Recession

Reviewed 12 August 2026 · Sourced from the NBER's Business Cycle Dating Committee, the BEA, the BLS and the Federal Reserve

What a recession actually is

Five months into a warehouse job, a man reads that the economy might be in a recession. He wants one thing out of that sentence: is his job about to go. He looks up what a recession is, and every page says the same thing — two consecutive quarters of falling GDP. He checks the last two quarters. Both positive. He closes the tab no better off, and the definition he just read is not the definition.

Here is the real one, from the body that dates U.S. recessions. The National Bureau of Economic Research defines a recession as “a significant decline in economic activity that is spread across the economy and that lasts more than a few months.” Three words in that sentence carry the whole test, and the committee names them: depth — the decline has to be significant, not a rounding error; diffusion — it has to be spread across the economy rather than confined to one industry; and duration — it has to last more than a few months.

Those three are not a checklist. They trade against each other, in the committee's own words: “while each of the three criteria—depth, diffusion, and duration—needs to be met individually to some degree, extreme conditions revealed by one criterion may partially offset weaker indications from another.” That sentence is why the two-month collapse of spring 2020 was designated a recession despite failing the duration test outright. The committee said so at the time: “the unprecedented magnitude of the decline in employment and production, and its broad reach across the entire economy, warrants the designation of this episode as a recession, even if it turns out to be briefer than earlier contractions.”

That is unsatisfying, and it is the honest answer. The ingredients of the judgment are all published monthly, by name — which is the second half of this page.

Why the two-quarters rule is wrong

This is the most repeated wrong fact in economics coverage, and the NBER answers it directly on its own FAQ page. It frames the question as something the press says, not something it uses: “The financial press often states the definition of a recession as two consecutive quarters of decline in real GDP.” The answer concedes the overlap and then refuses the rule: “Most of the recessions identified by our procedures do consist of two or more consecutive quarters of declining real GDP, but not all of them.”

Its reasons are structural rather than fussy. It works from monthly measures, because a quarter is a blunt instrument for locating a turning point in a month. It weighs how significant a decline is, not merely its sign — a quarter at −0.1% and a quarter at −6% are not the same event. And it does not treat GDP as the sole read on output: “in examining the behavior of production on a quarterly basis, where real GDP data are available, we give equal weight to real GDI.” Real gross domestic income measures the same economy from the income side, and the two disagree more often than people assume.

2022 is the case that settles it. On 28 July 2022 the BEA's advance estimate reported real GDP falling at an annual rate of 0.9% in the second quarter, following a 1.6% decline in the first. Two negative quarters. The word “recession” ran everywhere for a month. The NBER dated nothing, and has still dated nothing.

Then the revisions came in, which is the part nobody followed up on. The same two quarters, as first published and as the BEA series reads in August 2026:

Annual rates, seasonally adjusted. The second quarter is now positive. The two consecutive negative quarters the entire “technical recession” argument rested on do not exist in the current data. A rule that can be revised out from under you a year later is not a definition.

Who declares one, and how late

No agency of the U.S. government declares a recession. The NBER is “a private, nonprofit organization”, and it is its Business Cycle Dating Committee — a small group of academic economists — that identifies peaks and troughs in U.S. economic activity. A peak is the last month of an expansion, so it is the month the recession starts. A trough is the last month of the contraction, where the next expansion begins.

The committee dates those months retrospectively, and does not pretend otherwise: “There is no fixed timing rule because the committee waits long enough to avoid any doubt about the existence of a peak or trough.” Its own account of the lag between a turning point and its announcement is 4 to 21 months, and the 2020 recession shows both ends of that range inside one cycle:

Read the second row again. The recession ended in April 2020, and that was confirmed in July 2021. For fifteen months the honest answer to “is it over” was that nobody official would say.

Two things follow. Because the committee waits, it does not forecast — it publishes no probability of a future recession and never has. And as of August 2026 the most recent turning points in its chronology remain the February 2020 peak and the April 2020 trough. Nothing has been dated since.

