Key takeaways
- Since 1948 the S&P sat 20% or more below its prior peak in 20.4% of months, but in 39.8% of months when unemployment had risen a full point.
- US equities returned 6.6% a year in real terms from 1900 to 2024 against 0.5% on Treasury bills, a 5.9 times gap compounded over 30 years.
- JPMorgan Chase Institute data show families need 6.2 weeks of income for a simultaneous income dip and spending spike, an event with 1.5% monthly odds.
- Median US unemployment ran 11.0 weeks in June 2026, yet 27.3% of the unemployed had been out of work for 27 weeks or more.
- 55% of US adults held three months of rainy-day savings in 2024, while 42% of UK adults could not cover three months of expenses.
Two colleagues are made redundant in the same week. Both had six months of spending set aside. One kept every penny of it in a savings account. The other had moved half of it into a global equity fund three years earlier and then stopped thinking about it. Which of them made the better call?
You can't say yet, and neither can they, because the answer turns on something neither of them chose: what the market happened to be doing in the month the letter arrived. That's the whole argument in one sentence. It's worth taking apart slowly.
Start with what the cautious colleague gave up. On the 125-year US record, cash compounds at about 0.5% a year after inflation. Equities compound at 6.6%. Over three decades that gap multiplies the money by roughly 5.9 times. Leave six months of spending in cash for thirty years and you end up with about seven months of spending in Treasury bills — or about 41 months of it in equities.
That's the case for putting part of the fund to work. The case against is just as concrete. Job loss isn't random. It clusters in the same months that equity markets are down, which is exactly when selling hurts most.
Both claims are true. So the useful question isn't which one wins. It's how much each one costs the other, and the answer rests on two datasets anyone can check.
What the buffer is actually absorbing
Picture the year you've just had rather than the disaster you're insuring against. Most household shocks are small and frequent, and they arrive on a schedule you'd half recognise.
The JPMorgan Chase Institute studied monthly income, spending and balances for more than six million Chase families between 2013 and 2018. In a given month, 23% of families saw an expenditure spike and 11% saw an income dip. Those events arrive about every four months and every nine months. Each needs less than three weeks of take-home income to absorb, at 2.6 weeks and 2.8 weeks respectively.
The rare event is the one people picture when they say "emergency". A simultaneous income dip and expenditure spike happened in 1.5% of family-months, or about once every five and a half years. That one needs 6.2 weeks of income, and 65% of families didn't have it.
Split your fund along that line and the investing question changes shape. The first three weeks of income get touched several times a decade, so they can't sit in anything that moves. The next several months get touched rarely, and for plenty of households never. That's the slice where the argument actually lives. The companion piece on how income volatility changes the right emergency fund size handles the sizing question, so this one sticks to the part you don't touch.
The opportunity cost is real and it's large
If you want to check the return side yourself, here's exactly what it's built from. The Dimson-Marsh-Staunton database, published each year in the UBS Global Investment Returns Yearbook, tracks US asset returns from 1900 to 2024. Across those 125 years US equities returned 9.7% a year in nominal terms, long bonds 4.6% and Treasury bills 3.4%, against inflation of 2.9%. A dollar in equities grew to $107,409. In bills it grew to $67, with the inflation index ending at $37.
So bills raised purchasing power about 1.8 times over 125 years. Equities raised it roughly 2,900 times. That's the opportunity-cost case in two numbers.
Now walk your own money through it. Say you size your buffer off the national average: ONS Family Spending puts average weekly household expenditure at £676.60 in the year to March 2025, so six months of it is about £17,600. That figure is an illustration, but the arithmetic on it isn't. Held for 30 years at the historic real bill rate, your £17,600 is worth roughly £20,400 in today's money. Held at the historic real equity rate, roughly £120,000. The distance between those two numbers is what the certainty costs you. The same compounding that makes a one-point fee difference so expensive over 30 years works on cash drag too.
