What a Recession Is, and Why Nobody Can Tell You When One Starts

The official call comes from a committee of economists, months after the fact, and "two quarters of falling GDP" is a rule of thumb it doesn't use. Before the call there are warning lights. Since 2022 all three have gone off, and no recession followed.

In one sentence

A recession is a significant decline in economic activity that spreads across the economy and lasts more than a few months, as dated, month by month and long after the fact, by the Business Cycle Dating Committee of the National Bureau of Economic Research, a private nonprofit.

No law defines the word and no agency declares one. The committee judges three things: how deep the decline is, how widely it spreads and how long it lasts, and a lot of one can partly make up for less of another. It watches six monthly measures: real personal income less government transfers, payroll jobs, employment from the household survey, real consumer spending, inflation-adjusted manufacturing and trade sales, and industrial production. There is no formula. Lately it has leaned most on income and payroll jobs. The peak it names is the last month of the expansion; the trough is the last month of the recession.

New articles and charts arrive in one short email on Monday mornings. Get the weekly digest

The analogy

A drought. One dry month is just a dry month. Nobody can say a drought began in May until June, July and August have also come up dry, and the official declaration arrives after farmers have already lost a crop.

The committee works the same way. It waits until a decline is deep, broad and long enough that later revisions won't undo it, which takes months:

RecessionPeakPeak announcedTroughTrough announced
1981–82July 1981Jan. 6, 1982 (6 months later)Nov. 1982July 8, 1983 (8 months)
1990–91July 1990Apr. 25, 1991 (9 months)Mar. 1991Dec. 22, 1992 (21 months)
2001Mar. 2001Nov. 26, 2001 (8 months)Nov. 2001July 17, 2003 (20 months)
2007–09Dec. 2007Dec. 1, 2008 (12 months)June 2009Sept. 20, 2010 (15 months)
2020Feb. 2020June 8, 2020 (4 months)Apr. 2020July 19, 2021 (15 months)

Twice, in 1991 and 2020, the recession was already over when its start was announced. In 2001 the announcement came in what turned out to be its last month. The almanac's gray recession bars come from this record, so they always arrive late.

Why two quarters isn't the rule

"Two quarters in a row of falling GDP" is a newspaper rule of thumb, not the definition, and the committee says plainly that it doesn't use it. It usually agrees with the committee, but not always, in both directions.

Recessions without two down quarters. In today's data, 10 of the 12 recessions since 1948 had at least two consecutive quarters of falling real GDP. The two that didn't are 1960–61 and 2001. In 2001 output fell, rose, fell and rose: −1.3%, +2.5%, −1.6% and +1.1% at annual rates. Jobs were falling every month, and the committee called it.

Two down quarters without a recession. In 2022 the first estimates showed GDP falling at a 1.4% rate in the first quarter and, by late July, 0.9% in the second, with the first revised to −1.6%. Over the same six months employers added about 2.5 million jobs and unemployment fell from 4.0% to 3.6%. Much of the first-quarter drop came from a jump in imports, weaker exports and businesses rebuilding inventories more slowly, items that swing GDP without telling you much about jobs or spending. Revised data now show the first quarter at −1.0% and the second at +0.6%: by current figures, there were never two. The committee never called a 2022 recession.

GDP has two other problems as a recession clock. It comes quarterly and is revised for years (see Revisions and vintages), and it measures output from the spending side only. The committee gives equal weight to the income side, gross domestic income, which often tells a different story in the same quarter.

How it reaches your kitchen table

Jobs, first and longest. A recession is felt mostly as layoffs, and unemployment keeps rising after the recession has officially ended. It went from 5.0% to 10.0% in 2007–09, peaking in October 2009, four months after the trough. In 2001 it went from 4.4% to 6.3%, peaking in June 2003, a year and a half after the recession ended. In 2020 it jumped from 3.5% to 14.8% in two months.

Raises, a year or two later. Pay growth slows as job seekers outnumber openings. Hourly pay was rising about 3.8% a year in 2008; by October 2012 it was 1.2%.

Prices and rates: down. Recessions usually cool inflation, sometimes past zero. Consumer prices were up 5.5% from a year earlier in July 2008 and down 2.0% a year later. The Fed cuts: from 5.25% in mid-2007 to nearly zero by December 2008, and from about 1.6% to nearly zero in March 2020. Savings and money-market yields fall with it within weeks; mortgage rates follow the 10-year yield, which usually falls too.

Not evenly. Most people keep their jobs in a recession. The cost lands hard on the ones who don't, and on people trying to get their first job or change jobs. See You're Looking for Work.

The warning lights

Because the official call comes so late, people watch indicators that have turned before or at the start of past recessions. Three are on the almanac's charts.

The Sahm rule: unemployment's direction. Take the three-month average of the unemployment rate and compare it with its lowest three-month average of the previous year. A rise of half a point or more is the signal. Claudia Sahm, then an economist at the Federal Reserve Board, designed it in 2019 not as a forecast but as a trigger for automatic stimulus payments, so that relief could start without waiting for the committee. In the real-time version on FRED, built from the unemployment figures as first published, it crossed 0.5 within four months of the start of every recession since 1970. It also crossed twice without one: for one month in late 1976, and in 2024, when the July figure, released August 2, put it at 0.53. It peaked at 0.57 in August and fell back below 0.5 by October. In October 2026 it read 0.00.

