A short thread on what to expect from tomorrow& #39;s jobs report.

TL;DR: two metrics that have done well in predicting employment so far are telling two different stories: one a significant deceleration from August, the other a big acceleration. /1
Two high-frequency employment metrics from private companies--one from Homebase, one from Kronos--have generally followed each other so far this recession.

But in August, the two started diverging significantly. Homebase is on the left, Kronos is on the right. /2
Homebase data has flatlined in recent weeks. Pairing their data with UI claims yields a forecast of +290K jobs in September, not seasonally adjusted.

Kronos meanwhile has accelerated. Using them instead leads to a +2.4 million forecast.

For reference, August was +1.4 million /3
As you can see, there& #39;s no great way to adjudicate these forecasts using historical data. Both measures have done well in the past. My attempt to mix the two with UI claims using a principal component analysis, perhaps predictably, almost exactly splits the difference. /4
So why are the two measures suddenty so different? Well, for starters, they& #39;re two different companies with two different samples.

But also, there& #39;s a fundamental difference in how the two indices are constructed that may explain the gap. /5
The Homebase data is based on a fixed panel of businesses that are present throughout the entire historical period. If Homebase gets more clients, those businesses are *not* added to the Homebase sample. If a client *leaves* Homebase, they& #39;re dropped from the entire sample. /6
The opposite is true of the Kronos data. As clients sign up with or drop Kronos, they& #39;re added to or removed from the index in real time.

So what implication does this have? /7
In an economy where *rehiring* among existing businesses is the primary driver of employment growth, you& #39;d expect the Homebase methodology to do better. It& #39;s a fixed sample and is unaffected by how popular Homebase becomes as a business product. /8
But if the dynamics change and hiring among *new or reopening businesses* becomes a bigger factor, Homebase might start underrepresenting jobs growth, and Kronos might do better, because the latter& #39;s sample allows for new entrants. /9
Unfortunately, we don& #39;t have a great handle on business formation in real-time. High-frequency @uscensusbureau data suggests business applications that will likely lead to payroll employment are falling, but still above pre-COVID trend. /10
Some other metrics are consistent with flatlining. Mobility, for example, as summarized by the Dallas Fed& #39;s index, hasn& #39;t improved much recently. Womply data suggests there hasn& #39;t been a rise in the number of small businesses actually open. /11
UI has show a bit of movement, though some of the fall in regular claims has just been benefit exhaustion and subsequent shifts into extended programs. /12
There are also other possible confounding factors. For example: this year Labor Day fell during the CPS reference week, an event which only happens once every 7 years. /13
Looking at all the high-frequency data, I lean towards September coming in slower than August but still positive. But there is an upside risk case around that too (and, as always, a downside one). /FIN
You can follow @ernietedeschi.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: