Friday, February 6, 2009

U-3: NSA vs. SA

The unemployment report from BLS came out today and it was ugly, even with the headline reporting of the seasonally adjusted numbers. What stuck out at me when I looked into the A-12 table was the difference between the non-seasonally adjusted (NSA) and seasonally adjusted (SA) values. They were respectively 8.5% and 7.6%. Consider the following stories from the last part of 2008:

- "November retail hiring was 53 percent lower than a year ago, when retailers added nearly 458,000 holiday workers, compared with 217,200 hires last month, according to an analysis by Challenger, Gray & Christmas Inc. (Source: Boston Globe - Dec. 5, 2008)

- "Department stores hired 88,000 fewer people this November compared with 2007, and clothing and accessories stores cut 65,000 jobs, according to the Labor Department. (Source: Forbes - Dec. 11, 2008)

There are tons more of these stories if you're inclined to look. If you read a local Picayune, Gazette, or News you'll likely recall similar headlines. So with that backdrop, how valid is the U-3 SA number? As ever on this blog, to the charts!

The first chart looks at the monthly trend in the difference between the NSA and SA numbers. You would expect there to be a pattern because of the whole idea behind the NSA value. However, there is some variation inside in the monthly differences which would be expected. The question is whether or not this is a systematic pattern or just noise. The next chart can shed some light on that question.

This chart requires a little bit of explanation. It uses the difference values for all the January numbers from 1989-2009 and plots them against the U-3 NSA value. The line fits reasonably well with an R2 = 0.805. This certainly seems to imply some correlation with the background unemployment level - a higher U-3 number tends to have a higher difference between the NSA and SA values. I think the stories noted above explain the reason behind this phenomenon.

The SA model would assume that post-holiday season there would be more seasonal workers in the job market just as there should have been more seasonal workers employed in Q4. However, this year that doesn't seem to have been the case. This year, holiday hiring was muted at best and thus the substantial differences between the two measurements. Bottom line: the most recent period does not fit the pattern that the SA modeling assumes.

What to look for in February? Predicting a bump in the U-3 SA number would seem like a no-brainer at this point. The question to me is how much of the difference between the SA and NSA values will be bridged. I would not be surprised to see the NSA number actually hold steady and the SA value to creep up one or two tenths to maybe 7.7 - 7.8%. (see update below)

One other note: the participation rate slid further to 65.4% (SA) and 65.5% (NSA). These are 20 year lows. I've omitted discussion of that for brevity's sake but these are also important to watch as signs of a very anemic labor market. I've touched on that before in my original post to this blog.

Update: I have been thinking more about this and want to revise my expectations of SA unemployment. Here's why: First, the normal SA model appears to expect a fall in February NSA employment figures and compensates upward a bit as a result. I don't believe that this year, there will be a fall in February NSA unemployement and the NSA number is already very elevated. Second, the ongoing mass-layoff stories would point to a higher value. So my revised estimate is 7.9-8.1% for SA unemployment in the next report.


energyecon said...

Mr. S,

Great post - that answers a question that has been bugging me - quantifying the scale of the SA, which 'seemed' bigger than in the past...but show me the numbers. Awesome, and thanks.