Does a Downturn in Household Survey Employment Better Presage a Recession than One in the Establishment Survey?

I sometimes see this assertion. It clearly has relevance given the pattern we have as of June 2022 data. The red arrows indicate local maxima.

Figure 1: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 7/24/2022. Source: BLS via FRED. 

The argument is based on the conjecture that the BLS birth/death model (see discussion here) used to process data from the establishment survey is most likely to go off track at turning points, and in particular, during the Great Recession and the covid pandemic.

Looking at the last four recessions, we see the following patterns:

Figure 2: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 7/24/2022. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via FRED, NBER.

Figure 3: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 7/24/2022.  NBER defined peak-to-trough recession dates shaded gray. Source: BLS via FRED, NBER.

Figure 4: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 7/24/2022. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via FRED, NBER.

Figure 5: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 7/24/2022. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via FRED, NBER.

In two cases, the civilian employment series (household survey) peaks before the nonfarm payroll series (establishment survey), that is the 1990 and 2007 recessions; in the former, case the household peak is a whole three months earlier. In one case, the establishment series peaks first, and another, they are tied (2020).

Now, these are the final revised series, as of 7/24. They are not what economists would’ve seen as they were contemplating the possibility of a downturn. What did the series look in “real time”, i.e., as observers were contemporaneously assessing the onset of a recession? The corresponding figures are shown in Figures 6-9.

Figure 6: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 6/7/1990. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via ALFRED, NBER.

Figure 7: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 5/4/2001. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via ALFRED, NBER.

Figure 8: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 2/1/2008. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via ALFRED, NBER.

Figure 9: Civilian employment over age 16 (blue, left log scale), nonfarm payroll employment (tan, right log scale), both in 1000’s, seasonally adjusted, both as of 4/3/2020. NBER defined peak-to-trough recession dates shaded gray. Source: BLS via ALFRED, NBER.

In real time, the household series turns one month earlier than the establishment in two cases (2001, 2007), and twice the turning points are the same time (1990 and 2020). In the revisions, the civilian series peaks are moved earlier once (1990), and later by two months (2001). The NFP peak is moved later once (2001 recession).

I do not think one take too much from the current flattening out of the household series. With this in mind, here’s a recap of the two series, plus the household series adjusted to a NFP concept. This latter series was developed in response to conservative criticism that the conventional establishment series, relying on the birth/death model, was prone to error in the wake of the 2001 recession (i.e., critics argued employment was underestimated). The research series is shown in red below, and is not subject to the same problems with estimating the birth and deaths of firms (but does suffer from the same relatively small sample size associated with the household survey).

Figure 10: Civilian employment over age 16 (blue), civilian employment adjusted to nonfarm payroll concept (BLS series LNS16000000)  (red), nonfarm payroll employment (tan), Bloomberg consensus of 7/25 (tan square), all in 1000’s on log scale, seasonally adjusted, all as of 7/24/2022. Source: BLS via FRED, BLS, Bloomberg, and author’s calculations. 

The June civilian employment was down, while civilian employment adjusted to NFP concept was up (-315, + 131), while NFP was up 372. The Bloomberg consensus for July as of today is +255.