TL;DR: Claims for unemployment insurance are showing signs of minor weakening.
In the rest of this post I cover:
Claims for Unemployment Insurance
Which Industries Are Due for Downward Revisions?
More below chart.
1. Claims for Unemployment Insurance
We spent much of May and June worried about a small deterioration in continuing claims (that was somewhat exaggerated by residual seasonality). Then, in late June, the data started turning around, as growth decelerated (though again, the improvement was exaggerated by residual seasonality in reverse). And now continuing claims are again growing a little faster (which residual seasonality is masking).
These fluctuations are all pretty small - from 4% Y/Y to 6% Y/Y. I and others attached a lot of meaning to them back in late spring when we were worried about them spiraling out of control, but that didn’t materialize. So I’m going to be a little more measured this time. I’d rather not see any acceleration at all, but we’re basically talking about less than 40K in weekly extra continuing claims if it continues for a whole year.
I’m a strong believer that you should not discount any high-quality data point.1 Continuing claims are growing a little faster should make us a little less confident in the “labor market stability” thesis I’ve written about (which is underwritten by the JOLTS and CPS data). But I don’t think they undermine that thesis entirely.
I mentioned residual seasonality a lot in the first paragraph, so a quick reminder: even if continuing claims are indeed deterioration a little faster than they were before the spring, the published series’ trajectory should be pretty flat for the remainder of the year. If that happens it will paint an overly optimistic portrait of what is really happens.
One other data quirk that I’ve finally systemized: 1st prints of the continuing claims data have been getting revised downward by about 9K per week (0.4%). That’s a tiny change, but in a world where we’re trying to tease out the difference between 4% and 6% it’s “hot take significant”.2
Enough about continuing claims. What about initial claims? For about a month, initial claims were running below year-ago levels. I was hopeful that layoffs had come down a little. But that short period of lower layoffs seems to be abating. I’m not particularly worried, but we’ve been used to good news on this front and this is a little less good.
I’ll wrap this up with two of my usual charts. The first is the nowcast of unemployment due to permanent layoff. It’s flattish in August so far, though I’m pondering to what degree seasonality in these two published series is aligned.
The second is the ongoing disconnect between Google searches (historically, a useful nowcast of initial claims) and actual initial claims. I have some unverified hypotheses for this but no actual proof.
2. Which Industries Are Due for Downward Revisions
In a few weeks (September 9th) we’ll have our preliminary estimate for the benchmark revision to the level of nonfarm payroll employment (NFP) in March 2025. This preliminary estimate will be primarily based on 4 quarters of data from the Quarterly Census of Employment & Wages (QCEW). In early 2026, after the QCEW itself is revised, a final benchmark revision will be incorporated into the history of the nonfarm payroll employment time series.
As of June, we have 3 quarters of that data in hand. Historically, that data has been sufficient for a decent forecast of the 4 quarter sum - and currently, that forecast is for a large negative revision. A few months ago I estimated that based on the 3 quarters of QCEW data we have in hand, we should expect a downward revision of about -790K to the level of NFP as of March 2025 - or alternative, a downward revision of about 65K per month to employment growth in the year to March 2025. You can read more about that here. I’d attach wide error bands to that estimate!
A related question of interest is “which industries will be affected by the revision”? I took a crack at estimating this by comparing, through December 2024, the gap between year-over-year growth in NFP-by-industry and QCEW-by-industry. It’s the same method I used for the aggregate employment benchmark revision estimate, but on an industry basis. One caveat is that my aggregate employment revision estimate incorporated historical average misses to come up with the 65K/month number, something I did not do with the industry estimates. So these numbers are probably collectively too negative.
You can see the results below. The gist is that downward revisions will be broad-based, but pain will not be eventually distributed. If you take my estimates very literally, in absolute terms we should see the largest revisions in transportation & warehousing (-129K), accommodation & food services (-105K), wholesale trade (-99K), retail trade (-91K), construction (-65K), administrative & other business services (-62K), and durable goods manufacturing (-61K).
In percentage terms, the ranking is slightly different: transportation & warehousing (-1.8%), wholesale trade (-1.6%), information services (-1.1%), other services (-0.9%), construction (-0.8%), and durable goods manufacturing (-0.8%).
The only positive revision of non-trivial size that showed up in this approach was in management of companies & enterprises (+42K; +1.6%). I wonder if this reflects a consolidation of business activity from satellite establishments into corporate headquarters.
To reiterate, this is a rough directional guide to where we might see the biggest revisions. I’ve never really tested this approach before at the industry level so caveat emptor! I’m looking forward to comparing these numbers to the final results.
Properly weighted, you should not discount lower-but-not-zero-quality data points either.
I’m going to try to make “hot take significant” a thing in the numerate corners of social media.
Great stuff as always, Guy!