by danielhanchen
You might have saw jobless claims spiked from 198K to 228K, and much higher than 200K estimated. Worse long term unemployment spiked dramatically from 1,689K to a crazy 1,823K, hugely higher than 1,699K estimated.
Below RED is the new revisions, which I think revised the past 5 years of data, and BLUE is the OLD data from last week. You can see why there’s a spike – it’s not due to an increase in job losses, but rather due to a new seasonal adjustment method.
In fact, jobless claims technically DECREASED (RED) using the new metholodogy. July 2022 to December 2022 shows jobless claims was revised dramatically downard.
Jobless Claims RED=new. BLUE=old.
As per www.dol.gov/ui/data.pdf:
t says that OUTLIER effects were adjusted. There are 2 main models for seasonal adjustment – MULTIPLICATIVE and ADDITIVE. Multiplicative models are good at modelling changes when the economy is in reasonably calm phase, but will screw up in huge spikes. So, an Additive model is used for recessions / COVID.
BLS is still using an additive model I think (so presumably they still think COVID is causing a distortion in labor markets), except they changed the definition of what is an outlier and what isn’t.
The more shocking data release was long term unemployment claims, which came in at a super large 1,823K vs 1,699K estimated. The previous week was 1,689K which got revised UP to 1,817K.
Again below RED is NEW seasonally adjusted data, and BLUE is OLD last week model. The wild increase again is not that dramatic considering the seasonal adjustments. However, one interesting point is the trend is going up.
Continuing Jobless Claims RED=new, BLUE=old
TLDR: Jobless claims did NOT increase, but DECREASE, however, the new seasonally adjusted factors has increased the trend of jobless claims. Likewise, the sudden jump in continuing claims is also fake. HOWEVER, the trend of claims and continuing claims is not good. In fact, claims have NOT stagnated as shown in the old model, but from February 2023 (pre SVB et al), claims already ticked up.