The first ten years of retirement explain 77% of the withdrawal rate a portfolio can sustain over thirty. The last twenty years explain 5%. Both figures come out of the same simulation, run by Wade Pfau, and together they describe the problem most retirement plans handle worst.
The same lopsidedness shows up on the other side of the retirement date. In Pfau's model, the compounded return over a saver's first fifteen working years explains 6% of the pot they reach. The return over their final fifteen working years explains 65%.
Stitch those together and you get a window of roughly a decade either side of the retirement date that carries most of the lifetime risk. That isn't a metaphor. It's what happens when the balance is at its largest and the money still going in is at its smallest.
Why the exposure arrives so late
Compounding is back-loaded, and the risk riding on it is back-loaded too. Vanguard's How America Saves 2025 covers nearly five million defined contribution participants. At the end of 2024 the average balance for participants under 25 was $6,899. For those aged 35 to 44 it was $103,552. For 45 to 54, $188,643. For 55 to 64, $271,320.
Run a 10% market fall through those numbers. The under-25 participant is down $690. The 55-to-64 participant is down $27,132, which is close to four times the younger one's entire account. Neither behaved differently. One simply had more money exposed.
Fresh contributions can't close that gap. Vanguard puts the average total contribution rate, participant plus employer, at 12.0% of pay in 2024. Participants aged 55 to 64 deferred 9.3% of their own income, the highest rate of any working age band. Both figures are healthy. Neither is large beside a quarter-million-dollar balance moving 10% in a year.
British data is thinner but points the same way. The Office for National Statistics found that private pension wealth not yet in payment was most valuable in the 55-to-64 age band, with a median of £107,300 across April 2018 to March 2020. Wealth peaks just before it starts being spent.
What the regressions actually measure
Pfau's lifetime model is deliberately plain. Five hundred hypothetical individuals draw overlapping sixty-year return paths from one long simulated series with an arithmetic average of 7% and a standard deviation of 20%, roughly the historical behaviour of the S&P 500. The compound average works out at 5%. Everyone saves 15% of a constant real salary for thirty years, then withdraws for thirty more. Nothing else varies: no job losses, no health shocks, no behavioural mistakes.
At a fixed 5% return with no volatility, that saving rate produces a pot worth exactly ten times salary. Add volatility and the five hundred outcomes ranged from 2.6 times salary to 81.1 times, with a median of 11.2. Sustainable withdrawal rates over the following thirty years ranged from 1.6% to 20.7%, median 6.3%.
That spread is the finding. These people did the same thing for sixty years. What separated them was the order returns arrived in, and the regressions locate where the order bites. Pfau also isolated each individual year. The return in the first year of retirement, on its own, explains more than 14% of the eventual sustainable withdrawal rate.
Historical data lands in the same place. Using US returns for 55 rolling thirty-year retirements starting between 1926 and 1980, with a 5% real withdrawal, a 1% fee and a 60/40 portfolio, Pfau regressed the thirty-year result on what was left after each elapsed year. After five years, remaining wealth and cumulative inflation already explained about 50% of the outcome. After ten years, about 73% of the final wealth and about 80% of the sustainable withdrawal rate. Explanatory power passed 90% at roughly year eighteen, then weakened as more portfolios hit zero.
Sequence risk itself isn't news, and the mechanism has been laid out before in this analysis of how identical returns in a different order produce opposite outcomes. The narrower claim here is about distribution. The risk isn't spread evenly across a thirty-year retirement. It sits at the front, and the front can be measured.
The working half of the window
The decade before the retirement date is where this turns awkward, because that's precisely when most default funds cut equity exposure. Anup Basu and Michael Drew tested that head-on in the Journal of Portfolio Management in 2009.
They modelled a participant starting on a $25,000 salary with a 9% contribution rate, 4% salary growth and a 41-year career. Returns were resampled 10,000 times from US stock, bond and bill data covering 1900 to 2004. They built four conventional lifecycle strategies, all starting at 100% stocks and switching linearly into bonds and cash after twenty, twenty-five, thirty or thirty-five years. Then they paired each with a mirror image holding the same assets for the same number of years, in reverse order.
The mirrors won nearly everywhere. The strategy holding bonds and cash first and stocks last produced a median final pot of $1,425,387, against $1,160,225 for its lifecycle twin. That's a gap of $265,162 for identical time spent in each asset class. At the 75th percentile the gap widened to more than $700,000, or 41%. The margins for the other three pairs were $270,763, $176,531 and $121,584.
The accumulation paths explain why. For the first twenty years the two approaches were hard to tell apart. They only separated late. Early allocation barely moves the outcome because there's barely any money for it to move, and the authors are explicit that they aren't recommending the mirrored strategies. They built them as a fair test, holding time in each asset class constant so the comparison isolates when the risk was taken rather than how much.
