Almost everything written about recession indicators still claims the yield curve has predicted every US recession without exception, and that the Sahm Rule has never produced a false signal. As of 2026, neither claim survives.
Both broke in the same cycle. The yield curve inverted and stayed inverted for longer than at any point in modern records. The Sahm Rule triggered on a weak jobs report. No recession followed either signal, and enough time has now passed that every historical lead time has expired.
That changes how much weight these signals deserve from here.
Who actually declares a recession
Start here, because the popular definition is wrong and it matters.
You have heard that a recession is two consecutive quarters of falling GDP. That is a journalistic rule of thumb with no official standing in the United States. Nobody who dates recessions uses it.
The body that does is the Business Cycle Dating Committee of the National Bureau of Economic Research, a private non-profit. Its definition is a significant decline in economic activity, spread across the economy, lasting more than a few months. It weighs three things: depth, diffusion and duration, and notes that extreme severity in one can compensate for weakness in another.
It looks at real personal income less transfers, non-farm payroll employment, household-survey employment, real personal consumption expenditure, real wholesale and retail sales, and industrial production. GDP is considered and it is not decisive, partly because it is quarterly and heavily revised.
The committee announces late, on purpose. It waits for data revisions rather than being fast. The 2020 recession is the clearest case. The committee announced on 8 June 2020 that a recession had begun in February 2020. By then the recession had already ended, in April, though that end date was not announced until July 2021.
So the official arbiter told the public a recession had started roughly two months after it had finished. Whatever you do with recession information, it cannot be waiting for the declaration.
The yield curve, and the record it lost
The yield curve compares what the government pays to borrow for a long period against a short one. Normally long-term rates are higher, because lending for longer is riskier. When short-term rates exceed long-term rates, the curve is inverted, and that has historically been the single best recession signal available.
The two spreads people watch are the ten-year against the two-year, and the ten-year against the three-month. Research at the Federal Reserve has generally favoured the three-month version.
Across the six completed signals from 1978 to 2019, the record was close to perfect and the lead time was long: a median of roughly 15 months from first inversion to the start of the recession, and an average of about 14. That range, very roughly a year to two years, is what made it useful and what makes it easy to misuse, because a signal that can be right eighteen months early is indistinguishable in real time from one that is wrong.
Then came the episode that ended the run. The ten-year against three-month spread inverted on 25 October 2022 and did not return to normal until 13 December 2024. That is longer than any inversion in modern records, exceeding the previous longest, in 1980 to 1982, by a wide margin. It was also the deepest since 1981, at around 108 basis points.
No recession followed. Not during the inversion, and not in the period afterwards during which every historical lead time would have expired.
The indicator is not worthless. The sample was always tiny. Six or seven signals is not a track record in any statistical sense, and a perfect score across seven events was never as impressive as it sounded. What the last few years established is that an inverted curve can reflect expectations of falling inflation rather than expectations of contraction, which are very different things with the same shape.
The Sahm Rule, and why its own creator doubted it
The Sahm Rule is more recent and was designed for a different job. Claudia Sahm, then an economist at the Federal Reserve, built it to identify a recession that has already begun, in real time, so that support payments could be triggered automatically without waiting for a committee.
It is admirably simple. Take the three-month moving average of the unemployment rate. Compare it with the lowest three-month average of the previous twelve months. If it has risen by 0.50 percentage points or more, the rule triggers.
The logic is that unemployment does not drift upwards gently. Once it starts rising it tends to keep rising, because job losses reduce spending, which costs other people their jobs. A half-point rise has historically meant the process had started.
In August 2024, following a weak July jobs report, the rule triggered at 0.53 percentage points. There was no recession.
Sahm herself said publicly that she did not think the economy was in one, and explained why her own rule was likely to be wrong that time: the rise in unemployment was driven substantially by more people entering the labour force rather than by people losing jobs. The rule cannot tell those apart. It sees the unemployment rate going up.
It had also produced a false positive before, in November 1976, with no recession following. Two misfires against a handful of correct calls is a rather different proposition from the perfect record the rule is often credited with.
The indicators worth watching anyway
None of this makes the data useless. It makes no single series a forecast.
| Indicator | What it measures | What it is good for | Where it fails |
|---|---|---|---|
| Yield curve | Long rates minus short rates | Long lead, roughly one to two years | Inverted for 26 months to 2024 with no recession |
| Sahm Rule | Rise in unemployment from its recent low | Real-time identification, not prediction | Cannot separate job losses from a growing labour force |
| Initial jobless claims | New unemployment filings each week | Weekly, barely revised, genuinely current | Noisy, and moves late in the cycle |
| ISM manufacturing PMI | Survey of purchasing managers | Fast, and turns early | Below 50 for long stretches with no recession |
| Leading Economic Index | Ten forward-looking series combined | Designed for exactly this | Fell for around two years to 2024 while the economy grew |
| Credit spreads | Extra yield on risky corporate debt | Reacts fast when funding tightens | Widens in market panics with no recession |
The last few years broke several of them at once and in the same direction, which is not coincidence. The post-pandemic economy featured an unusual combination of falling inflation, a growing labour force and a manufacturing slowdown alongside a strong service sector, and most of these indicators were calibrated on an economy that did not look like that.
