That is, the forwarded proposal Kevin Drum of Jabberwocking. Verbatim:
So: have we really created 2.5 million new jobs since March? I’m not sure I believe it.
The chart he posted succinctly sums up his argument:
resource: Drum (2022).
The stylized fact that the family series adds almost nothing to the information on cyclical behavior or employment contained in the institutional series is a point made by practitioners (see Furman, 2016or China Aeronautics and Space Administration, 2017). However, the possibility of measurement error still exists, especially during these periods. Non-seasonally adjusted nonfarm payrolls (NFP) employment and total employment covered by the Quarterly Census of Employment and Wages (QCEW) showed some signs of misreporting. Numbers vary for various reasons related to employee coverage, but overall trends should provide some insight.
figure 1: Log ratio of nsa NFP to QCEW (blue) and 2001-19 mean (red dashed line). Dates of peak-to-trough recessions as defined by NBER are shaded in gray. Source: BLS via FRED, Bureau of Labor StatisticsNBER, and author’s calculations.
In 2021 and the first half of 2022, NFP has the following What the historic relationship with QCEW means.
While the Establishment Survey is exactly that – a survey – the QCEW is a census and should in principle cover all covered employment through things like tax records. This data is used, among other things, to benchmark institutional series.The last preliminary benchmark is for March 2022 (but not yet in the official series – I’m at this postal). Note in Figure 2 that the ratio falls sharply in early 2022 and picks up in June. This suggests to me that as of June, the nsa NFP series is pretty close. Now, the QCEW lags significantly (data through June was released on 22nd November), so we can’t really use the QCEW to make a judgment on the July-November data.
Given this, what has been the composition of net employment growth measured differently since March 2022 (when was the last base month)? In addition to the NFP series officially established by the BLS, the QCEW series (seasonally adjusted by myself) and the family series adjusted according to the NFP concept are shown in Figure 2.
figure 2: Employment change in NFP (black), QCEW adjusted using the authors’ Census X-11 seasonal adjustment (dark blue), QCEW using the authors’ multiplicative moving average difference (sky blue), and using households adjusted for the NFP concept The employment series (red) are all 1000, relative to 2022M03. Source: BLS via FRED, Bureau of Labor Statistics, Bureau of Labor Statisticsand the authors’ calculations.
The Family Employment Study series adjusted to the NFP concept is not an official series and is described below here. A caveat: given that the household employment series is based on a relatively small sample, the variability associated with it – and thus by extension its derived series – will be relatively large.
As of June, it was not clear whether NFP was under or over estimated relative to the QCEW figures, given the mixed results provided by the seasonal adjustment process. The adjusted household series shows that only about 800,000 NFP jobs were created as of November, compared with 2.7 million in the official NFP series.
One way to address the seasonal adjustment problem (exacerbated by the extreme pandemic-related decline) is to look at the 12-month change. This obviously undercuts recent developments, but at least we can see trends.
image 3: The 12-month employment change relative to March 2022 is 1000 for both the non-seasonally adjusted NFP (black), QCEW (dark blue), and the NFP concept-adjusted household employment series (red). Source: BLS via FRED, Bureau of Labor Statistics, Bureau of Labor Statisticsand the authors’ calculations.
All three series are well matched through 2022M06. Both non-farm payrolls and adjusted households are down for November 2022. Which one do you believe? As mentioned earlier, Furman, 2016 and China Aeronautics and Space Administration, 2017 Almost all the weight is put on building the series.A more formal analysis shows that the almost total weight of the initial firm series for real-time analysis is given by Goto, Jacobs, Sinclair & Van Nord (2021)(BTW, if I look at the final data for the NFP and adjusted home series, statistically the adjusted home series seems to respond to the gap between the two series, while the NFP series does not ; formally, the adjusted series responds to an error correction term.)
So for now, while I think it’s an open question what (NFP) job growth is, I think (1) it could be positive and (2) it could slow.