Two caveats, and they pull in opposite directions. This century has been worse for cash than the long record implies. Measured from 2000, global equities returned 3.5% a year in real terms with an equity risk premium over bills of 4.3%, which implies bills lost roughly 0.8% a year in real terms. And nobody holds a fixed emergency fund untouched for 30 years. Yours gets spent, rebuilt and resized as your income changes, so treat that £120,000 as the outer edge of the argument rather than a forecast.
The counter-argument: the buffer fails when it's needed
Here's the objection, stated as strongly as it deserves. You draw on an emergency fund when your income stops. Income stops most often in recessions. Equity markets fall in recessions. So a partly invested buffer is smallest exactly when it's called on, and you're the one forced to sell near a low.
Is that true, or does it just sound true? It's testable. Robert Shiller's monthly S&P composite price index combined with the BLS unemployment rate gives 921 months from January 1948 to September 2024. For each month I measured how far the index sat below its previous peak, and how much the unemployment rate had risen over the prior twelve months.
Across all 921 months, the index was 20% or more below its prior peak in 20.4% of them. In the 166 months where unemployment had risen at least a full point year on year, that share was 39.8%. For a 10% drawdown the two figures are 40.0% and 56.6%. For a 30% drawdown, 8.4% and 20.5%.
Average drawdown across all months was 10.6%. In the rising-unemployment months it was 17.5%. The odds of sitting within 5% of a high fell from 49.8% to 33.7%.
So the objection is right in direction and roughly right in size. A deep drawdown was about twice as likely when the labour market was deteriorating. That isn't a weak correlation you can wave away.
Where the objection overstates itself
Twice as likely still leaves most bad labour-market months without a 20% drawdown. In 60% of them the index was less than a fifth below its peak. And the timing is messier than the story implies, because equity markets usually turn before the labour market does.
Imagine you'd been laid off at the worst moment available in each postwar downturn — the exact month unemployment peaked. Where would your invested slice have been sitting?
| Episode | Unemployment peak | Rate | Index vs prior peak |
|---|---|---|---|
| 1974-75 | May 1975 | 9.0% | -23.9% |
| 1980-82 | Nov 1982 | 10.8% | 0.0% |
| 1990-92 | Jun 1992 | 7.8% | -1.9% |
| 2001-03 | Jun 2003 | 6.3% | -33.5% |
| 2008-10 | Oct 2009 | 10.0% | -30.7% |
| 2020 | Apr 2020 | 14.8% | -15.7% |
In 1982 and 1992 the market had already recovered by the time unemployment peaked, so you'd have sold at no discount at all. In 2003 and 2009 it hadn't, and your buffer would have been down about a third. The 2020 row understates the damage, because a monthly average price smooths the intramonth low.
There's a pattern underneath. The market low came one to seven months before the unemployment peak in five of the six episodes. The exception was 1990-92, where the gap ran 20 months. Anyone laid off on the way down still sells into weakness. But these are two series that move at different times, not a single switch that fails.
How long the money has to last
Duration decides how much of a drawdown you actually realise. So how long do you think you'd be out of work? In June 2026 the median US unemployment spell ran 11.0 weeks, with the unemployment rate at 4.2%. A cash slice of a few weeks plus a modest buffer covers that. But 27.3% of the unemployed had been out of work for 27 weeks or more, and that tail is where a buffer drains to nothing.
The Federal Reserve's 2024 household survey found that 68% of prime-age adults laid off in the prior twelve months were working again by the time they responded. Most layoffs resolve inside a year. Market recoveries often don't. The Yearbook dates the full recovery from the 1929 crash to February 1945, fifteen and a half years. Real US equity values stayed underwater for over ten years after the 1973-74 fall, seven and a half years after the 2000 peak, and four years after the February 2009 low. The study of how long bear market recoveries take in nominal and real terms traces those windows in detail.