What the committee actually looks at

The committee's raw material is public, monthly and free. Its determinations rest, in its words, on “a range of monthly measures of aggregate real economic activity published by the federal statistical agencies and other organizations.” The six it names are these:

Notice what dominates that list. Two of the six are employment, counted two different ways, and two more are spending. Real GDP is not on it at all, because GDP is quarterly.

Notice what is absent: the stock market, house prices, consumer confidence and the yield curve. None of them measures current activity. Markets are a forecast, and forecasts are what the committee refuses to make.

Inflation is not on the list either. A recession is defined on real activity — volumes, after prices are removed. Prices are a separate question, handled through inflation and the consumer price index, and an economy can contract while prices rise or expand while they fall.

Recession, depression, technical recession, slowdown

Four words get used interchangeably, and only one of them has a body behind it.

The depression row is worth sitting with. There is no threshold — no 10% decline, no three-year rule, no agency that certifies one. What the NBER's own table does show is scale:

Forty-three months against two, both recessions in the same chronology. The name of the episode has never been the useful information. Depth and duration are, and those are what the table holds.

The signals people watch instead, and how each fails

Because the official answer arrives late, everything used in real time is a substitute. Each has a track record, each has been wrong, and the wrong half is the part worth knowing.

The yield curve

Short-term Treasury yields paying more than long-term ones. It has the longest documented record on this list, and a documented miss: the curve inverted in 2022, stayed inverted into 2024, and the NBER has dated no recession since April 2020. The Federal Reserve Bank of San Francisco puts the historical delay between the spread turning negative and a recession beginning at six to twenty-four months — wide enough to contain most of a bull market. Full treatment, including the one documented false positive, is on yield curve inversion. What it is really pricing is rate cuts — central banks answer a downturn by cutting the federal funds rate, which is why rate expectations and recession expectations are so tangled.

The Sahm rule

A labor-market trigger rather than a forecast. It fires when “the three-month moving average of the national unemployment rate (U3) rises by 0.50 percentage points or more relative to the minimum of the three-month averages from the previous 12 months.” Its origin matters: the economist Claudia Sahm proposed it in Direct Stimulus Payments to Individuals (Hamilton Project, 2019) as an automatic trigger for sending stimulus payments, so relief would not wait on Congress. It was never built to tell an individual whether a recession had begun.

Its failure mode is documented by its own author. The rule triggered in 2024 and no NBER-dated recession followed. Writing on her own newsletter in December 2023, Sahm cautioned that “the Sahm rule is a historical pattern, and this time could easily be different,” that “even if it does trigger, a recession is not a done deal,” and — the line worth keeping — that “being data-driven is good, but being data-ridden is not.” The originator warning against mechanical use of her own indicator is the most useful sentence on this page.

Initial jobless claims

First-time unemployment filings, published weekly, the fastest labor read available. Its failure mode is noise: holidays, plant shutdowns, strikes and storms all move it, so a single week is close to meaningless. Only a multi-week direction carries information — and establishing one gives back the speed advantage that made you look.

A leading index

The best known is The Conference Board's Leading Economic Index, a proprietary composite from a private organization — not a government series, and not free. Its failure mode is on the record: its recession signal was tripped through 2022 and 2023, no NBER-dated recession followed, and by 19 August 2024 The Conference Board's own release stated that “the six-month annual growth rate no longer signals recession ahead.”

Credit spreads

The extra yield lenders demand from risky borrowers; credit markets often notice trouble before stock markets do. The failure mode is asymmetry — spreads stay tight right up until they are not, so a calm reading is an absence of evidence, not an all-clear.

One thing you will not find here: a hit rate. No percentage of the form “right 80% of the time” appears on this page. The U.S. record since 1945 holds roughly a dozen recessions, and a percentage computed on a dozen episodes is a story, not a statistic. Where a figure like the six-to-twenty-four-month range appears above, the publishing body is named beside it.

One failure mode belongs to all of them at once. The February 2020 recession was caused by a pandemic, and no spread, claims number or composite index contains information about a virus. The 2019 inversion was followed by a recession that had nothing to do with the mechanism the inversion measures. Being followed by a recession and predicting one are not the same event.