Eleven recessions is too small a sample
There have been eleven US recessions since 1950. Eleven. An indicator that has correctly signalled eight of them is working from a sample that would not support a conclusion in any other field.
Worse, the eleven are not comparable events. A recession caused by an oil embargo, one caused by deliberately raising interest rates to 20% to break inflation, one caused by a housing and banking collapse, and one caused by governments closing the economy for a public health emergency have almost nothing structurally in common. Fitting a rule across all four and expecting it to catch the fifth is optimistic.
So when you see a claim that something has predicted every recession, translate it: this has happened seven or eight times, and the thing worked most of them. It is not a forecast.
How to prepare without forecasting
The useful response to recession risk is not prediction. It is arranging your finances so the forecast does not matter.
Cash is the whole ballgame. The damage a recession does to a household is concentrated in job loss, and the difference between a bad year and a catastrophe is whether you can pay for several months without income. Three to six months of expenses does more than any indicator will.
Your job security is your real exposure. Far more than your portfolio. Somebody with tenure and a diversified portfolio is less exposed than somebody with a volatile commission income and the same investments. That assessment is specific to you and no economist can make it.
Do not de-risk a portfolio on a forecast. Anyone who sold on the 2022 inversion sat out a substantial rise waiting for a recession that did not arrive. The cost of acting on a false signal is real, and it is paid immediately.
Fix borrowing costs where you can. Variable-rate debt is the channel through which macroeconomic changes reach a household budget fastest.
Watch claims, not commentary. If you follow one series, initial jobless claims are weekly, barely revised, and describe something concrete. Most of what is written about the others is written to be published rather than to be right.
Where this fits with everything else
For the market half of this, including why a falling market is not the same thing as a recession, see bear market and bull market, and how to tell which one you are in.
If your balance has already fallen, what to do when your 401(k) is losing money deals with that directly.
And if a redundancy round is the specific thing you are worried about, what happens to your 401(k) if your company goes bankrupt or is acquired covers a rule that can make unvested employer contributions entirely yours.
Where to check the data yourself
Every series here is published free by the government or the Federal Reserve, updated automatically, and chartable without an account. Anybody quoting these numbers at you is reading the same pages.
- Ten-year minus three-month Treasury spread, from the Federal Reserve Bank of St. Louis, with recessions shaded so the lead times and the 2022 inversion are both visible. Search the same site for T10Y2Y, SAHMREALTIME and ICSA to chart the other series here.
- NBER business cycle dates, the official start and end of every US recession since 1854, alongside the announcement dates that show how far behind the declarations run.
Frequently asked questions
Is a recession two quarters of negative GDP?
Not in the United States. That is a media shorthand with no official standing. The NBER defines a recession as a significant, widespread decline in activity lasting more than a few months, and it weighs employment, income, spending, sales and production alongside GDP.
Does an inverted yield curve mean a recession is coming?
It used to be the best available signal, with a median lead of about 15 months across six signals from 1978 to 2019. The 2022 inversion was the longest and deepest in modern records and no recession followed, so the perfect record no longer holds.
What is the Sahm Rule?
It triggers when the three-month average unemployment rate rises 0.50 percentage points above its lowest three-month average of the previous year. It was designed to identify a recession already underway rather than to predict one, and it triggered in August 2024 without a recession following.
Who officially declares a recession?
The Business Cycle Dating Committee of the National Bureau of Economic Research, a private non-profit rather than a government body. It announces well after the fact: the start of the 2020 recession was announced in June 2020, by which point the recession had already ended.
Which recession indicator is most reliable?
None reliably. Every one of them is fitted to eleven post-1950 recessions with very different causes, which is far too small a sample for confidence. Initial jobless claims are the most current and least revised, which makes them the most useful for seeing what is happening now rather than guessing what happens next.
How should I prepare for a recession?
Cash reserves covering several months of expenses, an honest assessment of how secure your own income is, fixed rather than variable borrowing where possible, and no change to a long-term portfolio on the basis of a forecast. The last of those is the one that most often costs people money.
Why did so many indicators fail recently?
Largely because the post-pandemic economy combined falling inflation with a growing labour force and a weak manufacturing sector alongside a strong service sector. Most of these measures were calibrated on economies that did not look like that, and several were misled in the same direction at the same time.