Sequence Risk: Same Returns, Opposite Outcomes

11 min read

Key takeaways

  • Take the 30 real annual returns of a US 60/40 portfolio for 1993-2022, put $1,000,000 in, add nothing and take nothing out: the original order and the reversed order both end at exactly $5,116,739. Order changed nothing.
  • Same 30 returns, same two orders, but $40,000 a year withdrawn in today's money: the original order ends at $2,413,717 and the reversed order at $1,520,100 — a gap of $893,617, or 37%, on an identical 6.13% arithmetic average return.
  • The effect runs both ways. A saver putting in $10,000 a year for the same 30 years ends with $675,755 in the original order and $899,160 reversed — 33% more from the sequence that hurt the retiree.
  • Javier Estrada reshuffled the 1966-1995 returns 10,000 times and found that only 6.1% of the reorderings failed a 4% withdrawal plan. The order history actually delivered was unusually hostile, not typical.
  • Morningstar's December 2025 research puts the safe starting withdrawal rate for a rigid inflation-adjusted plan at 3.9%, and at 5.7% under its constant-percentage and endowment methods, which reset spending off the portfolio balance each year.

Order is irrelevant to money left alone, and decisive the moment money moves

Two people retire on the same day with $1,000,000 each. Over the next 30 years they get exactly the same 30 annual returns — the same numbers, the same average, the same best year and worst year. One receives them in the order history delivered. The other receives them backwards. How do they end up anywhere but the same place?

If neither of them touches the portfolio, they don't. Multiplying a starting balance by 30 growth factors gives the same answer whatever order you multiply in, and that is not an approximation — it is arithmetic. A lump sum left alone is completely indifferent to sequence.

Everything changes when money is moving. A withdrawal taken during a fall sells units that never come back, so the recovery applies to a smaller base. A contribution made during a fall buys units cheaply, so the recovery applies to a larger base. The returns are identical either way; what differs is how much capital was present to receive each one. That is sequence risk, and it is a property of cash flows, not of markets.

The sequence: 30 years of real 60/40 returns, run forwards and backwards

The return series is built from Robert Shiller's Irrational Exuberance dataset, the monthly file he has maintained since 1871. Two columns do the work: the real total-return index for the S&P Composite, and his real total-return series for 10-year US Treasuries. Blending them 60/40 and rebalancing every December produces 30 annual returns for the calendar years 1993 through 2022.

Three things about those returns need stating, because most illustrations of this idea skip them. They are real — already adjusted for inflation using the BLS consumer price index, which Shiller's file carries. They are total return, so dividends and coupons are reinvested, not price-only. And they are measured December to December, from monthly figures that are averages of daily closes rather than month-end prices, which is how Shiller's series is constructed.

The window is the most recent complete run of 30 calendar years in the file, picked by rule rather than by outcome. It isn't the worst 30 years in the record, and choosing it deliberately would have stacked the demonstration.

The arithmetic average of those 30 real returns is 6.13% a year; the compound average is 5.59%. Over the same window, US consumer prices roughly doubled — the CPI index rose from 141.9 in December 1992 to 296.797 in December 2022, a factor of 2.09, or 2.49% a year. Working in real terms means a withdrawal of $40,000 keeps the same purchasing power throughout; in cash terms it would have risen to about $83,700 by the end of the window.

With no withdrawals, both orders end at exactly $5,116,739

Start with $1,000,000, contribute nothing, withdraw nothing, and let the 30 returns run. In the original order the portfolio ends at $5,116,738.84. Reversed, it ends at $5,116,738.84. Not close — identical, to the last cent the arithmetic can carry.

This is where a great deal of retail writing on the subject goes wrong. Sequence risk is routinely described as though bad early returns are inherently more damaging than bad late ones. For an untouched lump sum that claim is simply false. The 2008 crash and the 2022 drawdown do the same damage whether they arrive in year 1 or year 30. Anyone still accumulating who hasn't started drawing down owns a sequence-neutral asset, at least with respect to the money already invested.

