The Economics of Recession: A Survey
Part 2/10, August 2022 
2. What is a recession?
It almost seems unnecessary to start with the definition of a term so often bandied about as is “recession.” We all talk about the last recession and compare it with previous recessions or with the Great Depression. We all read about recessions in the media, reflexively absorbing the content of the news stories without stopping to think about a formal definition of the term. But is it really clear to everyone what it is that the macroeconomist considers a recession and how economists determine in practice whether a particular economy has entered or exited a recession?
When someone is pressed to define the term, a frequent reply is that a recession occurs when there are two or more consecutive quarters of negative GDP growth. Before 1990, the reference might have been to GNP growth, and a more careful respondent might clarify that it is real rather than nominal growth we care about, but this simple rule of thumb has been a common perception for a very long time. However popular, the two-negative-quarters rule is not the standard by which economists, the press and historians tend to identify recessions in the United States and many other industrial countries. In fact, if we compare periods identified by this rule with periods officially classified as recessions, the difference is quite remarkable.
In the United States, as earlier described, recessions are identified by the NBER. This authority to designate periods as recessions does not derive from law or government decree, but from de facto acceptance as official by economists, government, and media. The NBER, a private nonprofit research organization, came to perform this function in effect because Professor Wesley Clair Mitchell, one of its founders and its first president, had a long-term interest in business cycle research. This quasi-official concept of business cycle evolved throughout much of the twentieth century into the standards we observe today.
The NBER approach to the business cycle has two basic elements. First, the cycle alternates between expansion and contraction of aggregate economic output resulting from concurrent movements in many economic activities. Thus, a recession is the outcome of roughly simultaneous slowdowns in many sectors of the economy with resulting effects that are discernable in overall economic activity. Second, business cycles are recurrent, and the expansion and contraction phases are persistent, though the length of the business cycle and its phases may vary from case to case. In its original formulation, the NBER defined a cycle as lasting from over one year to about twelve years and these guidelines continue to apply in current practice.
Some features of the NBER definition are purposefully left vague. For instance, economic activity may be represented by aggregate output or by the levels of income, employment, expenditures and sales. The lack of specificity forces the economic analyst to track all of these measures, and others as well, to try to identify a cycle. Moreover, it is not enough to look at aggregate measures of activity. The careful analyst will call a recession only when the slowdown is pervasive across the economy, not just in a single sector. Finally, observed cyclical changes in economic activity have to persist over periods longer than one year and in particular must be distinguishable from intra-year seasonal fluctuations.
How does the NBER business cycle chronology compare with the popular two-negative-quarters rule? If we look at the period from 1947 Q2 to 2021 Q2 in the United States, 46 out of a total of 297 quarters were identified as being part of a recession either by the NBER or by the two-negative-quarters rule. Of the 46 quarters, the two methods agree in only 23 cases, one half of the quarters identified. There are 3 cases in which the two-negative-quarters rule called a recession but the NBER did not, whereas there are 20 cases in which the opposite held. The main reason for this difference is that many recessions contain at least one quarter of positive, even if meager, real growth interspersed among the negative growth quarters. Clearly, the two-negative-quarters rule would undercount those cases, but they may be included by the NBER in their search for persistent declines in various sectors of the economy over extended periods.
Two of our readings discuss the NBER definition of a business cycle at a general conceptual level. First, one of the framers of the NBER process, Geoffrey H. Moore, answers the question “What Is a Recession?” [R1] In this article, he provides both some historical context for the NBER method as well as a discussion of some of the practical issues involved in business cycle dating. In the more recent article “What Is a Recession?: A Reprise” [R2], Allan Layton and Anirvan Banerjee give a restatement and defense of the NBER method from a twenty-first century perspective, retaining the emphasis on multiple indicators and arguing against the two-negative-quarters rule.
To provide a detailed introduction to the NBER methodology, the readings include a few of the seminal writings of NBER researchers published during the gestational period of the approach. Two excerpts from the classic book by Burns and Mitchell, mentioned earlier, are referenced here. One is the first chapter from the original book, entitled “Working Plans” [R3]. In the very first page of the reading, the authors state the “official” definition of a business cycle, adapted from Mitchell’s earlier 1927 book on the topic but “with modifications suggested by experience in using it.” In 1946, their concept of the cycle still included four phases (expansions, recessions, contractions, and revivals) of which only expansions are still known by the same name. Modern recessions are closest to what Burns and Mitchell call “contractions.” The rest of the reading discusses their general approach to identifying the phases of the business cycle.
In the fourth chapter of the original book, Burns and Mitchell [R4] get down to brass tacks. They illustrate the graphical, tabular, and statistical techniques they used to analyze “specific” cycles in individual industries as well as their approach to putting everything together to come up with dates for the aggregate business cycle. Just as importantly, they provide in Table 16 a set of concrete monthly and quarterly business cycle dates for the United States, France, Great Britain, and Germany covering the period from December 1854 to May 1938, as well as dates going back to 1834 at an annual frequency.
Interestingly, Burns and Mitchell do not report in the table all four phases of the business cycle identified their definition. Instead, they focus on what they call “reference dates,” namely “peaks” and “troughs” of the cycles, which correspond roughly to what they refer in their conceptual discussion as “recessions” and “revivals.” With today’s conventions, we identify a recession as the period from the month or quarter after the peak to the month or quarter of the trough. To Burns and Mitchell, these were “contractions.” They classified the remaining periods – from the month or quarter after a trough to the following peak – as expansions, just as they would be today.
