The Average Retiree’s RMD, by Age

There is a question people ask that nobody seems to have answered with actual data: what does a required withdrawal look like for an ordinary retiree? Not the formula, which is one division and is written up in full elsewhere on this site. The dollars.

The ingredients have been sitting in public for years. The IRS publishes, every year, the number of taxpayers holding IRAs in each age band and the total year-end fair market value of what they hold. Divide the second by the first and you have an average balance per age band. Divide that by the life-expectancy factor the regulations assign to that age and you have the withdrawal the table would call for.

Nobody publishes the second division. So I ran it, and this page is the result, along with a limitation serious enough that I want it in your hands before the table rather than after it.

The IRS balance column counts every kind of IRA together — traditional, Roth, SEP and SIMPLE — and a Roth IRA requires no withdrawal at all during the owner’s lifetime. So a straight division overstates the requirement. What makes that answerable rather than a shrug is that the size of the overstatement is published too: a second table in the same IRS release splits the identical pool of dollars by plan type, and Roth accounts hold 11.54 percent of it. So the table below carries both figures — the ceiling, and the ceiling scaled by the share of IRA dollars that actually carries a lifetime distribution requirement — and says which is which. No model, no estimated share, no assumption about anybody’s account. Every input traces to a spreadsheet you can download from a .gov domain and check against me.

Where do these figures come from?

Two public sources, one of which is a regulation and one of which is a spreadsheet.

The balances come from the IRS Statistics of Income program, in the series on accumulation and distribution of individual retirement arrangements. Table 4 of that release reports two columns I care about: the number of taxpayers in each age band with an end-of-year fair market value on file, and the total dollars they hold. It is built from Form 5498 filings, the information return a custodian sends the IRS about an IRA it holds.

The divisors come from the Uniform Lifetime Table at Treas. Reg. §1.401(a)(9)-9(c), which the regulation describes as setting out “the applicable denominator that applies for lifetime distributions to an employee in situations in which the employee’s surviving spouse is not the sole designated beneficiary.” That is the table almost every account owner uses, and the full table sits on its own page here, rendered from a sourced module rather than typed.

The arithmetic joining them is the one the regulation specifies. Treas. Reg. §1.401(a)(9)-5(a)(1) says the required amount “is equal to the quotient obtained by dividing the account balance … by the applicable denominator.” The balance on top is meant to be last December’s closing figure — for an IRA, Treas. Reg. §1.408-8(b)(2) uses “the account balance of the IRA as of December 31 of the calendar year preceding the calendar year for which distributions are required to be made” — and a year-end fair market value from Form 5498 is exactly that measurement, taken across millions of accounts at once.

I checked the age bands against the table’s own total row before using them. They sum to the published total on both count and dollars, to the last digit printed. That is not a claim that the IRS sampling is right; it is a check that I read the columns I thought I was reading.

What does the required withdrawal look like at each age?

Below the arc of it, in dollars, for the most recent year the IRS has published.

Read the second column as the average balance for the age band containing that age. The third is that balance divided by the divisor the regulation assigns to that single age — the ceiling, which is the answer if every dollar were subject to distribution. The fourth is the same figure scaled by 88.46 percent, the published share of IRA dollars that actually carries a lifetime requirement, and it is the closer of the two to what an ordinary account holder owes.

Both right-hand columns are my own arithmetic rather than government statistics, and neither is anybody’s actual requirement.

Three age bands are doing all the work there, because that is the resolution the IRS publishes: one figure for ages 70 through 74, one for 75 through 79, and one for everybody 80 and older. So the balance holds still inside a band and then steps down at the boundary, which is an artifact of the reporting rather than a thing that happens to an account on a birthday.

The bottom four rows need a warning label, and this is the weakest joint in the table. Ages 80, 85, 90 and 95 all draw on a single published average — one figure covering everybody from 80 to the end of life. Inside that range real balances almost certainly fall with age, because withdrawals have been compulsory for a decade or more by then and the required share climbs the whole time. The average I am dividing does not fall, though. So the rise across those four rows is the divisor working alone, applied to a balance the reporting has frozen, and the last rows overstate for a second reason on top of everything else on this page. If you take one figure away from here, take one from the seventies, where the bands are five years wide and the freeze is short.

What is real is the shape inside each band. Hold the balance constant and the required amount climbs anyway, because the divisor underneath it keeps shrinking. Between the two ends of the eighties the same average balance produces a withdrawal about a third larger, and by the mid-nineties it has roughly doubled from where the decade began. Nothing about that requires a market, a return, or a decision. It is the table doing what the table is designed to do, which is distribute an account over a life expectancy rather than preserve it.

The other thing worth noticing is how modest the early figures are. A household in its seventies with an average-sized IRA is looking at a withdrawal in the middle five figures — real money, and nothing like the tax event people brace for. The households for whom this is genuinely disruptive are not average households, which is the next problem.

Why is the average nothing like the typical?

Because retirement balances are enormously skewed, and the IRS does not publish the number that would show it.