The chart shows those two identical bars beside what happens next.

Add a $40,000 withdrawal and the same returns end $893,617 apart

Now give each retiree a fixed real withdrawal of $40,000, taken at the start of every year — 4% of the starting balance, held constant in purchasing power, which is the classic construction and also the most rigid. Both take out $1,200,000 in total, in today's money. Nothing else differs but the order.

The retiree who got the original 1993-2022 order ends with $2,413,717. The retiree who got it backwards ends with $1,520,100. Same returns. Same average. Same withdrawals. A gap of $893,617, which is 37% of the luckier outcome and more than 22 years of spending at that rate.

Retiree A and Retiree B: $1,000,000 start, $40,000 real withdrawn at the start of each year, US 60/40 real total returns for 1993-2022 in original and reversed order. Balances are end-of-year, in constant purchasing power.
YearA: return yearA: real returnA: balanceB: return yearB: real returnB: balance
11993+8.64%$1,042,9582022-20.15%$766,582
21994-5.27%$950,1392021+7.91%$784,044
31995+29.47%$1,178,3542020+13.52%$844,635
41996+11.22%$1,266,0562019+17.51%$945,559
51997+21.19%$1,485,8332018-3.30%$875,636
61998+19.47%$1,727,3532017+11.46%$931,385
71999+7.36%$1,811,4642016+4.68%$933,080
82000-0.89%$1,755,7402015+1.21%$903,929
92001-6.62%$1,602,1092014+12.26%$969,841
102002-8.72%$1,425,9082013+12.96%$1,050,357
112003+12.02%$1,552,5192012+9.86%$1,109,933
122004+6.08%$1,604,4602011+4.25%$1,115,445
132005+1.71%$1,591,1452010+9.09%$1,173,224
142006+7.47%$1,667,0802009+12.28%$1,272,407
152007+3.04%$1,676,4912008-15.86%$1,036,930
162008-15.86%$1,376,9212007+3.04%$1,027,205
172009+12.28%$1,501,1222006+7.47%$1,060,990
182010+9.09%$1,593,9662005+1.71%$1,038,405
192011+4.25%$1,620,0682004+6.08%$1,059,095
202012+9.86%$1,735,7922003+12.02%$1,141,608
212013+12.96%$1,915,5822002-8.72%$1,005,558
222014+12.26%$2,105,5152001-6.62%$901,611
232015+1.21%$2,090,6092000-0.89%$853,963
242016+4.68%$2,146,5261999+7.36%$873,833
252017+11.46%$2,347,8971998+19.47%$996,190
262018-3.30%$2,231,6341997+21.19%$1,158,787
272019+17.51%$2,575,4781996+11.22%$1,244,294
282020+13.52%$2,878,2621995+29.47%$1,559,195
292021+7.91%$3,062,7281994-5.27%$1,439,190
302022-20.15%$2,413,7171993+8.64%$1,520,100

The interesting detail is that neither retiree ever came close to ruin. Even the unlucky one finished with 1.5 times the real capital they started with, having spent $1.2m along the way. This 30-year window was kind to both of them.

The mechanism is the withdrawal rate nobody chose

Look at what happens in the first year alone. The original-order retiree got 1993's +8.64% and ended the year with $1,042,958. The reversed-order retiree got 2022's -20.15% and ended with $766,582.

They then each take the same $40,000. For the first retiree that is 3.84% of what is left. For the second it is 5.22%. Neither of them decided to raise their withdrawal rate by 1.4 percentage points; the market decided it for one of them. And because the plan is fixed in real terms, that elevated rate persists until returns rebuild the base.

Compounding does the rest. Across the first five years the original order delivered +79.6% cumulative real growth; the reversed order delivered +11.2%. The identical 25 years that follow are then applied to two very different amounts of surviving capital. Nothing about the second retiree's investments was worse. There was simply less left to invest.

