Key takeaways
- The 3-to-6-month range traces to financial-planning practice, not to research. A 1995 paper in Financial Counseling and Planning attributes it to planners and a textbook, then adopts it as a yardstick — one earlier study measured households against 6 months of spending, the next against 3.
- US median unemployment duration was 6.1 weeks in April 2000 and 25.2 weeks at the June 2010 peak, the highest reading since the series began in 1967. In June 2026 it was 11.0 weeks.
- The mean runs above the median by between 1.37 and 2.24 times across the rows below, and the gap is tightest in the worst years. In June 2010, 44.9% of unemployed Americans had been out of work 27 weeks or more, against 27.5% in May 2026 and 10.7% in April 2000 — and 26 weeks is where benefits stop in most states.
- How you earn separates households more than how much you earn. In the Fed's 2025 survey, 58% of self-employed adults said their monthly income varied, against 28% of employees, and 22% versus 10% said that variability had made bills hard to pay.
- Retirement, pensions and Social Security took 11.7% of the average US consumer unit's $78,535 of expenditures in 2024, and $6,684 of that was payroll deductions for Social Security — a bill that stops dead with the paycheck.
The range came from financial planners, and the research borrowed it
You want a number. How many months of spending, held in cash, before the rest goes to work? The answer already in circulation is three to six months, and it does have a paper trail — just not the kind most people assume.
A 1995 paper by Y. Regina Chang and Sandra Huston in Financial Counseling and Planning opens its discussion of adequacy like this: "Financial planners recommend a range anywhere from two or three months to six months, and in some cases a year's worth, of living expenses." The citations attached are a 1991 Wall Street Journal article and Garman and Forgue's personal-finance textbook.
They adopted the planners' number as a measuring stick. The same paper records that Hanna, Chang, Fan and Bae (1993) tested households against "a 6 months of spending guideline" while Hanna and Wang (1995) used "a 3 months of spending guideline". Chang and Huston themselves picked three months of income, and found only 32% of US households cleared it in either 1983 or 1986. Johnson and Widdows (1985), the field's definitional reference, settled what an emergency fund is — holdings available to cover spending without drastically altering the standard of living after an income disruption — and said nothing about how big.
So the arrow runs from practice to research, not the other way. Chang and Huston concede the point in passing: there's "no consensus regarding the exact amount".
The agencies decline to settle it either. The Consumer Financial Protection Bureau's guide to building an emergency fund names no figure. Its own 61-page research report on emergency savings goes further by omission: across the whole document the phrases "rule of thumb", "months of expenses" and "three to six" don't appear once. It sorts consumers at a month of income instead — 24% with nothing set aside, 39% with less than a month, 37% with at least a month.
The Federal Reserve does use three months, but as a ruler rather than a target. Its Survey of Household Economics and Decisionmaking calls savings sufficient to cover three months of expenses "one common measure of financial resiliency".
None of that makes the range wrong. It makes it untested.
How long unemployment actually lasts, and how far the cycle moves it
The US Bureau of Labor Statistics publishes both the median and the mean number of weeks that unemployed people have been looking for work. In June 2026, with the unemployment rate at 4.2%, the median was 11.0 weeks and the mean 25.5 weeks.
Neither number sits still. At the top of the early-1990s cycle, in December 1992, the median was 9.3 weeks. At the top of the post-2008 cycle, in June 2010, it was 25.2 weeks — the highest reading in a series that begins in July 1967. The mean peaked later still, at 40.7 weeks in July 2011.
| Month | Median weeks | Mean weeks | Mean ÷ median | Unemployed 27 weeks or more |
|---|---|---|---|---|
| November 1975 | 9.5 | 16.6 | 1.75 | 21.0% |
| May 1983 | 12.3 | 20.5 | 1.67 | 24.9% |
| December 1992 | 9.3 | 19.0 | 2.04 | 21.4% |
| June 2003 | 11.5 | 19.9 | 1.73 | 22.8% |
| June 2010 | 25.2 | 34.5 | 1.37 | 44.9% |
| March 2021 | 20.3 | 29.6 | 1.46 | 43.0% |
| May 2026 (current cycle high) | 11.6 | 26.0 | 2.24 | 27.5% |
| April 2000 (tight market) | 6.1 | 12.4 | 2.03 | 10.7% |
A rule expressed in months has to survive that whole range. Three months of expenses covered the median job search with room to spare in 1992 and in 2000. It didn't cover it in 2010 or in 2021.
The gap between the mean and the median closes in the worst downturns
The mean always sits above the median here. That's what a long right tail does to an average: most spells are short, a minority run very long, and the long ones drag the mean without moving the middle. But the gap isn't a constant, and it isn't a factor of two. Read down the fourth column and it runs 1.75 in 1975, 1.67 in 1983, 2.04 in 1992 and 1.73 in 2003 — then 1.37 in 2010 and 1.46 in 2021, the two worst rows in the table.
That inversion is the useful part. Long spells stop being the exception, the median climbs to meet the mean, and the ratio compresses. Treating the mean as roughly double the median gets the shape backwards in exactly the years that matter.