An average is total dollars divided by total people. In a population where a small number of accounts hold a very large share of the dollars, that average sits well above the middle of the distribution — the balance where half of account holders are above and half below. The averages in the table are therefore pulled upward by large accounts, and the typical retiree in each band almost certainly holds less than the figure printed.

I want to be exact about the status of that statement, because it is the kind of thing that gets repeated as a finding. It is not one. SOI Table 4 publishes aggregate dollars and taxpayer counts per band and no percentiles at all, so the median balance by age is not in the data I am using, and I cannot show you how far below the mean it falls or prove that it does. What I can say is that the arithmetic here is mean arithmetic, and a mean is the wrong statistic for the question “what does this look like for someone like me.”

So do not read the table above as a benchmark you are failing or beating. Read it as the scale of the thing.

How much of the balance actually owes a distribution?

Just under 88.5 percent of it, and that figure is measured rather than assumed.

Start with why it matters. Treas. Reg. §1.408-8(b)(1)(ii) is about as plain as regulations get: “No minimum distributions are required to be made from a Roth IRA while the owner is alive.” A Roth IRA has no divisor, no required amount and no place in the calculation. But Form 5498 is filed for traditional, Roth, SEP and SIMPLE arrangements alike, so the age table reports all four as one number.

Table 1 of the same IRS release splits that number. It reports end-of-year fair market value by plan type for the same tax year, and its total row is identical to the age table’s — $14,583,955,520 thousand in both. (The four plan-type rows add to a thousand dollars more than that, which is rounding inside a table printed in thousands, on a total near fifteen trillion.) That identity is the part that makes this usable. The two tables are describing one pool of money from one set of filings, so the split can be applied to the whole directly, rather than being borrowed from a different survey of a different population and hoped to transfer.

The split, for tax year 2023:

SEP and SIMPLE accounts belong with traditional money, not with Roth, and filing them the other way would understate the requirement by about four points. Treas. Reg. §1.408-8(a)(4) settles it: IRAs receiving employer contributions under a SEP arrangement or a SIMPLE IRA plan “are treated as IRAs, rather than employer plans, for purposes of section 401(a)(9) and are, therefore, subject to the distribution rules in this section.” So three of the four rows above carry a lifetime requirement, and one does not.

Add the three and you get 88.46 percent of all IRA dollars subject to distribution, and 11.54 percent — the Roth row — exempt for as long as the owner is alive. That is the number scaling the right-hand column of the main table.

Now the limitation, which is real and which no arithmetic closes.

That share is an aggregate across every age, not a measurement of any one band. SOI publishes no plan-type by age cross-tabulation, so there is no published Roth share for people in their eighties, or their seventies, or any other slice. Applying 88.46 percent to a single age band treats a whole-population ratio as though it had been measured inside that band, and it has not been. The adjusted column is therefore a better figure than the raw ceiling and still not a measurement.

Which direction that error runs is worth one paragraph, and it turns on a date rather than a guess. Roth IRAs did not exist before 1998: they were created by section 302 of the Taxpayer Relief Act of 1997, Public Law 105-34, which added section 408A to the Internal Revenue Code and whose subsection (f) applies the change to “taxable years beginning after December 31, 1997.” Someone who is 85 now was in their late fifties when the accounts first appeared, and someone who is 95 was well past retirement. The oldest cohorts had far fewer working years in which to fund one.

So the Roth share among the 80-and-over band is very likely below the all-ages 11.54 percent, which means the adjusted column understates the requirement slightly at exactly the ages this article spends the most time on. Understating is the conservative direction for a ceiling, and it is one more reason to treat the older rows as the softest figures here.

If you want the number that is actually yours rather than a national ratio, your own statements have it, and the split takes about a minute to read off them. Every dollar you have converted is a dollar that has left this calculation permanently, which is a large part of why conversion work tends to cluster in the years between retiring and the first required withdrawal.

If it helps to have the numbers in one place, I keep them all in the Ultimate Retirement Guide.

Why does the average balance peak in the early seventies?

It does, and the reason is more interesting than the fact.

The band covering ages 70 through 74 holds the largest average balance in the whole table — larger than the band below it, and larger than either band above. After it, the averages fall.

The obvious story is that required distributions start drawing accounts down, and that is probably part of it. But this data cannot demonstrate it. Table 4 is a cross-section: it photographs everybody at once. The people in the 80-and-over band are not the people in the 70-to-74 band a decade later — they are a different cohort, who worked in different decades and saved under different rules. Spending, gifting, market history, charitable transfers and plain mortality all sit inside that downward slope alongside required withdrawals.

Separating those causes takes following the same people through time, and this is not that kind of data. So the correct thing to say is that balances are lower in the older bands, and required distributions are one of several forces consistent with it. Anything stronger is a life-cycle claim the data cannot carry.

There is a second lesson in that peak that does not depend on causation at all. The largest average balance in the table sits precisely at the age when the withdrawals become mandatory. Whatever is true about the individuals, the aggregate says the requirement arrives at the moment there is the most to distribute.

