How Recessions Are Dated

Brent O’Mara’s formative years taught him that conventional assumptions do not always hold. One of the commonest assumptions about the economy, that two quarters of falling output define a recession, fails the test of this entry. The figure at the start of the story is Wesley Clair Mitchell. He served as director of research at the National Bureau of Economic Research from its founding in 1920 until the middle 1940s, led a long research program on business cycles, and wrote the two works that became the standard references, a 1927 study and a 1946 monograph with Arthur Burns called Measuring Business Cycles.¹ The bureau published its first business cycle dates in 1929. In 1961 the Commerce Department began printing the bureau’s chronology of peaks and troughs in a monthly publication without endorsing it, and there has never been an alternative chronology compiled by the government.¹ Mitchell’s successors still date American recessions by his method, which is judgment over many measures, never a formula applied to one.

The committee and its method

The Business Cycle Dating Committee in its current form was launched by the bureau’s president, Martin Feldstein, in 1978. Its inaugural chair, Robert Hall of Stanford, served for 46 years before Valerie Ramey succeeded him in 2024, and the bureau’s president appoints its members, who are specialists in macroeconomics and business cycle research.¹ The committee’s own questions and answers define its job as maintaining a chronology of peak and trough months. A peak is the month in which a range of indicators reach their highest level before a significant decline; a trough is the month activity reaches its low and begins a sustained rise. A recession, in the bureau’s traditional definition, is a significant decline in economic activity that is spread across the economy and lasts more than a few months, and the committee reads that definition as three tests, depth, diffusion and duration, each of which must be met to some degree while an extreme reading on one may offset weakness on another.¹ Expansion is the normal state of the economy, and most recessions are brief.¹

The measures are monthly. The committee follows real personal income less transfers, nonfarm payroll employment, real personal consumption, manufacturing and trade sales adjusted for prices, household survey employment and industrial production, and it states flatly that there is no fixed rule about which measures contribute or how they are weighted.¹ The San Francisco Fed explained in 2003 why quarterly output gets little weight in the monthly chronology: it is measured only quarterly and is subject to continuing, large revisions, so the committee refers almost exclusively to monthly indicators and treats economy wide employment as the broadest of them.³ A separate quarterly chronology weighs the expenditure and income measures of output equally, and the quarter does not always contain the month.¹

Figure

How a turning point gets its date

1. 01
   WatchMonthly measures of real activity are followed: real personal income less transfers, payroll employment, real consumer spending, manufacturing and trade sales, household employment and industrial production.
2. 02
   WeighNo fixed rule says which measures count or how much; depth, diffusion and duration are each judged, and an extreme reading on one can offset a weaker one on another.
3. 03
   WaitThe committee waits for standard data revisions and for enough time to remove any doubt that a peak or trough has occurred.
4. 04
   AnnounceThe peak or trough month is published, along with a separate quarterly date that weighs output and income measures equally.

The two quarter rule that the financial press repeats plays no part in any of this. The committee said so in its April 25, 1991 announcement, which dated the start of a recession to July 1990 even though some key monthly measures were still rising during most of that quarter, and noted that the popular rule of thumb defining a recession as two successive quarterly declines in real output played no role in its procedure.² Its questions and answers give four reasons: activity is not identified with output alone, the decline must be significant and two small quarterly dips would not qualify, the main focus is the monthly chronology, and the income side estimate of output gets equal weight with the expenditure side, a difference that mattered in both 2001 and 2007 to 2009.¹ The recession of 2001 did not include two consecutive quarters of declining output at all.¹ The 1991 announcement also shows how untidy a peak can be: payroll employment peaked in June, real personal income in July, sales in August and industrial production in September, and the committee chose July.²

