Is Employment Growth (A Lot) Lower Than We Think?
Yesterday’s employment release indicated a deceleration in nonfarm payroll (NFP) employment growth. Data from other sources suggests that the deceleration over the last year is more marked than indicated by the establishment survey.
First, consider the deceleration relative to 2017-18 (stochastic) trend, in Figure 1.
Figure 1: Nonfarm payroll employment (bold blue), stochastic trend for NFP over 2017-18 (red line), Quarterly Census of Employment and Wages (QCEW) (bold brown), and QCEW adjusted using Census X-13/log transform w/Census X-11 ARIMA (green). Source: BLS via FRED, BLS and author’s calculations. Data [xls]
Notice that total employment growth as measured by the Quarterly Census of Employment and Wages (QCEW) has decelerated, but more markedly than the establishment survey measure. This is hard to see due to the seasonality in the series. Applying the Census X-13 seasonal adjustment procedure on the log transformed QCEW series confirms the deceleration.
Given that the QCEW data will be used to benchmark revise the establishment series come February, it’s of interest to consider what these QCEW data will imply. I use a regression of the establishment series on the seasonally adjusted QCEW data 2001-2019M06 to obtain an fitted establishment series. (The fit is unsurprisingly very good, adj. R-squared at 0.999, SER at 0.0015.)
The establishment, fitted, and forecasted establishment series are shown in Figure 2. The implied establishment series deviates strongly starting in 2019M01, and the forecast (based on an ARIMA(1,1,1) continues to do so. (The ARIMA regression adj. R-squared is 0.42, SER = 0.0015.)
Figure 1: Nonfarm payroll employment (bold blue), stochastic trend for NFP over 2017-18 (red line), August 2019 preliminary benchmark revision for March 2019 NFP (dark blue square), implied NFP using Quarterly Census of Employment and Wages (QCEW) (bold green), and forecasted QCEW implied NFP using ARIMA(1,1,1) (bold brown), with 90% forecast interval (gray lines). Source: BLS via FRED, BLS and author’s calculations. Data [xls]
Interestingly, the preliminary benchmark revision released in August 2019 for March 2019 levels matches almost exactly the implied value I obtain. The preliminary benchmark is not the last word, of course. However, the negative half million revision is larger in absolute value than typical, according to Oxford Economics’ Gregory Daco.
I know some observers will note that civilian employment from the household survey indicates nearly double the growth rate of that from the establishment (2% vs. 1.1%, m/m annualized, in log terms). However, y/y, household lags at 1.25% vs. 1.39%. In addition, as Furman notes, it’s the household series is almost irrelevant for inferring high frequency movements in the labor market.
What does this imply for business cycle indicators, particularly those followed by the NBER’s Business Cycle Dating Committee? The conventional indicators are shown in Figure 3. The alternative nonfarm payroll employment indicator is shown in Figure 4.
Figure 3: Nonfarm payroll employment (blue), industrial production (red), personal income excluding transfers in Ch.2012$ (green), manufacturing and trade sales in Ch.2012$ (black), and monthly GDP in Ch.2012$ (pink), all log normalized to 2019M01=0.Source: BLS, Federal Reserve, BEA, via FRED, Macroeconomic Advisers (12/30 release), and author’s calculations. Data [xls]
Figure 4: Nonfarm payroll employment implied by Quarterly Census of Employment and Wages (light blue), industrial production (red), personal income excluding transfers in Ch.2012$ (green), manufacturing and trade sales in Ch.2012$ (black), and monthly GDP in Ch.2012$ (pink), all log normalized to 2019M01=0.Source: BLS, Federal Reserve, BEA, via FRED, Macroeconomic Advisers (12/30 release), and author’s calculations. Data [xls]
The employment series now seems more in line with monthly GDP, industrial production and manufacturing and trade sales, at least through 2019M06. Note that even then, the 3 month change in employment so measured has not gone negative as of 2019M06, as one would expect in a recession.
All data in used in this post: Data [xls]
Disclosure: None.