What else can this data not tell you?

Four things, and they are worth knowing before you quote any of it.

It suppresses some cells. In the 2022 release, two of the youngest age bands are blank rather than zero — the IRS withholds cells where publishing them could risk identifying individual taxpayers, a routine disclosure-avoidance practice across the SOI program. Neither band matters for required withdrawals, but it is a fact about how this data is made: an empty cell is a decision, not a gap in the world.

It only sees IRAs. Workplace plans are not in this table. A household with an old 401(k) or 403(b) still sitting where they left it has a required withdrawal this article knows nothing about, and those plans do not aggregate with IRAs — which accounts combine and which never do is its own subject and the most common way a household ends up short.

It is per taxpayer, not per household. The calculation runs per person, and a couple with two IRAs appears here as two entries. Nothing describes a household total.

It is years old the day it is published. The most recent figures are for the 2023 tax year, and the ones before them for 2022, when the averages in every band were lower. Balances have moved since, and the table above is a picture of particular year-ends rather than a current statement of anything.

What should you do with your own number instead?

Run the same division on your own two figures, which will take less time than reading this section.

Take your traditional IRA balance as of the last day of last year — the figure on the December statement, not today’s, and not including Roth accounts. Find the age you reach during this calendar year, read the divisor across from it on the Uniform Lifetime Table page, and divide. The RMD calculator here runs the same table and lets you enter more than one account, which matters if your IRAs sit at several firms.

One input catches people out more than the rest, and it is not arithmetic. Which year is your first one is set by your birth year rather than by a single universal age. Treas. Reg. §1.401(a)(9)-2(b)(2)(iv) gives an applicable age of 73 for someone “born on or after January 1, 1951, but before January 1, 1959,” and paragraph (vi) gives 75 for someone “born on or after January 1, 1960.” Read those two boundaries next to each other and birth year 1959 falls between them, into a paragraph the regulation prints as “[Reserved]” — genuinely unresolved in the published text, so if that is your birth year it is a question for your tax preparer rather than one any article can close.

Compare the answer against your custodian’s figure, and if they disagree find out why in January rather than in December. A custodian can only see the accounts it holds.

What do people get wrong about this?

Four things, and only one of them is about the math.

Treating an average as a benchmark. The figures here describe an aggregate, and skew means the typical account holds less than the mean. Being below the number in the table is not a diagnosis of anything.

Reading the ceiling column when the adjusted one exists. The raw division includes Roth money that owes nothing, and about one IRA dollar in nine is Roth. If you carry a meaningful Roth balance of your own, even the adjusted column overstates your requirement, and the overstatement grows with every conversion you have made.

Assuming the withdrawal is flat. The most common misreading of the requirement is that it is a fixed percentage of the account. It is not. The divisor shrinks every year, so the required share climbs on its own, and it climbs fastest in the decade when a spouse’s death or a move into care is changing everything else too.

Reading a cross-section as a life story. The declining balances in the older bands describe different people, not the same people getting older. It is the single easiest mistake to make with published age-band data and it is made constantly, usually in the service of a point about retirees running out of money.

The reason to look at data like this at all is not to find out whether you are normal. It is to see the shape of a rule that will apply to you for the rest of your life, in dollars rather than in divisors, before it starts.

Sources

Everything above comes from four documents, and all four are free to open.

IRS Statistics of Income, Table 4, “Taxpayers with Individual Retirement Arrangement (IRA) Plans, by Age of Taxpayer,” tax years 2023 and 2022 — the taxpayer counts and end-of-year fair market values by age band, from files 23in04ira.xlsx and 22in04ira.xlsx in the SOI release on accumulation and distribution of individual retirement arrangements. Money amounts are printed in thousands of dollars. Pulled 14 August 2026.

IRS Statistics of Income, Table 1 of the same release, tax year 2023, file 23in01ira.xlsx — end-of-year fair market value by plan type, which is where the traditional, SEP, SIMPLE and Roth figures come from. Its total matches Table 4’s exactly. Pulled 14 August 2026.

Treas. Reg. §1.401(a)(9) — the Uniform Lifetime Table at §1.401(a)(9)-9(c), the division itself at §1.401(a)(9)-5(a)(1), and the birth-year applicable ages at §1.401(a)(9)-2(b)(2). Text checked against the eCFR on 14 August 2026.

Treas. Reg. §1.408-8 — the IRA rules: the prior-December-31 balance at (b)(2), the Roth exemption at (b)(1)(ii), and the SEP and SIMPLE treatment at (a)(4). Text checked against the eCFR on 14 August 2026.

Taxpayer Relief Act of 1997, Public Law 105-34, section 302 — the creation of section 408A and the effective date that makes 1998 the first year a Roth IRA could exist. Text read at govinfo.gov on 14 August 2026.

The two derived columns in the main table, the plan-type percentages, and every comparison between years are my own arithmetic on those sources. If you redo it and get something else, I would like to know.