The cleaner read on the tail is the share of the unemployed out of work 27 weeks or more. That threshold isn't arbitrary. The Department of Labor's comparison of state unemployment insurance laws notes that "most states provide a cap of 26 weeks of benefits". So the 27-week share is roughly a count of people whose benefits have run out.
Both figures describe spells still running, not finished ones
A March 1996 Chicago Fed Letter makes the point precisely: the published figures describe spells that are in progress on the survey date, not completed spells. Two consequences follow. Most of the people counted are still unemployed and keep accumulating weeks after they're counted. And a long spell gets counted repeatedly — a 26-week spell can appear in six consecutive monthly surveys, while a two-week spell appears at most once.
The second effect is the stronger one. In the Chicago Fed's worked example, December 1995's published mean stood at 16.2 weeks and the median at 8.2, but a randomly chosen worker becoming unemployed that month would on average have found work in something closer to 14 weeks.
Income volatility, not income level, is what separates one household from another
The rule treats all incomes as the same kind of object. The Fed's own survey says they're not.
The 2025 SHED, fielded from 17 to 28 October 2025, found that 30% of US adults had income that varied at least occasionally through the year. Split by how the income is earned and that average breaks apart: 58% of self-employed adults said their income varied from month to month, against 28% of those working for someone else. Twenty-two percent of the self-employed said income variability had made bills hard to pay in the previous 12 months, against 10% of employees.
The JPMorgan Chase Institute measured the same thing from bank records rather than self-report. Its Weathering Volatility study tracked 100,000 individuals' Chase accounts from October 2012 to December 2014 and found that 84% experienced month-to-month income swings larger than 5%. Its estimate was that a middle-income household needed roughly $4,800 in liquid assets to absorb a simultaneous income dip and spending spike, and that the median such household held about $3,000.
It's a buffer derived from observed volatility and observed expense shocks, not a round number of months applied to everyone. Two earners whose jobs are uncorrelated, an employee rather than a contractor, and a sector that doesn't lay off in the same quarter as everyone else are all inputs the three-to-six framing has no slot for.
"Months of expenses" is a smaller number than "months of spending"
The second input the rule leaves out is what share of spending actually survives a lost paycheck. The BLS Consumer Expenditure Survey puts average annual expenditures per US consumer unit at $78,535 in 2024. Here's where it went.
| Category | Dollars | Share of total |
|---|---|---|
| Housing (shelter $16,317, utilities $4,736, other) | 26,266 | 33.4% |
| Transportation | 13,318 | 17.0% |
| Retirement, pensions and Social Security | 9,222 | 11.7% |
| Food at home | 6,224 | 7.9% |
| Healthcare | 6,197 | 7.9% |
| Food away from home | 3,945 | 5.0% |
| Entertainment | 3,609 | 4.6% |
| Cash contributions | 2,292 | 2.9% |
| Apparel and services | 2,001 | 2.5% |
Retirement, pensions and Social Security, at $9,222, is the line that matters most here and gets discussed least. It's payroll-linked almost by definition: $1,991 of it is contributions to retirement plans and $6,684 is deductions for Social Security. That's 11.7% of expenditures an emergency fund never has to replace. BLS files this line inside a wider "personal insurance and pensions" heading worth $9,797, or 12.5%, and the $575 difference is life and personal insurance — which does keep falling due.
Add the categories that compress fastest — food away from home at 5.0%, entertainment at 4.6%, cash contributions at 2.9%, apparel at 2.5% — and 26.7% of the average consumer unit's expenditures are either paycheck-linked or squeezable within a month. On that split, a fund sized at 6 months of total spending is closer to 8 months of the bills that actually keep arriving. That split is an editorial judgement rather than a BLS category, and it moves enormously by household. Housing is 33.4% of expenditures and the least compressible line on the table, and a renter who can move faces different arithmetic from an owner with a fixed mortgage.
What households actually hold, in the US and in Europe
In the 2025 SHED, 63% of US adults said they'd cover a hypothetical $400 emergency expense using cash, savings or a credit card paid off at the next statement — unchanged for three years and down from a high of 68% in 2021. On the three-month question, 55% said they had set aside money to cover three months of expenses in a rainy-day fund, down from 59% in 2021; a further 15% thought they could cover three months by borrowing or selling assets; and 30% said they couldn't cover three months by any of those routes.
The distribution by income is where the rule visibly stops being one rule. Twenty-one percent of adults with family income under $25,000 had three months set aside, against 75% of those with income of $100,000 or more. The Survey of Consumer Finances adds the level: in 2022, the median family transaction-account balance was $8,000, among the 98.6% of families holding such an account, against median family income of $70,300.
The JPMorgan Chase Institute reached a much cheerier figure from 5.9 million households' bank records: 92% could cover a $400 shock, including 77% of the lowest income quartile. The gap between 63% and 92% is definitional, not contradictory.
