The Sahm Rule in Motion

Transcript

Data and method

The unemployment rate (U-3), U-6, employment from the household survey and the group rates come from the Bureau of Labor Statistics via FRED, seasonally adjusted (by the BLS where it publishes an adjusted series, otherwise with STL). Family-income deciles come from IPUMS CPS microdata. Recession dates, the committee’s measures and its words on method come from the NBER’s Business Cycle Dating Committee. Two versions of the rule are used. The record since 1950 and the grid of groups use the Sahm Rule as it is usually computed: a series’ three-month average less its lowest three-month average over the previous twelve months, signaling at half a point. The chart of who is affected first, the racial, educational and income turning points, the Black and White comparison, the income tenths in each recession since 1990, and the county maps use the website’s version, on its CPS series (and, for the maps, on each county’s rate): a six-month average less its lowest six-month average of the last twelve months, this month included, with the alarm at half a point. Readings use today’s revised data, not the data available at the time.

The Michez Rule slides follow Pascal Michaillat and Emmanuel Saez’s “Has the Recession Started?” (working paper 2024; Oxford Bulletin of Economics and Statistics, 2025) as the website’s Indicator Comparisons page does. The unemployment rate is unemployment over the labor force (UNEMPLOY and CLF16OV), and the vacancy rate is JOLTS job openings, shifted forward one month, over the labor force. Each is taken as a three-month average. The unemployment side is that average less its low over the thirteen months to date; the job-openings side is the high over the same months less the average. The rule is the smaller of the two, read against a band from 0.30 points, where a recession may have begun, to 0.80, where it has; the chance of recession rises in a straight line across the band, so the August 2024 reading of 0.54 is about one in two. The October 2025 gap in the household survey is filled by a straight line.

The county slides use the website’s county data: each county’s unemployment rate from the BLS’s Local Area Unemployment Statistics, via FRED, seasonally adjusted the same way as every other series on the site. A county’s timing in the 2007–09 recession is the month it crossed the line, counted from the NBER’s peak in December 2007; the fifths compare the average timing of each fifth of counties. The 22 bellwethers are the counties that crossed in each of the 2001, 2007–09 and 2020 recessions and at no other time since 1993; their timing is the average over those three. County poverty rates and the share of adults with a bachelor’s degree are Census Bureau estimates prepared by the USDA’s Economic Research Service (about 2019), as compiled in the open JsonOfCounties dataset.

Acknowledgments

This work would like to thank Claudia Sahm, Trevon Logan, Pascal Michaillat, Emmanuel Saez, Steven Durlauf, and James Stock, for immeasurable guidance and mentorship on this work.