The BLS releases data on employment in the previous month towards the end of the first week after the month ends. So last Friday morning we got a whole bunch of new data, including the unemployment rate and the number of new jobs. Those numbers are kept secret from everyone (even the White House and Congress) until they're released publicly to everyone at the same time.
And everyone who pays attention to this sort of thing was ... flabbergasted. Especially with the jobs number.
The report said the jobs market was fantastic in January. For example, there's "‘I’ve Never Seen Anything Like This!’ CNBC Absolutely Loses It Over Blockbuster Jobs Report Showing Half-a-Million New Jobs"
But everyone's expectations were that it would be really bad (due to the omicron surge, which really started to hit the week before Christmas, peaked in mid-month in the northeast, and which is only peaking just now in certain parts of the country). The report was expected to be so bad that on Monday, the White House Press Secretary Jen Psaki admitted as much and tried to talk down the press pool about what to expect (follow this link, and then search in that webpage for the word "frontrunning"). Here's Bloomberg from Thursday "U.S. Jobs Data Look ‘Particularly Tricky’ on Omicron, Revisions"
(Read more at: https://www.bloombergquint.com/onweb/u-s-jobs-report-set-to-be-a-doozy-with-omicron-revisions
Copyright © BloombergQuint").
As to the response, it's not normal for the Secretary of Labor to have to go on the cable news shows to defend their data releases (see "Labor Secretary Marty Walsh on US job growth soaring past expectations: 'Very transparent number'"). Or for quick posts that try to explain seasonal revisions (see "Yahoo Finance VideoWhy jobs reports revisions happen and what they mean"). In "Here Is What's Behind Today's Stunning Payrolls Beat" ZeroHedge went so far as to note that the released numbers were 3 standard deviations above the mean of several dozen publicly available forecasts of how the numbers would look.†
***
The conclusion that people came to is that the data is anomalous because of seasonal adjustment (in the commonly quoted data) applied to non-seasonally adjusted data (buried on the BLS website).
Everyone likes to use seasonally adjusted data because 1) it gets rid of seasonal swings, which are often larger than the changes you're interested in, and 2) it's a lot easier to use some data that someone else did the prep work on.
But, it's a dirty little secret in time series analysis that a lot of seasonal adjustment done by government agencies is ... junk. The reason for this is continuity. When data was first getting measured, and seasonality was noticed as a problem (say 75 years ago), they passed it off to engineers (hey, you guys do math!) and computer programmers (hey, you guys fix buggy stuff!) ... who came up with non-statistical, non-economic, and non-financial ways of adjusting things. (They even named the big government seasonal adjustment program ... X11 ... which sounds impressive and stuff). Later on, time series was developed as a field (starting in the 60's). The methods they came up with for dealing with seasonality were more solid, and not surprisingly, entirely different. But governments have hung on to those older methods to maintain continuity. In short, sometimes sh*t happens in going from non-seasonally adjusted to seasonally adjusted data (NSA and SA for short). Beware.
***
But, the cool thing is, you've learned some tools over the last month that can be used to seasonally adjust data. I don't normally do this in ECON 3020, but it's a lateral shift rather than a big leap. So we'll do this in class next week.
† Hopefully you've heard in business discussions (often about quality control) the phrase "six sigma". In statistics, sigma is the standard deviation. Six sigma refers to plus or minus 3 standard deviations around the mean: an observation should fall outside a ± 1 standard deviation range about a third of the time, outside a ± 2 standard deviation range about 5% of the time, and outside a ± 3 standard deviation range ... just about never (more precisely, 1 in every 275 times).
Update: Here's a quote from Phoenix Capital Research's briefing "There's fiction ... and then there's the January jobs report". And here's one from Rabobank "You just saw history being made."
No comments:
Post a Comment