Unemployment rate with the Sahm indicator and its 0.5 threshold. The indicator crosses the line at or just after the start of each recession, and in 2024 without one. Full chart and context →

Jobless claims: layoffs, weekly. New claims for unemployment benefits arrive every Thursday, so they move first. A common rule watches for the four-week average to rise about 50,000 above its low of the past year. By this site's count, using the four-week average since 1967, that happened within about a year of the start of all eight recessions since, and about ten other times when no recession followed, among them after Hurricane Katrina in 2005 and in mid-2023. Claims catch layoffs, not a freeze in hiring, so they can stay low in a weakening market (see What jobless claims are). The four-week average was about 198,000 in the week ending November 14, 2026, near the low end of its history.

New jobless claims, four-week average, with recessions shaded. Each recession starts with a rise; so do several false alarms. Full chart and context →

The yield curve: what bond buyers expect. The 10-year Treasury yield minus the 3-month yield is normally positive, because lenders want more to tie money up for longer. When it falls below zero, an "inverted" curve, bond buyers expect the Fed to cut rates, usually because they expect a slowdown. The Federal Reserve Bank of New York turns the spread into a probability of recession twelve months ahead. On October 2026's spread it gave 12.4% for October 2027; on the spring 2023 spread its formula gave about 70%. Since the daily series begins in 1982, the curve first inverted 11 to 23 months before each of the four recession peaks, and gave one false alarm, five days in 1998. Then came the inversion of 2022–24: from October 25, 2022 to December 12, 2024, 534 trading days, by far the longest in the series (the next, in 2006–07, lasted 217), and as deep as −1.89 points on May 4, 2023. It dipped below zero briefly several times in 2025. On October 29, 2026 it stood at +0.80 points. The almanac's line is the monthly average, computed from the Fed's 10-year and 3-month yields; FRED publishes the daily series as T10Y3M.

The 10-year Treasury yield, with the 10-year minus 3-month spread in the panel underneath. Below the zero line, the curve is inverted. Full chart and context →

What they say now. As of late October 2026, none of the three is lit: the Sahm indicator is at zero, claims are low and the curve is upright. Growth was strong in the third quarter, at a 3.4% pace, while hiring stayed thin (see Growth Without Hiring). The lights describe the job market and the bond market's expectations. They can't see a shock from outside, an oil embargo or a virus, before it hits; the curve's 2019 inversion counts as a hit for 2020, though few think the bond market foresaw a pandemic.

Where the analogy breaks

A drought comes down to one number, rainfall. A recession is a judgment across six measures with no formula, which is why careful people can look at the same months and disagree until the committee speaks, and sometimes after.

And a drought declaration does something: it releases aid. The committee's announcement triggers nothing. No federal program switches on when it speaks; extended unemployment benefits, for instance, turn on by each state's own unemployment rate. Closing that gap, with help that starts by itself when unemployment turns up, is the problem Sahm's rule was invented to solve.

What people argue about

Two quarters or the committee. In July 2022, with a second down quarter about to be reported, the White House Council of Economic Advisers wrote that two quarters of falling GDP is neither the official definition nor how economists judge the business cycle, and that the broad measures (jobs, spending, production, income) were not signaling a recession. A week later Senators Tim Scott, Ted Cruz and Cynthia Lummis introduced a resolution affirming two consecutive quarters of falling GDP as the definition, and accused the administration of trying to redefine the word to avoid admitting a recession. Revisions have since removed the second down quarter and the committee never called one. Whether a definition nobody can apply until a year later serves the public better than a crude rule anyone can check is the part the revisions didn't settle.

Whether the warning lights failed. All three lit up in 2022–24 and no recession came. On the curve: Campbell Harvey, whose 1986 dissertation tied the yield curve to future economic growth, argued in March 2024 that calling it a false signal was premature, since recessions have usually started after the curve turns upright again, and that the curve can soften the downturn it predicts, because companies that see it trim investment and hiring early. By October 2024 he allowed that this might be the time it gives a false signal, in part because so many people now watch it. Two Federal Reserve Board economists, Eric Engstrom and Steven Sharpe, had argued in 2022 that the most-watched spread, 10-year minus 2-year, gives a muddled signal pushed around by risk premiums, and that a near-term spread tracking what markets expect the Fed to do over the next 18 months or so gives a cleaner one. Economists at the San Francisco Fed had argued earlier, in 2018, that a favorite reason for discounting an inversion, that the Fed's bond buying had pushed long-term rates down, doesn't hold up in the data. On unemployment: Sahm herself wrote in July 2024 that her rule was likely overstating the weakness, because immigration and the pandemic had swollen the number of people newly looking for work, and said in August that the U.S. was not in a recession despite the indicator bearing her name. Pascal Michaillat and Emmanuel Saez, of UC Santa Cruz and UC Berkeley, built a version from unemployment and job openings together and, on August 2024 data, put the chance that a recession had already begun at 48%. One side reads 2022–24 as the pandemic scrambling indicators built on older patterns; the other, as the indicators catching a real slowdown that was softened, by companies' early caution, as Harvey argues, or by the Fed's rate cuts in late 2024. The next recession will be the test.

See also

Sources

The weekly digest

One email on Monday mornings: the week's new articles and charts, the numbers that moved most, and what comes out in the week ahead. No ads, and no tracking of opens or clicks.

Past issuesHow addresses are kept