Subscribe to Our Blog
Considering the widespread speculation that the economy is in a recession, we thought it might be useful to compare the economic factors that the National Bureau of Economics (NBER) Business Cycle Dating Committee uses to determine when the economy has peaked, and when it has reached a trough. While it is commonly believed that two consecutive quarters of decline in GDP define a recession, the NBER actually uses a broad set of monthly indicators to identify the start and end of a recession. For the past 2 decades they have relied heavily on real personal income less transfer payments, and nonfarm payroll employment, but they also consider retail and wholesale trade (in constant dollars), the household employment survey, industrial production, and real consumption of services. In addition, the duration of a downturn is taken into consideration.
The radar plot shown below displays the average monthly growth of these factors for each recession from 1980 through the Great Recession. The pandemic recession was omitted due to its anomalous nature. When you look at these, there is very little indication that the economy in the 4Q is in recession.
The dashed line indicates the most recent observation of each series. Based on these observations the U.S. economy is not in a recession. The unemployment rate increased 0.2% in October, and the most recent print industrial growth was negative, but all of the other factors are still growing.
To get a sense of the relative importance of the different factors we estimated a logistic regression using the 7 series in Figure 1 as explanatory variables, and an indicator variable (1 for recession, 0 otherwise) as the dependent variable. We used data from Jan. 1959 to present, so there are 9 recessions. The average marginal effects are in the following table.
All of the variables except retail sales and wholesale trade are statistically significant. The labor variables have the greatest impact. An increase in the unemployment rate from 3.7 to 4.7% will increase the probability of a recession by 29%. A 2% decrease in non-farm payroll employment will increase the probability of a recession by 74%. (The average decline in non-farm payroll employment was 1.9% for the recessions in the data set, excluding the pandemic recession which saw a decrease of 14%.) A 1% decrease in real personal income less transfer payments has a much smaller impact, increasing the probability of recession by 2.5%.
The model estimates of the probability of a recession are shown below. In general, this simplistic model does a fairly good job of capturing the recessions, although there are a few notable false positives. For instance, in March 1960 when the unemployment rate increased by 0.6% the probability was almost one. Interestingly, the actual recession started the following month. Overall, 94% of the area of the ROC plot (now shown) lies below the curve suggesting that the model does a reasonable job of correctly predicting recessions.
For the current level of the indicators, the probability of a recession is about 10 percent. The variable to watch is the unemployment rate. The marginal effect is large, and the series is notoriously asymmetric, meaning that it increases at a much faster rate than it decreases. The average monthly increase during the first 6 months of a recession is 0.16% with a standard deviation of 0.2%. That’s why keeping an eye on the claims data will be particularly useful to seeing what is happening in close to real time.