Waiting is part of the method. The committee allows time for the standard data revisions and waits long enough to remove any doubt that a turning point has occurred. The shortest lag between a turning point and its announcement was four months, for the February 2020 peak; the longest was 21 months, for the March 1991 trough, announced in December 1992.¹ The March 2001 peak was not announced until November 26, 2001, and the November 2001 trough not until July 17, 2003, when the committee observed that the eight month recession had been slightly shorter than the postwar average of eleven.³ The June 2009 trough was announced on September 20, 2010, after revisions to the national accounts had clarified the 2009 path of output and income.⁷⁹ The committee has not changed a previously announced date since it was created, though it would if it concluded a date was wrong.¹

Figure

Months between a turning point and its announcement

• February 2020 peak4
• March 2001 peak8
• June 2009 trough15
• March 1991 trough21

The record, what changed and what did not

The chronology reaches back to a peak in June 1857. FRED, the St. Louis Fed’s data service, shades every one of its graphs with the bureau’s dates and publishes the full list, 34 pairs of peak and trough months from 1857 to the February to April 2020 contraction, and keeps a lighter shading on a recession until the committee has fixed its end.⁵ The San Francisco Fed counted 32 recessions since the mid 1850s when it answered the question in early 2008, ten of them since 1945, and reported that from the middle 1940s to 2007 the average recession lasted 10 months and the average expansion 57, so that a complete cycle ran about five years and seven months.⁴ The shortest postwar recession ran from January to July 1980; the two longest in that span, November 1973 to March 1975 and July 1981 to November 1982, each lasted 16 months; the longest expansion ran 120 months, from March 1991 to March 2001.⁴ The St. Louis Fed’s education podcast adds the extremes of the whole record: the shortest expansion of ten months in 1919 and 1920, and the longest recession, more than 65 months from October 1873 to March 1879, known as the Long Depression.⁶

The causes vary as much as the lengths. History.com counts thirteen recessions since the Depression and attributes the 16 month downturn of 1973 to 1975 to an oil embargo and wage and price freezes that produced stagflation, with output down 3.4 percent and unemployment nearly doubling to 8.8 percent; the six month recession of 1980 to a second oil shock and inflation of 13.5 percent; and the 1981 to 1982 recession, which gave Americans the phrase double dip, to a central bank that pushed interest rates to 21.5 percent in 1982 to break inflation, with output down 3.6 percent and unemployment above 10 percent.⁸ The committee itself does not recognize a double dip as a category: two contractions are either separate recessions or one, decided by the strength of the upturn between them, which is how the 1980 and 1981 recessions came to be counted separately.¹ The Great Recession of December 2007 to June 2009, at 18 months, became the longest since the Second World War, and the Bureau of Labor Statistics, which calls the bureau the official arbiter of American recessions, counted ten recessions between 1948 and 2011.⁹⁷ The committee also declines to identify depressions as a separate class; the 1930s contraction from August 1929 to March 1933 is simply the worst recession on its list, followed by a second from May 1937 to June 1938.¹ The collapse of 1929 and the financial crisis each have an entry of their own.

Figure

The record as the San Francisco Fed summarized it in 2008

recessions identified since the mid 1850s

32

recessions between 1945 and the end of 2007

10

average length of a recession from the mid 1940s to 2007

10 months

average length of an expansion over the same period

57 months

What changed over a century and a half is the shape of the record. The St. Louis Fed’s lesson notes that recessions became less frequent and shorter over time, and the committee’s chronology confirms it, with contractions of several years in the nineteenth century giving way to postwar recessions that averaged under a year.⁶ The 2020 episode changed one more thing: the committee concluded that a decline so deep and so widely diffused should be classified as a recession even though it lasted only two months, and it judged that unemployment insurance claims, which it normally weighs lightly, had been unusually informative in March of that year.¹

What did not change is the method. The dates are still set by a small group of economists reading many monthly series, still announced long after the fact, and still indifferent to the two quarter shortcut; the unemployment rate, which rose for 15 months after the March 1991 trough and for four months after the June 2009 trough, is still a lagging guide to the turning points rather than the turning point itself.¹ A recession, on the bureau’s chronology, is a judgment about direction, not level, and the first months of an expansion can leave activity well below the previous peak, as payroll employment did until May 2014.¹ The eight month recession of 2001 that the committee dated so carefully is the setting for the next entry.