Sequence of Returns Risk Explained
Quick answer: Sequence of returns risk is the way the order of good and bad years can change how long a portfolio lasts once money is being taken out of it. In the worked example below, the same ten yearly returns, shuffled into opposite order, leave $690,000 in one case and $1,255,000 in the other. With no withdrawals, both orders end at exactly the same balance. The example is hypothetical, and nothing on this page forecasts anyone's retirement.
If you would rather watch the effect than read about it, the Monte Carlo simulation retirement calculator on the home page runs thousands of different orderings of returns for your own inputs and shows how the outcomes spread out.
Education, not financial advice. This page explains a method and shows what a model produced from stated assumptions. It does not tell you what to buy, sell, save or spend. For your own situation, ask a qualified professional.
The idea in plain terms
Most retirement projections you meet in everyday life use one number: an average yearly return. For example, type in a single yearly return, multiply it out, and a smooth line climbs across the page. Real markets do not move in a smooth line. They bounce, and the bounces arrive in some particular order.
For a saver who is not touching the money, the order makes no difference to the final balance. Growing a balance by one set of yearly returns is a chain of multiplications, and a chain of multiplications gives the same answer whichever way you line the factors up. That is why the no-withdrawal row in the example below shows identical endings.
Taking money out changes that. A withdrawal is usually a fixed number of dollars, not a fixed share of whatever is left. When a bad year comes early, the balance shrinks, and the next withdrawal is then a bigger slice of a smaller pot. Whatever is left has to carry the next withdrawal and also climb back. When the same bad year comes late, there is a large balance to absorb it and fewer years of withdrawals still ahead.
Think of the first years of withdrawals as the stretch with the least room for error. Early on, every dollar taken out has the most remaining years in which it could have compounded, and a loss in those years is carried forward by every withdrawal that follows. Late in retirement, fewer future withdrawals still depend on what the balance does. The model does not label any year as the dangerous one; the pattern shows up because the same withdrawal schedule meets a different balance depending on what came first.
Planners call this sequence of returns risk, and you will also see it written as the order of returns. The word "risk" is slightly misleading, because nothing about it is a mistake or a fault in a plan. It is simply the part of the uncertainty that matters most in the years right around the day withdrawals begin.
Two orders, same ten returns
Here is a deliberately tidy example. Take a pot of $1,000,000 and a plan to withdraw $60,000 at the start of each of ten years, with the return for that year applied to whatever remains. The ten yearly returns are the same in both runs. Only their order differs, and the list averages 7.4% a year. These are made-up numbers chosen to make the point visible, so treat the whole example as hypothetical.
- The low-first order runs -18%, -8%, +0%, +5%, +9%, +12%, +15%, +17%, +20%, +22%.
- The high-first order runs +22%, +20%, +17%, +15%, +12%, +9%, +5%, +0%, -8%, -18%.
| Year | Bad years first: return | Balance | Good years first: return | Balance |
|---|---|---|---|---|
| 1 | -18% | $771,000 | +22% | $1,147,000 |
| 2 | -8% | $654,000 | +20% | $1,304,000 |
| 3 | +0% | $594,000 | +17% | $1,456,000 |
| 4 | +5% | $561,000 | +15% | $1,605,000 |
| 5 | +9% | $546,000 | +12% | $1,730,000 |
| 6 | +12% | $544,000 | +9% | $1,821,000 |
| 7 | +15% | $557,000 | +5% | $1,849,000 |
| 8 | +17% | $581,000 | +0% | $1,789,000 |
| 9 | +20% | $625,000 | -8% | $1,590,000 |
| 10 | +22% | $690,000 | -18% | $1,255,000 |
| End, with withdrawals | $690,000 | $1,255,000 | ||
| End, no withdrawals | $1,905,000 | $1,905,000 |
Read the balances year by year. In the low-first run, the balance after each year's withdrawal and return is $771,000, $654,000, $594,000, $561,000, $546,000, $544,000, $557,000, $581,000, $625,000, $690,000. The pot never gets back to the million it started with. It bottoms out in year six, and although the last four years bring the biggest gains in the list, those gains land on a balance that has already shrunk. The run ends at $690,000.
In the high-first run the balances are $1,147,000, $1,304,000, $1,456,000, $1,605,000, $1,730,000, $1,821,000, $1,849,000, $1,789,000, $1,590,000, $1,255,000. Early gains build a cushion that is well above the starting million for most of the decade. The balance peaks in year seven, then falls for the last three years as the two worst returns arrive, and ends at $1,255,000.
Same ten returns, same average, same withdrawals, and a gap between the endings that is large relative to either one. Now remove the withdrawals and run both orders again. The low-first order ends at $1,905,000, and so does the high-first order ($1,905,000). The order stopped mattering the moment nothing was being sold.
Two details are worth noticing. First, the example is not a story about one run having "better" returns, because the lists are identical. Second, the loss in the low-first run is not made up later by the strong years, because later gains apply to fewer dollars. The model is not doing anything exotic here, it is only subtracting and multiplying in a particular order.
What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.
What real history shows
The tidy example is built by hand. The dataset behind this site's calculator lets you look at real calendar years instead, so here is a second illustration using it. The numbers come from Aswath Damodaran's annual returns workbook from NYU Stern, which carries stock, bond and gold returns for 1928 through 2025, and every return is converted to a real (after inflation) return. The replay below is not part of the calculator itself. It is a fixed example: a $1,000,000 portfolio in a 60/40 mix of stocks and bonds, a first-year withdrawal of 4 percent of the starting balance ($40,000), then the same buying power each year after that, for 30 years.