The case against treating this decade as special
The strongest objection is that a fragile decade only matters if bad decades happen, and they hardly ever do. Javier Estrada argues exactly that. Using S&P 500 real returns from 1900 to 2019, annualising at 6.5% with 20.0% volatility, he ran 91 overlapping thirty-year retirements on a 4% inflation-adjusted withdrawal. Four failed. That's a 4.4% failure rate.
The four failures began in 1929, 1966, 1968 and 1969. Their annualised real returns over the first ten years were 0.7%, -2.4%, -2.6% and -3.4%, against 6.3% across all periods. So the early decade did the damage. Estrada then reshuffled each failing period's own thirty returns 10,000 times. The 1929 retiree's returns failed in 8.8% of the rearrangements, and the other three in 6.1%, 2.0% and 1.2%. In over nine cases in ten, the same returns in a different order would have been survivable.
He also checks whether the usual scare illustrations are plausible. A widely reproduced example implies annualised real returns of -25.4% over three years, -19.7% over five and -11.0% over ten. Nothing close has happened. The worst three-year run was 1929 to 1931 at -23.5%. The worst ten-year run was 1911 to 1920 at -4.4%, followed by 1965 to 1974 at -3.8%. His conclusion is that retirees should be informed about sequence risk without obsessing over it.
That's a fair hit on the rhetoric and a weaker one on the mechanism. A 4.4% failure rate is a base rate for one country, and the country with the best equity record in the sample. Evidence on how the 4% rule held up across seventeen countries shows far higher failure rates elsewhere. And the reshuffling result concedes what it's meant to weaken. If reordering the same returns converts failure into survival nine times out of ten, order is doing the work.
What de-risking costs
The counter-argument that really bites is about the remedy rather than the diagnosis. Cutting equity exposure into the fragile decade is expensive, and the cost is easy to miss because it shows up as an absence.
Basu and Drew's downside table makes it concrete. Lifecycle strategies did protect the worst outcomes, but only just, and only there. At the 10th percentile the lifecycle version beat its mirror by about 8%. At the 90th percentile, exactly as likely, the mirror was 55% ahead. At the 5th percentile the advantage from switching early was 19% for the earliest switcher and 2% for the latest. Around the 15th percentile the difference mostly disappeared, and by the 20th percentile the mirrors were ahead again.
Estrada found something comparable on the retirement side. Across 81 rolling thirty-year retirements from 1900 to 2009, starting with $1,000 and withdrawing $40 a year in real terms, a US portfolio held at 100% equities failed 3.7% of the time and left a mean bequest of $3,077. A static 60/40 failed 4.9% of the time and left $1,437. The all-equity version was safer on the failure measure and left more than twice as much behind. For the world market index the gap was wider: 6.2% failure and $2,367 for all-equity, against 16.0% and $1,062 for 60/40.
His glidepath work also disagrees with the best-known remedy. Pfau and Kitces argued for a U-shaped path, cutting equities into retirement then raising them again. Estrada's international evidence, covering nineteen countries over 110 years, points to the opposite shape, and he says so plainly: it is "the opposite of the U-shaped glidepath that Pfau and Kitces (2014) suggest." Two careful researchers, one question, contradictory answers. That's the honest state of the evidence on what to do about the fragile decade, as opposed to whether it exists.
The size of the drawdown is only half of the cost anyway. The other half is time, and the record on how long bear markets have taken to recover since 1968 matters more at 62 than at 32, because a retiree is selling into the recovery rather than buying through it.
What would change the conclusion
Three things, and the first is the biggest.
Rigid spending is baked into all of it. Pfau's regressions assume a fixed real withdrawal, and Estrada's failure counts do too. A retiree who cuts spending after a bad opening year interrupts the compounding of the damage, which is the entire point of the research on what a 5.2% starting withdrawal rate costs under spending guardrails. If spending flexes, the concentration figures overstate the danger. The exposure doesn't vanish, because the balance is still at its peak, but the link from a bad decade to a broken plan gets weaker.
Second, the 77% figure is model-dependent. It comes from Monte Carlo draws where each year's return is independent of the last. If real returns mean-revert even slightly, a poor first decade is partly repaid later and the concentration is overstated. The check on that is the historical version of the same test, which gave about 80% after ten years on actual US data since 1926. The two agree closely, so the independence assumption isn't carrying the result on its own. A different return regime, or a sample that isn't dominated by the twentieth-century United States, could still move it.
Third, the accumulation-side result assumes contributions stay modest relative to the balance. That holds for most people, and Vanguard's age bands show it clearly. It doesn't hold for a late, very high earner whose final-decade contributions are large next to what came before, or for someone whose pension wealth is mostly a defined benefit promise rather than a market-linked pot. For them the fragile decade is genuinely less fragile, because less of their retirement income is exposed to a price at a point in time.
What would settle the remaining dispute is evidence on glidepath shape that survives outside the US record. Estrada's nineteen-country test is the closest thing available, and it favours static allocations over clever paths. Until something better arrives, the defensible reading is that the concentration of risk in the fragile decade is well evidenced, while the right response to it isn't.