Lots of data comes out every day. Some matters greatly, some less, some not at all.
Nonfarm payrolls matter a lot. Especially in the current environment of pivots, recessionary will-they-won’t-theys and shifting market narratives, the US’s flagship jobs report is arguably the most important regular data release around.
The latest headline NFPs had the US adding 517,000 jobs in January, or losing 2.5mn on a non-seasonally-adjusted basis. Choose your own adventure.
These statistically processing difficulties are one thing, but there’s also the danger that NFPs might be just generally, well, wrong.
That’s the takeaway from a new note by Standard Chartered’s global research team, which has examined at the striking gulf between employment figures as pronounced in the Current Employment Statistics (CES) release (which contains the NFPs), and the slower-moving Quarterly Census of Employment and Wages (QCEW).
Both are published by the Bureau of Labor Statistics, but the NFPs are not replaced by the QCEW data when it becomes available — unlike national accounts and productivity readings, which do retroactively integrate QCEW.
StanChart’s Steve Englander and Sarah Hewin write:
BLS every month makes a forecast for NFP of how many jobs are created by firms that open and firms that close. Their methodology is reasonable, but the estimate can be way off. Five months later QCEW comes out with a direct count because firms enter the UI system when they open and drop out when they close. It looks like there was a big gap between the estimate and reality in Q2 2022. We also think that Q3 2022 NFP job gains were overestimated, but there is risk of offsetting errors elsewhere. If Q3 QCEW implies softer jobs growth along with Q2, it means that 2022 jobs growth may be far less pronounced than advertised.…
We are a little puzzled that BLS leaves money on the table by not updating NFP as QCEW becomes available, given how much more authoritative QCEW is.
The crux of the problem is BLS’s Net Birth-Death (B-D) model, which is integrated into the CES. As the BLS site explains:
There is an unavoidable lag between an establishment opening for business and its appearance on the sample frame making it available for sampling. Because new firm births generate a portion of employment growth each month, non-sampling methods must be used to estimate this growth.
It attempts to address this issue with a two-step process:
Based on research that indicates the net contribution of businesses opening and closing as quite small, job losses due to business deaths are excluded (with the assumption that these losses are being offset by gains elsewhere).
An “auto-regressive integrated moving average”, updated quarterly, is used to estimate any impacts not accounted for in step 1.
The potential problems with this are probably quite obvious: if business deaths or births suddenly stop neatly netting out for some reason, this adjustment could skew the headline figures. As SC notes:
The QCEW, by contrast, directly measures firm openings and closings from administrative data…
We think the B-D model gave a very different estimate of net job creation by new and closing firms than the subsequent QCEW observation in Q2-2022, and we think this likely occurred again in Q3-2022.
“[V]ery different” is perhaps shown better by this chart:
Englander and Hewin continue (our emphasis):
The Q3-2022 B-D adjustment added 294,000 NFP jobs. This seems a little high to us, as the QCEW had not showed this many Q3 net job gains from openings less closings other than in 2020 during the initial recovery from COVID…
We cannot preclude that the H2 NFP data is correct, but we think it is – most likely – somewhat overstated. That would mean a decent downward benchmark revision in employment levels when benchmark revisions are implemented at the beginning of 2024…
A regression analysis suggests that QCEW net job creation from opening and closing firms comprises about 12% on average of job creation from existing firms that are expanding or contracting (Figure 3). This means QCEW data could suggest that the NFP employment level is too high for Q3-2022, and even that the Q3 NFP employment gains were overstated, leading investors to question the tight labour-market narrative…
On balance, we think the QCEW data is likely to suggest that NFP job gains were overstated in Q3-2022 as well as in Q2-2022.
Anything that might shift the tight labour market narrative would indeed be pretty major in the current environment, as persistent apparent hiring strength and low-trending unemployment are the main reason Jerome Powell keeps having to crack out his grumpy face.
Will the BLS change its approach to integrate QCEW? StanChart certainly think it should:
To be clear, it is not a criticism of the BLS that the B-D model estimates can be off. The alternative of ignoring firm births and deaths is unlikely to be more accurate. The BLS B-D adjustment captures typical net job shifts from openings and closings but cannot capture actual job creation. There is bound to be error…
What puzzles us is why the BLS does not update NFP with QCEW data as the QCEW data becomes available. It is understandable that the BLS uses a model estimate when direct estimates are not available, but less clear why QCEW reads on net job creation from opening and closing firms are not integrated as they become available. It would make US data more coherent, in our view.