Start that example in 1966 and the money runs out in 1991, ending at $0. Start it in 1982 and the money does not run out, and the portfolio ends at $5,360,000 in today's dollars. The two starting years were not picked at random. In this dataset, 1966 is the start year whose 30-year window ran out of money soonest, and 1982 is a start from which the same plan finished far ahead.
It is tempting to read those two lines as a clean case of the sequence effect, but they are not one, and it matters to say why. The 1966 window averaged 4.8% a year in real terms across the 30 years, while the 1982 window averaged 8.0%. Those are different averages, not the same set of returns in a different order, so part of the gap between the two endings is simply that one window earned less. What the pair does show is the same pattern as the hand-built example: the first ten years of the 1966 window averaged -1.0% a year, and the first ten years of the 1982 window averaged 12.1%. A weak opening decade, with withdrawals under way, is hard for a portfolio to recover from. This is a statement about what happened in this dataset under these assumptions, not about what any later start year will do.
For the full comparison of how a replay of past years differs from random draws, the page on Monte Carlo vs historical simulation goes through the same data both ways.
What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.
What tends to reduce the exposure
None of what follows is a recommendation, and none of it is free. These are the ideas that people who study withdrawals most often discuss when they talk about the order of returns, along with what each one trades away.
Spending that can flex. If the dollars taken out can drop after a poor year, the balance has a better chance to recover, because fewer shares or dollars leave while prices are low. The cost is plain: your spending becomes less predictable. How far a plan can flex is a personal question, and a model cannot answer it. The page on the safe withdrawal rate shows how fixed, percentage-based and guardrail-style withdrawals behaved side by side in one simulation, including how often the spending in the guardrail version had to be cut.
A cash buffer. A common idea is to hold a reserve of cash so that a few years of withdrawals do not have to come from stocks after a fall. One 2011 conference deck co-authored by Evensky describes a reserve of two years of living expenses. The reserve earns less than stocks do in a good stretch, so the buffer has a price. In the model on this site, adding a cash bucket made little difference to the success rate, and the bucket strategy page reports exactly what the numbers show.
Spreading across more kinds of assets. Stocks and bonds do not always fall together, and other assets may behave differently again. How much that helps depends on the period and on the assumptions, and a mix can also fall short of the stocks-only result in a long run of strong stock years. Some readers ask whether gold belongs in that conversation. The page on gold in a retirement portfolio lays out what gold did in dated stretches of the dataset, including the long stretch when it earned nothing after inflation, and shows that a model can go either way depending on the assumption.
Earning a bit longer. In the calculator, the retirement age is an input. Moving it changes how many years of savings go in and how many years of withdrawals follow. The page on what a Monte Carlo simulation cannot see uses the same default plan to show how large that one input can be.
If you want the longer version of that story, including who the firm says it serves and which questions are worth putting to any company in the field, the page on the Augusta Precious Metals gold IRA gives an honest fit passage and keeps its own claims attributed to the company.
How to see the effect in a simulation
A Monte Carlo run does not replay a single sequence. It builds thousands of them, and some of those open with several poor years in a row. When you read a result, the fan chart is where that shows up: the lower band is made of the paths that started badly. The more weight a plan puts on money withdrawn in the first years, the further that lower band tends to drop.
A few habits make the picture clearer. Look at the lowest band, not only the median, because the median path hides the bad openings. Look at the age at which failing paths ran out, which the results card reports. And try the same plan with a different retirement age or a lower spending number, since those two inputs tend to move the result more than any tweak to the mix. The page on what a Monte Carlo simulation is explains how the thousands of paths are built, and the retirement Monte Carlo calculator lets you change one input at a time.
One caution about the model itself: it draws each year's return separately from the last. Real markets sometimes show runs, where a weak year is followed by another weak one, and a model that draws years independently can understate how long a rough stretch lasts. That is one of the limits spelled out on the page about what a Monte Carlo simulation cannot see.
What this can't tell you
This page can't tell you whether your own retirement will begin in a good or bad stretch. It also can't tell you how large your first-decade returns will be, how your spending will move, or whether the future will resemble any window in a dataset that starts in 1928 and holds only a few non-overlapping decades. The worked example uses ten tidy returns picked to make a point. The history uses one portfolio mix, one withdrawal rule and the assumptions shown. Real life adds taxes, health costs, changes in spending and income that does not come from the portfolio, and none of that is in the numbers here.
What the page can do is name the mechanism so that a score on a calculator makes more sense. A success rate is partly a measure of how often the early years go badly in the simulated paths, and that is a reason to read it as a description of the model's assumptions and not as a forecast.
Questions readers ask
Is sequence of returns risk the same as having a low average return?
No. The example above holds the average fixed at 7.4% and still produces very different endings, because the order changes how many dollars are left when each return arrives. A low average return is a separate problem, and the two can occur together or apart.
Does the order of returns matter while I am still saving?
Less in the same way. In this model, savings are added before each year's return is applied, so a weak early year hits a small balance, and a weak late year hits a large one. With contributions going in rather than coming out, the arithmetic of the withdrawal example does not apply in the same direction.
Can a Monte Carlo model show sequence risk if its returns are random?
Yes. Each simulated path draws its own order of yearly returns, and a share of the paths open with weak years. The lower bands and the failure ages in the results come from those paths. The limit is that the draws are independent from year to year, so long runs are rarer than in some real periods.
If a plan shows a high success rate, is sequence risk gone?
No. A high score means that most simulated paths, under the assumptions shown, ended with money. The paths that did not are often the ones with a weak start. Sequence effects are still inside the score, only less frequent than in a plan with a lower score.