A saver gets the mirror image, and the bad start is worth 33% more

Run the same arithmetic with the cash flow reversed and you get a result that rarely gets said out loud. Take someone contributing $10,000 a year in today's money, at the start of each year, into the same 60/40 portfolio, starting from zero. Over the same 30 returns they pay in $300,000 either way.

In the original 1993-2022 order they finish with $675,755. In the reversed order — the one that opens with 2022's 20% real loss — they finish with $899,160. That is 33% more money from the sequence that cost the retiree 37%.

It's the same mechanism with the sign flipped. Contributions into a fallen market buy more units, and those units are still there when the good years arrive. A saver's largest balance sits at the end, so late returns dominate; a retiree's largest balance sits at the start, so early returns dominate. Sequence risk isn't a market phenomenon that happens to strike retirees. It's a property of the direction your cash flow runs.

Run a genuinely hostile sequence and one retiree runs out

The 1993-2022 window was chosen by rule, and it was generous. History has harsher runs. Applying the same construction to 1966-1995 — a period whose arithmetic average real return was 5.06% and whose compound average was 4.26% — the untouched lump sum again lands in the same place both ways, at $3,493,735.

The retiree taking $40,000 real doesn't. In the original order the portfolio falls to $459,678 by the end of 1974 and is exhausted in 1992, the 27th year. Reversed, the same 30 returns leave $1,786,828 standing after 30 years. One order funds the plan with money to spare; the other runs dry three years early.

That result is also a warning about how sensitive these demonstrations are to their assumptions. William Bengen's 1994 paper — the origin of the 4% rule, reprinted by the Journal of Financial Planning — found that a 4% inflation-adjusted withdrawal from a 50/50 portfolio had never been exhausted before 33 years in any starting year since 1926, with 4.25% capable of failing in as little as 28. His 1966 retiree lasted 33 years. The one modelled here lasts 27. The difference is not a contradiction so much as a demonstration: Bengen used Ibbotson data with intermediate-term Treasuries, which held up far better through the 1970s than the 10-year Treasuries in Shiller's file, and different withdrawal timing. Change the bond sleeve and the answer moves by six years.

The strongest counter-argument: a mirrored sequence is a teaching device, not a probability

Sequence risk is real, and it's also routinely overstated. Three objections deserve their full weight.

The first is that a reversed sequence is an artefact. Markets never offered anyone the reverse of 1993-2022, and the 30 returns used here can be arranged in 30 factorial ways — more than 2.6 x 10^32 distinct orders. Javier Estrada's Sequence Risk: Is It Really a Big Deal?, using an all-equity portfolio and S&P 500 real returns from 1900 to 2019, makes this concrete. Across 91 overlapping 30-year retirements, a 4% inflation-adjusted plan failed four times — a 4.4% failure rate. He then reshuffled the returns of each failing period 10,000 times. The 1929-1958 returns failed in 8.8% of orderings; 1966-1995 in 6.1%; 1968-1997 in 2.0%; 1969-1998 in 1.2%. His 6.1% is not a like-for-like companion to the 1966-1995 run above: Estrada holds 100% stocks throughout, chosen deliberately because sequence risk hits the most aggressive allocations hardest, while the portfolio here is 60/40. The two settle the same question about ordering, not the same question about a mix. Those retirees were not typical victims of a bad decade. They drew unusually unlucky orders of a set of returns that mostly worked. Estrada also finds the correlation between a retirement's sustainable withdrawal rate and its first-15-year return is 0.91, against 0.85 for the first ten years — so the danger window is far longer than the "first few years" the phrase usually implies, which makes any hedge against it correspondingly more expensive.

The second is that nobody actually withdraws like this. A fixed real withdrawal that never flexes is a modelling convenience, not a description of behaviour. David Blanchett's analysis of the RAND Health and Retirement Study, a longitudinal survey of US retirees' actual spending, found real retiree spending falls by an average of 0.96% a year between ages 60 and 90, tracing a "retirement spending smile" that dips through the middle years. In his simulations a 4% starting withdrawal had a 73.3% success rate held rigidly in real terms, rising to between 79.9% and 91.1% once spending followed the observed curves instead. Morningstar's December 2025 research reaches the same place from the other direction. Its estimate is 3.9% for a retiree who wants perfectly steady inflation-adjusted spending over 30 years, with a 90% chance of money remaining. On the same 40% equity/60% bond portfolio, its constant-percentage and endowment methods start at 5.7%. Those 1.8 percentage points are the price of rigidity.

