Does this mean we should expect a recession in next month’s data or the month after that? Maybe, maybe not.
figure 1: Civilian employment (blue, left logarithmic scale) and nonfarm employment (tan, right logarithmic scale) and nonfarm employment (tan, right logarithmic scale) aged 16 and older as of November 3, 2023, are seasonally adjusted in units of 1,000. Source: BLS via FRED.
Looking at existing data for the past four recessions, one finds that in two cases the household series peaked before the economy was established, in one case three months earlier.However, the immediate results were somewhat different, with observers simultaneously assessing the onset of the recession
Figure 2-5 shows the immediate evaluation (see this postal Discuss final versus immediate data).
figure 2: Civilian employment (blue, left logarithmic scale) and non-agricultural employment (tan, right logarithmic scale) over the age of 16, both in units of 1,000, seasonally adjusted as of June 7, 1990. NBER-defined recession peak-to-trough dates appear gray. Source: BLS, via ALFRED, NBER.
image 3: Civilian employment (blue, left logarithmic scale) and non-agricultural employment (tan, right logarithmic scale) over the age of 16, both in units of 1,000, are seasonally adjusted as of May 4, 2001. NBER-defined recession peak-to-trough dates appear gray. Source: BLS, via ALFRED, NBER.
Figure 4: Civilian employment (blue, left logarithmic scale) and non-agricultural employment (tan, right logarithmic scale) over the age of 16, both in units of 1,000, are seasonally adjusted as of February 1, 2008. NBER-defined recession peak-to-trough dates appear gray. Source: BLS, via ALFRED, NBER.
Figure 5: As of April 3, 2020, civilian employment (blue, left logarithmic scale) and nonfarm employment (tan, right logarithmic scale) aged 16 and older are seasonally adjusted in units of 1,000. NBER-defined recession peak-to-trough dates appear gray. Source: BLS, via ALFRED, NBER.
In real time, there are two cases (2001, 2007) where the family series begins a month earlier than the founding date, and the two turning points are at the same time (1990 and 2020). In the revision, the peak of the civil series was advanced once (in 1990) and postponed by two months (in 2001). The NFP peak was delayed once (2001 recession).
Extrapolating from the past is always a dangerous enterprise.We now know the response rate for first readers of the report Institutional investigation dropped sharply, from 60% pre-pandemic to 42% in June (although last read Not that bad, about 94%).The home series has also experienced Response rate drops70% in July 2023, and 82% before the outbreak.
In general, most practitioners place almost all weight on the institutional series, which makes sense because the standard deviation of the (annualized) monthly growth rate for the household series since 2000 is 3%, while for institutions it is 2% (I have omitted 2020M02-2022M01).
Regardless, it makes sense to cross-check the data. The following are several employment growth indicators corresponding to the NFP concept, among which the QCEW series does not rely on the BLS birth-death model and survey response rate. The Philadelphia Fed’s early benchmark relies on survey data after the first quarter.
Figure 6: NFP employment incorporates the author’s calculated preliminary benchmark (bold blue), the Philadelphia Fed’s early benchmark (red), household employment adjusted to conform to NFP concepts (tan), and the author’s adjusted QCEW covered employment based on Census X-13 ( green), all in thousands, seasonally adjusted. Source: BLS via FRED, Philadelphia FedBureau of Labor Statistics, BLS-QCEWand the author’s calculations.
Another check is on private employment using ADP data.
Figure 7: Total private nonfarm employment, from BLS (tan, right axis) and ADP (blue, left axis). Source: BLS, ADP, via FRED.
