We have written about the predictive performance of two financial variables (the yield curve slope and the gap between the fund’s rate and its estimated neutral rate) in predicting recessions.1 Let’s now consider two variables directly related to economic activity itself: real GDP growth and the unemployment rate. The first is about what is called “stall speed” and the second is related to what has become known as the Sahm Rule.
Stall Speed: A Low-Growth Threshold
The analogy here, of course, is to airplanes. If a plane loses velocity, there is some threshold beyond which it can’t maintain sufficient lift and descent cannot be avoided. The same may be true of the economy. Indeed, Jeremy J. Nalewaik finds that the notion of stall speed has “some empirical content” as a predictor of recessions.2 He describes the concept this way: “By a stall speed, we mean particular values…such that when these values are reached during an expansion, the economy has tended to move into a recession within a fairly short time span.”
He finds that “output tends to transition to a slow-growth phase at the end of expansions before falling into a recession.” That’s intuitive! The economy likely doesn’t transition from a high-growth phase directly to a recession without passing through a low-growth phase. Of course, we need to identify some threshold for how low that growth rate has to be. He looks at the behavior of quarterly annualized growth rates of output leading up to recessions. He finds that the results are stronger for GDI compared to GDP. With respect to GDI, he finds that 64% of the observations of below-1% growth during expansions came in the four quarters preceding a recession. With respect to GDP, on the other hand, the corresponding figure is 48%. He then designs Markov-switching models in which output transitions to a low-growth phase before recessions. He makes an interesting suggestion based on the result that a model using industrial production growth didn’t identify a stalled phase: “This suggests that an important part of some stall phases may have been concentrated in services, a fact that may help explain some of the relative superiority of GDI over GDP in identifying stall phases.” Overall, he finds the stall speed concept is only moderately useful in forecasting recessions because there are a lot of false positives. He finds that adding the yield curve slope, change in housing starts, and the change in the unemployment rate to his model reduces false positives and improves recession forecasting.
The Unemployment Rate: The Sahm Rule
Another indicator of a recession is a rise in the unemployment rate. Claudia Sahm, a Fed Board economist, has studied this concept and wrote about it earlier this year as part of a joint project between Brookings and The Washington Center for Equitable Growth, the latter of which she’ll join as the director of macroeconomic research beginning later this month.3 Her suggested indicator has gotten some attention and has become
1 See “The Yield Curve and Recession Risk: Is This Time Different?” July 29, 2019.
2 Nalewaik, Jeremy J. (2011). “Forecasting Recessions Using Stall Speeds.” FEDS working paper 2011-24. 3 Sahm, Claudia (2019). “Direct Stimulus Payments to Individuals” Chapter in Recession Ready: Fiscal Policies to Stabilize the American Economy, edited by Heather Boushey, Ryan Nunn, and Jay Shambaugh. Washington, DC: The Hamilton Project and the Washington Center on Equitable Growth.
known as the Sahm Rule: A downturn is probably already occurring if the three-month moving average of the unemployment rate has risen at least 0.5 percentage points above its low point over the previous twelve months.4
The unemployment rate rises when growth falls below its potential rate. That signals a loss of momentum of the economy. A sufficient loss of momentum apparently becomes self-reinforcing, in effect, a point of no return. This rule has signaled every recession since 1970 with virtually no false positives. Very impressive. Sahm proposed using this as a trigger for an automatic stabilizer, her preferred method being direct stimulus payments to individuals.
However, that doesn’t seem as useful for predicting recessions ex-ante as it is for identifying recessions ex-post. It has tended to cross the 0.50-percentage-point threshold once the economy has already entered a recession, though fairly quickly, several months after the start. It still signals a recession well before one is officially recognized, however. Unfortunately, the de-facto official arbiter of dating recessions (the National Bureau of Economic Research’s Business Cycle Dating Committee) does not identify the start of a recession for quite some time after one has begun. For example, the last recession began in December 2007. The Business Cycle Dating Committee did not render its verdict until a year later, December 2008. In contrast, the Sahm indicator crossed the 0.50 threshold in April 2008, identifying the recession many months earlier. It turns out that economists are not only bad at predicting recessions; they are not even very good at quickly identifying a recession even after one has begun!
This indicator, shown in Figure 1 below, has recently been added to FRED as the “Real-Time Sahm Rule Recession Indicator.”
Still, we are keenly interested in anticipating a recession. That would facilitate, for example, an early turn to more aggressive easing by the Fed. In a note by Nunn et al., the Sahm recession indicator is used to estimate the probability of a recession.5 They use real-time estimates of the three-month average unemployment rate from 1970 to 2019 and exclude the months between when the NBER announces the beginning of a recession and when it announces the end of recessions (140 out of 592 months). In that sample, the probability
4“A Recession Is Coming (Eventually). Here’s Where You’ll See It First,” The New York Times, July 28, 2019. 5“How will we know when a recession is coming?” Ryan Nunn, Jana Parsons, and Jay Shambaugh, June 6, 2019.
of a recession at any unemployment rate is 12%. When all 592 months are included, the probability in any given month is 15%.
They give the following results for the likelihood of recessions over various horizons and for various values of the Sahm indicator:
Reproduction of a table produced by Nunn et al. 2019. See the original piece for further details. Source: BLS, Nunn, et al. 2019, Sahm 2019.
▪ When the three-month average unemployment rate is below the low for the unemployment rate over the previous 12 months, there is virtually no chance of a recession. The odds of one in the next 12 months is only 10%.
▪ If the three-month average unemployment rate has risen 0 to 0.09 percentage points above the 12-month low for the unemployment rate, there’s a 20% chance of a recession in the next 12 months.
▪ If it has risen 0.10 to 0.19 percentage points above the 12-month low, while it still is almost certain that a recession has not begun, the probability of a recession within the next 12 months is 33%.
▪ It gets more interesting when it has risen 0.20 to 0.29 percentage points above the 12-month low. In that case, the probability that the economy is in a recession is 11%, and the probability it will enter one within one year is 39% of the time.
▪ When it is 0.30 to 0.39 percentage points above the 12-month low, the probability that the economy is already in a recession is 40%. In that case, the probability of a recession within the next three, six, and 12 months is also 40%.
▪ When it is 0.40 to 0.49 percentage points above the 12-month low, the probability that the economy is already in a recession is 76%. In that case, the probability of a recession within the next three, six, and 12 months is also 76%.
▪ Finally, if the three-month average unemployment rate has risen by at least 0.50 percentage points above the 12-month low for the unemployment rate, the probability that the economy is already in a recession is 97%.
How About Today?
Today the Sahm indicator is zero! The unemployment rate rises in our forecast by several tenths from its low, three tenths from its current level. However, the corresponding Sahm rule indicator, using the 3-month average minus the 12-month low, doesn’t flash red because the rise in the unemployment rate is sufficiently gradual.
▪ The concept of stall speed has some empirical content. Based on historical experience, if quarterly output growth below 1% is observed, there are roughly even odds of a recession in the next year. In addition to the problem of false positives, however, not all recessions have been preceded by a slowdown of this sort.
▪ If the three-month average unemployment rate gets three-tenths above its recent 12-month low, that’s a recession alert, enough for the FOMC to respond as appropriate, taking into account inflation risk. If it gets four-tenths above its recent minimum, it’s a recession call, and we would predict a recession over the next 12 months, likely sooner. ▪ While each of these recession indicators is useful, a judgment of the likelihood of an impending recession will be improved by looking at the yield curve and rate gap as well, in particular predicted probabilities from probit models incorporating both.