The third is that a portfolio is rarely the whole picture. Morningstar's figure explicitly excludes Social Security and other non-portfolio income. A retiree whose fixed costs are largely covered by a state pension or a defined-benefit scheme is running a much smaller version of this experiment, because the part of their spending that must come from the portfolio is the part that can flex.

None of that makes the demonstration wrong. Morningstar's own 2025 work found that retirees who met poor returns in their first five years and did not adjust spending were much likelier to exhaust their savings than those whose first five years were positive. That is the same mechanism, showing up in a forward-looking simulation rather than a backward-looking one. What the mirrored sequence cannot do is tell you how likely it is.

What this demonstration cannot tell you

One historical sequence is one sample. Thirty returns, drawn once, from a single country over a single generation — and the US equity record is the survivor's record, not the world's. Applying the identical method to 1966-1995 rather than 1993-2022 changes the ending balances by millions, which is the honest measure of how much weight a single window can bear.

The construction embeds choices that all move the answer. A 60/40 mix rebalanced annually. Withdrawals at the start of the year rather than monthly. Zero fees, zero taxes, zero trading costs. Shiller's 10-year Treasury series rather than the intermediate-term Treasuries Bengen used. Run the same 1993-2022 test on a 100% equity portfolio and the two retirees end at $3,900,295 and $3,327,421 — the same direction, a far smaller gap, because equities and bonds fell together in 2022 but not in 2000-2002. And a real portfolio isn't a smooth annual return: someone drawing monthly through 2008 experienced something the yearly figures can't show.

The numbers here are also arithmetic on a fetched dataset, not a forecast. They describe what these returns did in these two orders. They say nothing about what any 30 years starting now will do.

What would change the conclusion

If withdrawals flex with the portfolio, most of the gap closes. The entire divergence above is generated by taking $40,000 out of a portfolio that had just fallen 20%. Blanchett's and Morningstar's numbers both point the same way: the damage is done by the rigidity, not by the returns. A plan that isn't rigid isn't running this experiment.

If the portfolio isn't the main income source, the effect shrinks in proportion. Sequence risk scales with how much of your spending the portfolio has to produce, which is why the same market can be a crisis for one retiree and an inconvenience for another with identical holdings.

If stocks and bonds keep falling together, the 60/40 construction here stops being a diversified case and starts being a leveraged bet on one factor. 2022 was the clearest instance in this dataset — equities at -20.14% real and 10-year Treasuries at -20.16% — and it's the single year doing the most work in the reversed sequence. A repeat of that correlation would make bad sequences more common than the historical record implies.

If you're still contributing, the sign flips entirely, and the reasoning above says a poor decade early is worth having. The uncomfortable implication is that the same market event is genuinely good news for one household and genuinely bad news for another, and the only thing distinguishing them is which way the money is flowing.

What the demonstration does establish is narrow and worth keeping: the ordering of returns is invisible in an average and invisible in a lump sum, and it's the dominant variable once withdrawals begin. Any figure quoted as an "average return" is silent on the thing that decided the outcome here. Tracking the balance and the withdrawal rate it implies — the 3.84% and 5.22% above — is what makes that variable visible, whether the tool is LedgerTouch or a spreadsheet. Related reading: how rebalancing rules compare over 92 years, and what badly timed cash flows cost investors — the same arithmetic, running through behaviour instead of retirement dates.