The dates for the United States given by Burns and Mitchell in 1946 are the same in most cases, or at least very close, to those currently in use by the NBER for the period in question. The exact present-day NBER dates were initially fixed by Geoffrey Moore in the article “Measuring Recessions” [R5]. Moore gives a detailed account of how he adjusted a few of the dates from the Burns and Mitchell chronology and how he dated cycles subsequent to the publication of the earlier work. He provides his final results in a table on the second page of the article even before he goes into the detailed analysis. In contrast to the Burns and Mitchell table, only monthly dates are given and only for the United States.
The NBER method has been very successful in practice in that it has assumed quasi-official status worldwide. Still, some economists have tried to construct “objective” procedures for dating recessions that do not rely on the judgment and discretion of a committee. Perhaps this view is another form of the “rules versus discretion” movement in economic policy or perhaps it is simply an attempt to achieve finality and put the process in autopilot mode. In any case, the literature contains various proposals for mechanical ways of dating recessions that do not rely on the judgment of individuals or committees. Recognizing the limitations of the two-negative-quarters rule, proponents have endeavored to come up with more sophisticated algorithms that may better approach the desirable characteristics of the NBER methodology.
To sort out recession quarters from expansion quarters, James Hamilton [R6] proposes “A new approach to the economic analysis of nonstationary time series and the business cycle.” The idea is to allow for Markovian switching between the two states of the economy in modeling the stochastic process that is assumed to generate real GNP growth. In a Markovian model, only the current state of the economy and the transition probabilities matter, rather than the full history of the variables, and the unobserved values of the model’s parameters may be statistically inferred from real GNP data. Hamilton shows that recession dates derived from his method for the United States are very similar to the NBER dates, even though the latter are not used in the derivation. At a quarterly frequency, some of the turning point dates match exactly, though others are off by as many as three quarters. The Markov switching model is conceptually attractive and produces reasonable results, but it is clearly not a perfect substitute for the NBER.
Michael Boldin [R7] compares the Markov switching approach with four other methods of dating business cycle turning points including the NBER approach, the two-negative-quarters rule of thumb and variations, and peaks and troughs extracted from two business cycle indices that were being produced independently at the time the article was published by the U.S. Department of Commerce and by NBER faculty affiliates James Stock and Mark Watson. Using the NBER dates in effect as a benchmark, Boldin finds that the Markov switching and Stock-Watson approaches performed well historically but that all of the mechanical methods had difficulties in identifying the 1990-91 recession. The article ends with the sobering thought that “It is unlikely that a be-all and end-all technique for dating business cycles can ever be developed.”
Harding and Pagan propose a more sophisticated variant of the two-negative-quarters rule in the article “A comparison of two business cycle dating methods” [R8], which draws on ideas from earlier research published by the NBER. The authors pit their non-parametric approach against the Hamilton Markov switching model and find that their method approximates the NBER dates more closely. They also argue that the non-parametric approach is more robust because, in contrast to Markov switching, it does not rely on the validity of an underlying statistical model.
One clear conclusion from research in this area is that finding a mechanical model that matches NBER business cycle dates exactly is an elusive goal, even though various models have come close using U.S. data since the 1950s. Perhaps motivated by that observation and perhaps also because of the wide acceptance of NBER dates, Issler and Vahid [R9] turn the focus of most of this research on its head. Instead of trying to find an index variable whose fluctuations match the NBER dates, they use the NBER dates to build a coincident indicator of the U.S. economy. Their method converts the “yes or no” NBER recession indicator into a quantitative variable whose cyclical fluctuations match the NBER dates as closely as possible.
In the final article of this section, Stock and Watson [R10] return many decades later to one of the basic questions raised in Mitchell’s early research on business cycle dating. Is it better to employ a top-down approach in which aggregate economic indicators are first examined and sectoral indicators are used later for corroboration or a bottom-up approach in which sectors are examined first and the results are later aggregated? Using tools from present-day econometrics, the authors tentatively conclude that Mitchell’s bottom-up approach is preferable.
Readings referenced from book The Economics of Recession
R1 Geoffrey H. Moore (1967), ‘What is a Recession?’, American Statistician, 21 (4), October, 16–9
R2 Allan P. Layton and Anirvan Banerji (2003), ’What is a Recession?: A Reprise’, Applied Economics, 35 (16), 1789–97
R3 Arthur F. Burns and Wesley C. Mitchell (1946), ‘Working Plans’, in Measuring Business Cycles, Chapter 1, New York, NY, USA: National Bureau of Economic Research, 3–22
R4 Arthur F. Burns and Wesley C. Mitchell (1946), ‘Dating Specific and Business Cycles’, in Measuring Business Cycles, Chapter 4, New York, NY, USA: National Bureau of Economic Research, 56–114
R5 Geoffrey H. Moore (1958), ‘Measuring Recessions’, Journal of the American Statistical Association, 53 (282), June, 259–316
R6 James D. Hamilton (1989), ‘A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle’, Econometrica, 57 (2), March, 357–84
R7 Michael D. Boldin (1994), ‘Dating Turning Points in the Business Cycle’, Journal of Business, 67 (1), January, 97–131
R8 Don Harding and Adrian Pagan (2003), ‘A Comparison of Two Business Cycle Dating Methods’, Journal of Economic Dynamics and Control, 27 (9), July, 1681–90
R9 João Victor Issler and Farshid Vahid (2006), ‘The Missing Link: Using the NBER Recession Indicator to Construct Coincident and Leading Indices of Economic Activity’, Journal of Econometrics, 132 (1), May, 281–303
R10 James H. Stock and Mark W. Watson (2010), ‘Indicators for Dating Business Cycles: Cross-History Selection and Comparisons’, American Economic Review: Papers and Proceedings, 100 (2), May, 16–9