The Challenges in Dating the End of Recessions

The Business Cycle Dating Committee of the National Bureau of Economic Research (NBER) is generally thought of as the authority on deciding when recessions in the United States begin (the business cycle peak) and end (the business cycle trough). Because the committee does not revise the peak and trough dates, it has an incentive to get them right the first time. Consequently, the committee tends to wait a substantial amount of time before determining when a recession begins and ends.

Waiting provides not only more data but also more accurate data. Variables such as gross domestic product and employment are typically revised. Waiting to call a turning point allows the committee to use revised economic data that might be released later. In fact, the average lags for the committee to announce peaks and troughs are 7.3 months and 15 months, respectively.The Business Cycle Dating Committee was created in 1978, so these averages are calculated using the lag times between peak/trough dates and announcement dates for all recessions post-1978.

One might wonder why the end of recessions takes longer to time than the beginning. Two common reasons are the following:

• Recessions tend to begin sharply and are therefore easier to identify.
• The committee uses multiple data series to evaluate business cycle dates that move more closely together at the beginning of the recession but are more discordant during recoveries.

While we do not have access to the committee's exact methodology, economists have employed a number of different models to date turning points. The figure below plots recession probabilities from December 2019 to February 2021 predicted by three such models.

The orange line plots a model estimated by Marcelle Chauvet and Jeremy Piger that uses data corresponding to a given month to estimate the probability that the U.S. was in a recession in that month. As data corresponding to a given month is typically available in the proximate month, this series reflects real-time probabilities published with a month delay. For example, the recession probability in March 2020 is estimated using March 2020 data, which would have been available in April 2020. This model uses four monthly economic indicators in its prediction: nonfarm payroll employment, the index of industrial production, real personal income excluding transfer payments, and real manufacturing and trade sales.There is some overlap between these series and those utilized by the NBER: real personal income less transfers, nonfarm payroll employment, real personal consumption expenditures, wholesale-retail sales adjusted for price changes, employment as measured by the household survey, and industrial production.

The blue line plots a revised model estimated by Chauvet and Piger that uses the most up-to-date vintage of data (April 2021) from the four aforementioned data series to estimate the probability that the U.S. was in a recession in each sample month.

The red line plots a model estimated by Ana Beatriz Galvao and Michael T. Owyang that uses real-time Weekly Economic Index (WEI) data corresponding to a given month to estimate the probability that the U.S. was in a recession in each month.The release date of the WEI time series employed to compute these real-time probabilities coincides with the monthly release dates of Chauvet and Piger recession probabilities as recorded in ALFRED. Real-time data on WEI were obtained from ALFRED. The WEI comprises ten indicators of real economic activity, covering consumer behavior, the labor market and production. The dotted red lines indicate the 68% credible intervals for the predicted probabilities.

Though each of these models uses different estimation methods or data, the recession probabilities of all three models move together at the start of the pandemic recession (February to April 2020); economic activity on all fronts dropped dramatically and the models all identify the end of an expansion in February 2020 and the beginning of a recession in March 2020.

Starting in May 2020, however, the recession probabilities predicted by the models diverge dramatically; the models do not concur on when the recession ended, or even if it has certainly ended as of February 2021. This asymmetry reflects why the NBER calls business cycle peaks relatively quickly but has much more difficulty calling business cycle troughs.

Notes and References

1. The Business Cycle Dating Committee was created in 1978, so these averages are calculated using the lag times between peak/trough dates and announcement dates for all recessions post-1978.
2. There is some overlap between these series and those utilized by the NBER: real personal income less transfers, nonfarm payroll employment, real personal consumption expenditures, wholesale-retail sales adjusted for price changes, employment as measured by the household survey, and industrial production.
3. The release date of the WEI time series employed to compute these real-time probabilities coincides with the monthly release dates of Chauvet and Piger recession probabilities as recorded in ALFRED. Real-time data on WEI were obtained from ALFRED.

Additional Resources

• On the Economy: How Well Do Consumers Forecast Inflation?
• On the Economy: Using Data to Show When Recessions End