What a Recession Is and Who Decides One Has Started

In the United States, a recession is the period between a peak and a trough in economic activity as dated by the Business Cycle Dating Committee of the National Bureau of Economic Research (NBER), a private, nonprofit research organization. The committee judges it from monthly income, jobs, spending and production data, not from a single GDP rule, and announces its dates only afterward: in the six recessions since 1980, the start was confirmed 4 to 12 months after it happened and the end 8 to 21 months after.

Key takeaways

The NBER defines a recession as a significant decline in economic activity that is spread across the economy and lasts more than a few months; depth, diffusion and duration must each be met to some degree, and an extreme reading on one can partly offset another.
Peaks and troughs are dated to months, mainly from real personal income less transfers and nonfarm payroll employment, with four other monthly series and quarterly GDP and GDI as cross-checks.
Two consecutive quarters of falling real GDP is a press shortcut, not the NBER test: the 2001 recession had no two such quarters in current BEA data, and the two early-estimate declines of 2022 were not a recession.
Since 1980, the recession's end was announced a median 15 months after it occurred; in 1990, 2001 and 2020 the start was confirmed only in or after the month the recession later turned out to have ended.
As of October 13, 2026, the NBER's latest announcement is still the July 19, 2021 call of the April 2020 trough; no new peak has been announced.

The NBER definition, word for word

The committee's dating procedure FAQ states the definition it has used for decades:

"a significant decline in economic activity that is spread across the economy and that lasts more than a few months" — NBER, Business Cycle Dating Procedure FAQ

Each phrase maps to a test: depth (how far activity fell), diffusion (how widely the decline spread) and duration (how long it lasted). The committee says each criterion must be met "to some degree," but that extreme conditions on one may partially offset weaker readings on another.

The 2020 downturn is the case that shows the trade-off. Activity fell for only two months, February to April, far short of the six months of the first half of 1980, the previous shortest recession. In its July 19, 2021 trough announcement, the committee explained that the unprecedented size of the fall in employment and production, and its reach across the whole economy, justified calling it a recession despite its brevity.

By the NBER's counting convention, the first month of a recession is the month after the peak and the last is the trough month, so a peak in February and a trough in April make a two-month recession.

Who sits on the dating committee

No US government agency dates recessions. The NBER's FAQ says there is no alternative chronology compiled or published by the government; its dates became the de facto reference in 1961, when the Commerce Department began printing them in its Business Cycle Developments publication without formally endorsing them. The committee in its current form was set up in 1978.

The NBER's president appoints the members. According to the NBER's committee page, the eight current members are Valerie Ramey of Stanford (chair since 2024), Kristin Forbes of MIT, Robert J. Gordon of Northwestern, James Poterba of MIT, Christina and David Romer of the University of California, Berkeley, James Stock of Harvard and Mark W. Watson of Princeton. The NBER says there is no fixed timing rule for its announcements and no fixed formula for weighting the data.

The indicators behind the dates

The NBER's business cycle dating page lists six monthly measures of real (inflation-adjusted) economy-wide activity used to pick peak and trough months:

Real personal income less transfers: income from work, businesses and assets, with government payments such as unemployment benefits removed because they do not come from production.
Nonfarm payroll employment: jobs counted in the Bureau of Labor Statistics (BLS) survey of employers.
Employment from the BLS household survey, which, unlike the payroll count, excludes people who are paid but on furlough, as the committee noted in 2020.
Real personal consumption expenditures, the monthly measure of household spending.
Manufacturing and trade sales adjusted for price changes.
Industrial production, which tracks factory, mining and utility output.

There is no fixed weighting, but the committee says that in recent decades it has put the most weight on the first two. It also makes a separate quarterly call using real gross domestic product (GDP, measured from spending) and real gross domestic income (GDI, measured from incomes), giving the two equal weight. The St. Louis Fed hosts a FRED dashboard that collects the six monthly series.

Month and quarter can differ. In its June 8, 2020 announcement, the committee dated the monthly peak to February 2020 but the quarterly peak to the fourth quarter of 2019, because March 2020 collapsed so sharply that first-quarter GDP, GDI and employment all averaged below the fourth quarter's levels. In the 2007–2009 recession, it dated the trough to June 2009 even though payroll employment kept falling until eight months later.

Why the verdict arrives so late

The committee describes its approach as retrospective. It waits until it is confident a peak or trough has occurred and until standard data revisions have come in, so that it will not have to move a date later.