That asymmetry is the sharpest version of the problem. Your unemployment spell usually runs months. A drawdown recovery often runs years. Selling assets to cover the rent at a depressed price is the same mechanism that makes the sequence of returns matter so much for retirement withdrawals, applied to a working-age household instead of a retired one.
What households actually hold
All of this assumes you have a fully funded buffer and are only deciding where to park it. Most people aren't there. The Federal Reserve's Survey of Household Economics and Decisionmaking found 55% of US adults had set aside three months of expenses in a rainy-day fund in 2024. Thirty percent said they could not cover three months by any means.
A widely repeated line holds that 37% of Americans can't cover a $400 emergency. That misreads the survey. Sixty-three percent said they'd cover it exclusively with cash, and most of the rest would use another method, usually a credit card carried as a balance. Only 13% said they couldn't pay it at all.
The UK picture is similar. The FCA's Financial Lives survey, fielded in May 2024, found 42% of UK adults could not cover living expenses for three months or more if they lost their main income. Ten percent had no cash savings at all, and another 21% held under £1,000. Nine in ten adults had some cash savings, at a median of £5,000 to £6,000.
Income variability compounds the problem. Twenty-nine percent of US adults said their income varied at least occasionally from month to month in 2024, rising to 59% among the self-employed. If that's you, your buffer does frequent work rather than tail work, and the frequently used part has no business carrying market risk.
The return on cash isn't only financial
There's a column in this ledger that a return comparison never shows you. Vanguard surveyed its clients in July 2024 and published the results in April 2025. Emergency savings were the strongest predictor of financial well-being in that sample. Holding at least $2,000 was associated with a 21% higher well-being score than holding none. Three to six months of expenses on top added a further 13%.
The behavioural figures are starker. Clients without emergency savings reported spending 7.3 hours a week thinking about money, against 3.7 hours for those holding at least $2,000. At work they lost 6.1 hours a week to financial stress, against 1.5 hours. And 51% reported rising financial stress year on year, against 15%.
What's an hour a week of not thinking about money worth to you? These are associations from one firm's client base, not causal estimates, and the sample skews wealthy. They still price something a return comparison misses. A buffer that can fall 30% isn't the same product as one that can't, even when its expected value is higher.
What would change the conclusion
Four things would move it.
Real cash yields come first. That 0.5% historic real bill return averages long stretches of negative and positive. If short rates settle above inflation for a decade, the opportunity cost shrinks and the case for investing weakens. If they sit below inflation, it widens.
The correlation estimate is second. It uses one country, one index and one measure of labour-market stress, starting in 1948. A different window, a different market, or unemployment insurance that replaces more income would move it. If you disagree with the 39.8% figure, there's a clear test available: re-run it on UK data.
Access to credit is third. Say you have a drawable mortgage offset, a cheap credit line, or a partner on stable pay. You're solving a different problem, because your buffer only has to bridge to the next source rather than fund the whole spell.
Fourth is how closely your job tracks the market. Think of someone paid partly in shares by a listed employer, or working in a deeply cyclical sector. They already carry market risk inside their human capital, so the conditional probability above understates their danger — their own job loss is tied more tightly to their own holdings than the aggregate figures suggest. The distinction between risk capacity and risk tolerance is the frame that catches it.
Where the evidence leaves it
Back to the two colleagues. Neither of them was obviously wrong, which is the honest answer and an unsatisfying one.
The data supports a split rather than a single verdict. The frequently drawn part of a buffer meets shocks every four to nine months, which is far too often for market risk to earn its 6% a year. The rarely drawn part meets them about once every five and a half years, and pays a compounding cost of roughly 5.9 times over thirty years to avoid a drawdown that historically showed up in about 40% of the months when it might have been called on.
Neither the size of that opportunity cost nor the strength of that correlation is seriously disputed. What's disputed is which one dominates. That turns on how deep your buffer is, how stable your income is, and how long you'd have to live off it.