balance reaches zero within the 30 years. Balances and returns are reported in today’s dollars, deflated by that same index.
Groups: start months are ranked by their first-decade real portfolio return (ties by date) and split into tenths of 81.
Bootstrap lens: returnModel: "bootstrap", the planner’s stationary block bootstrap (mean block 24 months) of its July 1990 to May 2026 panel, pseudo-random seeds 1–2000, 25 paths each. Each seed runs twice on the same draws: with no withdrawals (the first-decade return) and with the plan (the outcome). Every path is the engine’s own. The bootstrap model has no month-by-month inflation record, so this lens raises spending and deflates by the plan’s 2% assumption.
What this is not: not personalized advice, and not a model of adaptive spending, cash buffers, glide paths or taxes.
Changelog
- 2026-10-15 — rebuilt on the planner’s engine. The first edition ran its own simulation (5-year blocks of annual 1928–2024 returns resampled 50,000 times, withdrawals at the start of each year, realized inflation) and reported 45.58% failure for the worst decile, 0.00% for the best and 6.02% overall, with medians of $49,522 and $5,355,467. The planner could reproduce none of them. The new figures come from the planner’s engine, on the inflation that actually happened, as before. They differ in withdrawal timing (monthly) and in the data (real months replayed in order instead of resampled annual blocks). The first edition’s numbers are kept in summary.json.
- 2026-06-02 — first edition.
Is your plan exposed to a bad first decade?
Run your own plan through the same engine: the planner’s Historical Sequences mode replays every start month in this study, and the stress test shows forecast-based paths.
Stress-Test Your Retirement Plan →Related research
- → Does the 4% Rule Survive Forward-Looking Forecasts? — the companion study on how lower expected returns (not just their order) reshape safe-withdrawal math.
- → The 4% rule across historical retirement cohorts — the same question year by year since 1871, on realized inflation.
- → 510,000-path 51-State Retirement Monte Carlo — the same couple in every state; the median 30-year income tax and IRMAA runs from $162,597 to $309,718 (Oregon).
- → Social Security claiming age under uncertainty — another case where a single point estimate hides a full distribution of outcomes.
Frequently asked questions
Last updated 2026-10-15. Dataset license: CC-BY-4.0. Return data: the planner’s monthly panel from the Ken French Data Library and Shiller’s data. QuantCalc is an independent retirement-planning research project. Not financial, tax, or legal advice.