The table below lists the last eight NBER recessions. Announcement lags are the months between the turning-point month and the announcement month, taken from the NBER announcements and matching the elapsed-month counts in its FAQ. The unemployment column shows the monthly BLS unemployment rate (FRED series UNRATE) in the peak month and its highest level before the next recession began.

Peak	Trough	Length (months)	Peak announced (lag)	Trough announced (lag)	Unemployment at peak → high
Dec 1969	Nov 1970	11	Before the committee existed	Before the committee existed	3.5% → 6.1% (+2.6)
Nov 1973	Mar 1975	16	Before the committee existed	Before the committee existed	4.8% → 9.0% (+4.2)
Jan 1980	Jul 1980	6	Jun 3, 1980 (5)	Jul 8, 1981 (12)	6.3% → 7.8% (+1.5)
Jul 1981	Nov 1982	16	Jan 6, 1982 (6)	Jul 8, 1983 (8)	7.2% → 10.8% (+3.6)
Jul 1990	Mar 1991	8	Apr 25, 1991 (9)	Dec 22, 1992 (21)	5.5% → 7.8% (+2.3)
Mar 2001	Nov 2001	8	Nov 26, 2001 (8)	Jul 17, 2003 (20)	4.3% → 6.3% (+2.0)
Dec 2007	Jun 2009	18	Dec 1, 2008 (12)	Sep 20, 2010 (15)	5.0% → 10.0% (+5.0)
Feb 2020	Apr 2020	2	Jun 8, 2020 (4)	Jul 19, 2021 (15)	3.5% → 14.8% (+11.3)

Lengths and dates from the NBER's expansions and contractions table and announcements list. Unemployment rates are seasonally adjusted, as published by FRED on October 2, 2026; changes in percentage points.

Across the six recessions with formal announcements, the median lag was 7 months for the peak and 15 months for the trough. Troughs take longer because the committee wants evidence that a recovery is lasting. In three of the six cases the recession was over, or ending, by the time its start was confirmed: the 1990 peak was announced a month after the March 1991 trough, the 2001 peak was announced in November 2001, the trough month itself, and the 2020 peak was announced two months after the April 2020 trough.

Two negative quarters: a shortcut

A common rule of thumb says a recession is two consecutive quarters of falling real GDP. The NBER's FAQ says most of its recessions do include such a stretch, but not all, and gives four reasons it does not use the rule: it looks at a range of indicators rather than GDP alone; it requires a significant decline, so two small dips need not qualify; its main chronology is monthly; and in quarterly data it weights GDI equally with GDP.

Where the shortcut and the dates diverged

2001, a recession without two negative quarters. In the Bureau of Economic Analysis (BEA) data as published on FRED on September 30, 2026, real GDP growth alternated: −1.3% in the first quarter of 2001, +2.5% in the second, −1.6% in the third and +1.1% in the fourth (annualized rates). Real GDI, by contrast, fell in each of the last three quarters of 2001, the kind of gap the NBER cites when it says GDI matters.
2022, two negative quarters without a recession. BEA's advance estimate of July 28, 2022 showed real GDP falling at annual rates of 1.6% in the first quarter and 0.9% in the second. Meanwhile, from December 2021 to June 2022, nonfarm payrolls (FRED series PAYEMS) rose by about 2.5 million and unemployment fell from 3.9% to 3.6%, although real personal income less transfers slipped about 1.7%. The NBER dated no recession, and current BEA data now show the second quarter at +0.6% annualized, so the two-quarter decline no longer exists in the revised figures.

Earlier history has more mismatches. In current BEA data, real GDP alternated through the 1960–61 recession (−2.1%, +2.0%, −5.0% annualized from the second to the fourth quarter of 1960), and it fell in the second and third quarters of 1947 (−1.0% and −0.8%) without an NBER recession.

Reading a quarterly GDP figure

BEA headlines quarterly growth at an annual rate: the pace GDP would change over a year if the quarter's change repeated four times. Its FAQ on growth rates describes the compound formula:

annualized rate = (GDP_t / GDP_(t-1))^4 − 1
GDP_t      = real GDP in the latest quarter
GDP_(t-1)  = real GDP in the previous quarter
4          = number of quarters in a year

The second quarter of 2020 shows why the distinction matters. Real GDP (FRED series GDPC1) fell from $20,709 billion to $19,078 billion in chained 2017 dollars, a 7.9% drop within the quarter. Compounded four times, that is the −28.0% annual rate in current data. BEA's advance estimate of July 30, 2020 first put it at −32.9%; revisions since then account for the difference.