Sources

  1. Robert J. Shiller, ie_data.xls (current file, data through July 2026) — monthly S&P Composite real total-return price index (column J) and real total bond returns for 10-year US Treasuries (column S), plus CPI-U. All 30 annual real returns for calendar years 1993-2022 used in this article are computed from December-to-December values in this file; CPI 141.9 (Dec 1992) and 296.797 (Dec 2022). (img1.wsimg.com)
  2. Robert J. Shiller, Online Data — documentation for the ie_data series: 'Stock price data are monthly averages of daily closing prices through January 2000, the last month available as this book goes to press' (Shiller's current documentation at shillerdata.com carries the same sentence without that closing qualifier); CPI-U from the US Bureau of Labor Statistics; series begins January 1871. Also hosts the Yale mirror of ie_data.xls, whose 1993-2022 annual returns are identical to the current file (the indices differ only by a constant rebasing factor of 1.0999). (econ.yale.edu)
  3. US Bureau of Labor Statistics, CPI-U All Items series CUUR0000SA0 (cu.data.1.AllItems) — December 1992 = 141.9 and December 2022 = 296.797, a factor of 2.0916 (+109.2%, or 2.49% a year), confirming the CPI column used by Shiller to deflate the return series. (download.bls.gov)
  4. William P. Bengen, 'Determining Withdrawal Rates Using Historical Data', Journal of Financial Planning, October 1994 (FPA reprint) — a 4% first-year withdrawal with inflation-adjusted increases from a 50/50 stock and intermediate-term Treasury portfolio 'in no past case has it caused a portfolio to be exhausted before 33 years'; 4.25% 'could exhaust a portfolio in as little as 28 years'; 3% never below 50 years; the 1966 start year gives 33 years and 1969 gives 36; data from Ibbotson SBBI 1992 Yearbook, start years from 1926. (financialplanningassociation.org)
  5. Javier Estrada, 'Sequence Risk: Is It Really a Big Deal?' (IESE Business School, October 2020; Journal of Investing 30(6), 2021) — S&P 500 annual real returns 1900-2019, annualized 6.5% real with 20.0% volatility; 91 overlapping 30-year retirements; a 4% inflation-adjusted plan failed 4 times, a 4.4% failure rate; reshuffling each failing period's 30 returns 10,000 times gives failure rates of 8.8% (1929-1958), 6.1% (1966-1995), 2.0% (1968-1997) and 1.2% (1969-1998); correlation of the maximum withdrawal rate with first-15-year return is 0.91 versus 0.85 for the first ten years. (blog.iese.edu)
  6. Amy C. Arnott, Christine Benz and Jason Kephart, Morningstar, 'What's a Safe Retirement Withdrawal Rate for 2026?', 3 December 2025 — 3.9% is the highest safe starting withdrawal rate for steady inflation-adjusted spending at a 90% probability of funds remaining over 30 years, for a 30-50% equity weighting, excluding Social Security and other non-portfolio income; prior base cases 3.3% (2021), 3.8% (2022), 4.0% (2023), 3.7% (2024); retirees tolerating spending fluctuation 'can start with a withdrawal rate of nearly 6%'; retirees hit by poor returns in the first five years who did not adjust spending were much likelier to exhaust savings. (morningstar.com)
  7. David Blanchett, 'Exploring the Retirement Consumption Puzzle', Journal of Financial Planning, May 2014 — using RAND Health and Retirement Study spending data matched to the RAND Consumption and Activities Mail Survey (CAMS), real retiree spending falls an average of 0.96% a year from age 60 to 90 (t = -4.31), tracing a 'retirement spending smile'; a 4% initial withdrawal rate has a 73.3% probability of success held constant in real terms, versus 79.9%, 86.0% and 91.1% under the observed spending curves. (financialplanningassociation.org)
  8. Robert J. Shiller, ie_data.xls (Yale mirror, data through September 2023) — independent copy of the same dataset used to confirm that the 1993-2022 annual real returns are unchanged across file vintages. (econ.yale.edu)

Research Disclosure

This content is for informational purposes only and does not constitute financial advice. Always do your own research or consult a qualified financial advisor before making investment decisions.

Published · Last reviewed . Data can revise after publication, so validate critical figures at source before making allocation changes.