What a recession does to somebody with no cushion

For most of the people this site is written for, a recession is not a macro debate. It is the thing that takes the job first. That is not a feeling — it shows up in the published data four separate ways, and they land at different times.

1. Unemployment does not rise evenly

The aggregate unemployment rate is an average, and averages hide who absorbs the damage. The BLS publishes it broken out by educational attainment for workers 25 and over. Two recessions, seasonally adjusted:

In 2007–2009 the rate for workers without a diploma rose 8.1 percentage points against 2.8 for degree holders. In 2020 it was 15.5 against 6.5. Both groups got hurt; one got hurt two to three times as hard, and it is the group with the least saved to absorb it.

2. Hours get cut before jobs do

Layoffs are the visible event and usually not the first one. Employers trim hours first, and the BLS counts that directly: people working part time who want full-time work and cannot get it. That count roughly doubled in both recessions.

That matters more than it looks. Somebody whose hours are cut is still employed, so the unemployment rate does not move for them and they do not qualify for most things that trigger on job loss. The income is gone anyway.

3. Credit tightens exactly when it is needed

The Federal Reserve surveys senior loan officers every quarter and asks whether they have tightened standards. The net percentage tightening on credit card loans specifically:

Tightening means lower limits on existing cards, fewer approvals, and thin files declined that would have been approved a year earlier. The mechanism is risk management, not malice, and the effect is the same either way: borrowing capacity contracts on the same schedule as income. Anyone whose plan for a bad month is a credit card holds a plan that shrinks exactly when the bad month arrives.

4. A hiring freeze is invisible and it targets one group

The last mechanism has no announcement at all. When a firm stops hiring, nobody is fired, no release goes out, and the payroll number simply stops growing. The person harmed is the one who does not have the job yet — a first job, or a return to work after prison, after caregiving, after treatment, after a move. A freeze shortens nobody's paycheck. It lengthens the search, and a longer search is what somebody with no cushion cannot fund.

The phrase for this is “last hired, first fired.” It is a saying, not a statistic, which is why it is filed as unverified below. What is measurable is what the three tables show.

What this site already has on it, all free, no account: emergency fund for how many months a given cushion actually covers and where the three-to-six-month figure comes from; Stage 1 · Survive for banking and income in the next thirty days; Stage 2 · Stabilize for bill triage and the first cushion; after financial collapse if the downturn already landed; and starting over after prison, written for exactly the position a hiring freeze punishes hardest.

What is confirmed, what is not, and what is convention

Every figure on this page sits in one of three buckets. Mixing them is how a page like this goes wrong.

What trips people up

• Using the two-quarters rule. The NBER's own FAQ names it as something the financial press says, and declines to use it.
• Treating 2022 as a recession that was somehow denied. It was not dated, and the second of the two negative quarters has since been revised to +0.6%. The premise is gone, not just the conclusion.
• Waiting for the announcement. They have run 4 to 21 months late. The April 2020 trough was confirmed on 19 July 2021 — fifteen months after the recession ended.
• Expecting a forecast from the NBER. It does not make one. Published recession probabilities come from other people's models, most often the New York Fed's yield-curve model.
• Assuming a recession means a market decline, or the reverse. The committee's six monthly measures contain no asset prices at all. Markets and the business cycle run on different clocks.
• Using “depression” as if it had a threshold. It has none, from the NBER or anyone else.
• Reading the aggregate unemployment rate as everyone's experience. In April 2020 it was 14.8% overall, 8.4% for degree holders and 21.2% for workers without a high school diploma.
• Watching only for layoffs. Hours go first, and a cut schedule never shows up in the unemployment rate. Involuntary part-time work went from 4.38 million in February 2020 to 10.88 million that April.
• Counting a credit card as the plan. Net card tightening hit 58.8% of banks in late 2008 and 71.7% in the third quarter of 2020.
• Trusting one signal because it worked last time. Every one on this page has a documented miss, which is why the site's economic page prints five with the failure mode next to each.

The figures on this page are checked against the source that publishes them, and dated. Published rates move after the release named above — the linked source always carries the current number. This page explains a term; it does not recommend a product.