None of this transfers cleanly across borders. Eurostat's 2024 figures put 30.0% of the EU-27 population in households unable to face an unexpected financial expense, but the country spread runs from 16.9% in the Netherlands, 18.3% in Malta and 19.2% in Czechia to 43.9% in Greece, 45.3% in Latvia and 45.6% in Bulgaria. The threshold is set relative to each country's own poverty line, so this isn't a like-for-like dollar comparison, but the ordering says something a US-derived rule cannot.
The case against a large cash buffer is stronger than it usually gets credit for
Cash has an opportunity cost, and over long horizons it's brutal. In Aswath Damodaran's returns dataset, $100 invested in 3-month US Treasury bills at the start of 1928 was worth $2,578.30 at the end of 2025. The same $100 in the S&P 500 was worth $1,157,598.95, and in 10-year Treasuries $7,752.88. Ninety-eight years isn't an emergency fund's holding period, and quoting that gap as though it applied to a rainy-day balance would be dishonest. But the direction is right and the drag compounds: a household carrying 12 months of expenses in cash from age 30 to age 60 isn't making a small allocation choice.
Right now the drag is unusually mild. The 3-month Treasury constant-maturity yield was 3.95% on 23 July 2026, against CPI-U inflation of 3.46% in the year to June 2026 — roughly half a point of real return. That gap has been far more negative in the past.
Some of the fund's job can also be done by something other than cash. An undrawn credit line costs nothing to hold. The JPMorgan Chase Institute found that 43% of low-income households unable to absorb small unexpected expenses might be able to pay off such shocks with more available credit. And for a household with two uncorrelated incomes, solid disability and health cover, and a stable employer, a very large cash balance may be insuring a risk that's already insured — while the cost of holding it is certain.
The objection to that case: credit is withdrawn exactly when it is needed
The counter is empirical, and in one specific state of the world it's decisive. Credit lines are not committed capital.
In the Federal Reserve's January 2009 Senior Loan Officer Opinion Survey, roughly 35% of domestic banks reported having trimmed existing consumer credit card account limits over the previous three months, and about 40% reported reducing the size of existing home equity lines of credit, on net. By the April 2009 survey, covering January to March, the card figure had risen to "approximately 50 percent (up from 35 percent in the January survey)", with HELOCs about unchanged at 40%. Those are the three months in which the median duration of unemployment went 10.7 weeks to 11.7 to 12.3, on its way to the 25.2-week peak. That's the whole problem in one sentence: a facility that shrinks as the risk rises isn't a buffer, it's a fair-weather promise.
The same correlation objection applies to the tidier version of the argument — holding the buffer inside a liquid investment account rather than in cash. It's liquid, and it earns more. It's also most likely to be down at the moment a redundancy round arrives, because layoffs and equity drawdowns share a business cycle. Selling into that is the mechanism that makes sequence risk so expensive in retirement, and it's why any argument for a smaller cash buffer rests almost entirely on how correlated a household's income is with its portfolio. For a tenured public-sector employee the correlation is close to nil. For someone paid partly in the stock of a listed employer it's close to one, which is a question about risk capacity rather than risk tolerance.
What this evidence cannot tell you
The duration data is backward-looking and heavily cycle-dependent. Six completed cycle peaks is a small sample, and the 2010 and 2021 readings were produced by two events unlikely to repeat in the same form.
The survey evidence is self-reported. SHED asks people what they'd do about a hypothetical $400 expense, and the roughly 29-percentage-point gap between its answer and the bank-record answer shows how much room sits between stated and revealed behaviour. The Chase data has the opposite problem: Chase customers aren't the population, the sample excludes the unbanked entirely, and the Weathering Volatility figures are now a decade old.
The two US datasets also don't divide into each other: the Consumer Expenditure Survey counts "consumer units" and the Survey of Consumer Finances counts "families", so putting a median $8,000 balance over a mean $78,535 of expenditures mixes a median with a mean across two sampling frames. And none of this speaks to the probability that any particular person loses their job, which is the input that would matter most and the one nobody has.
What would change the conclusion
If unemployment duration stopped being cycle-dependent, the case for sizing a buffer against the tail rather than the median weakens considerably. A world in which the median search reliably runs 9 to 11 weeks, as it did through the 1990s, is one in which three months is a defensible central estimate. The 25.2-week reading of 2010 is what makes the tail expensive to ignore.
If credit lines proved reliable in a downturn, the substitution argument gets much stronger and the sensible cash balance falls. The 2009 surveys are the evidence against, and they're one episode. A future recession that passed without banks cutting existing limits would earn that argument a serious re-hearing.
If the safety net changes, everything shifts. Almost every figure here is American, and a US answer transplanted into a country with long, earnings-related unemployment insurance and public healthcare is answering a question nobody asked.
The variables worth tracking aren't really months at all. They're the two the rule leaves out: what share of your spending survives a lost paycheck, and how correlated your income is with the assets you'd otherwise have to sell. A portfolio tracker such as LedgerTouch can show the second of those continuously; the first takes a bank statement and an afternoon.