Faster signals and other countries

Analysts watch faster signals such as the Sahm rule, built from the unemployment rate, and an inverted yield curve; neither is on the NBER's indicator list. Other economies use their own conventions. The UK's Office for National Statistics says two consecutive quarters of falling GDP, often called a "technical recession," has no official standing, and the ONS itself avoids declaring recessions. In the euro area, a committee run by the Centre for Economic Policy Research (CEPR) and the Euro Area Business Cycle Network dates turning points by quarter; in March 2021 it still could not say whether the recession that followed the fourth-quarter 2019 peak had ended.

What past recessions did to jobs

For households, a recession shows up mainly as lost jobs and lower real earnings. The table shows how two of the NBER's own indicators changed from peak month to trough month, and when unemployment topped out. These are historical measurements, not a guide to any future downturn.

Recession	Payroll jobs, peak to trough	Real income less transfers, peak to trough	Unemployment high (months after trough)
1969–70	−1.2% (−0.8 million)	+0.5%	6.1%, Dec 1970 (1)
1973–75	−1.6% (−1.3 million)	−5.0%	9.0%, May 1975 (2)
1980	−1.1% (−1.0 million)	−1.5%	7.8%, Jul 1980 (0)
1981–82	−3.1% (−2.8 million)	−0.1%	10.8%, Nov 1982 (0)
1990–91	−1.1% (−1.2 million)	−2.3%	7.8%, Jun 1992 (15)
2001	−1.2% (−1.6 million)	−1.7%	6.3%, Jun 2003 (19)
2007–09	−5.3% (−7.4 million)	−6.0%	10.0%, Oct 2009 (4)
2020	−14.4% (−21.9 million)	−8.2%	14.8%, Apr 2020 (0)

Payroll jobs from FRED series PAYEMS; real personal income less transfers from FRED series W875RX1 (chained 2017 dollars), as published on FRED on October 2 and September 30, 2026.

The damage varied widely, from about 1 million payroll jobs in 1980 to nearly 22 million in 2020. Unemployment often kept rising after the official end: 15 months after the 1991 trough and 19 months after the 2001 trough. And no single indicator tells the whole story; real income less transfers barely fell in 1981–82, when payrolls dropped by 2.8 million, and in current data it rose slightly across the 1969–70 recession.

Frequently asked questions

Has the NBER ever changed a recession date?

Not since the dating committee was created in 1978; none of its announced peaks or troughs has been revised. Before 1978 some turning points were revised, and the NBER says it would change a date if it concluded the earlier choice was wrong. That is one reason it waits for data revisions before announcing.

Is a depression a different category?

The NBER does not identify depressions separately; every downturn in its chronology is a recession. The episode generally regarded as a US depression contains two NBER recessions: August 1929 to March 1933, which lasted 43 months, and May 1937 to June 1938. The NBER describes the contraction that began in 1929 as widely acknowledged to be the worst in US history.

What does the NBER do with a double-dip recession?

It has no special category for one. Two close contractions are recorded either as two separate recessions or as one, depending mainly on how long and how strongly activity recovered in between. The committee treated 1980 and 1981–82 as separate recessions partly because major indicators bounced back in late 1980 and early 1981.

Does a recession end when output gets back to its old level?

No. The trough marks the month activity stopped falling, and the expansion starts the next month even if activity is still far below the previous peak. After the June 2009 trough, real personal income less transfers did not pass its earlier peak until July 2011, and payroll employment not until May 2014.

How long do US recessions usually last?

In the NBER's table, the 12 recessions from 1945 to 2020 averaged 10.3 months, against an average expansion of 64.2 months. The longest in that period was the 18-month 2007–2009 recession and the shortest was the 2-month recession of 2020. The full record since 1854 averages 17.0 months per contraction.

Checking the status yourself

As of October 13, 2026, the NBER's announcements page still ends with the July 19, 2021 trough call, and its dating page names February 2020 as the most recent peak and April 2020 as the most recent trough. In the latest data on FRED, real GDP grew at a 2.2% annual rate in the second quarter of 2026 and unemployment was 4.2% in September 2026. A short routine covers the official picture:

Check the NBER announcements list for a new peak date; the NBER offers email alerts, which require a free account.
Scan the six monthly indicators on the FRED dashboard, starting with real personal income less transfers and payroll employment, the two the committee weights most.
When a GDP report comes out, note whether the headline is annualized, compare it with GDI, and expect revisions; a single negative quarter, or even two, is not a recession call.
Treat any announced peak as a date in the past; on the record since 1980, the recession may be over by the time it is official.