Back to Zero
by Tom Bozzo
The unexpected drop in the unemployment rate for January made me more than usually curious about the household survey results, and things there actually look OK for a change. The flat unemployment rate between November and December ’09 was the less-than-virtuous result of declining labor force participation pacing the decline in employment. This month’s decline in the unemployment rate reflects increasing labor-force participation and employment-to-population ratios; unemployment levels and underemployment rates [1] are also down. The unemployment decline appears to be statistically significant based on the BLS’s (inexact) guidance on sampling variability in household survey estimates.
The headline employment figure, in contrast, is not a statistically (or qualitatively) significant result, hence the summary’s “essentially unchanged” language. The everlovin’ net birth-death model is subtracting more jobs this month than it did in January ’09 — -427,000 jobs vs. -356,000 a year ago — so if we’re actually at a turning point expect this adjustment to be a drag on measured employment. The mild upturn in manufacturing employment follows other strong data from that sector, and it’s also not unexpected to see temporary employment accounting for the measured service sector growth.
Brad DeLong has been scratching his head over seasonal adjustment to the unemployment claims series; I’m wondering about what’s showing up as a December employment dip. The not-seasonally-adjusted data usually features a small November-to-December drop. Last year’s -538,000 was less than in ’08 (around -1 million) but more than previous years. A December dip after a strong November doesn’t feel right, so go figure.
Calculated Risk and Spencer have graphs and much more discussion.
[1] I expect the last paragraph of the Times story to be revised later to reflect this.
The 1.36 million benchmark revision to the establishment data makes the seasonnally adjusted 20,000 job drop look like a drop in the ocean. The headline number could easily be off by an order of magnitude.
Underemployment rates are only down on a SA basis. The contrast in U-6 is interesting:
SA Dec: 17.3
SA Jan: 16.5
NSA Dec 17.1
NSA Jan 18.0
Forgive me for thinking that the second derivative is going to remain negative for a few more months, and that Brad DeLong’s “equinox” prediction is more likely to be Autumnal than Vernal (and, if I were betting, Vernal 2011 at best)–unless you’re talking about Job Creation in the abstract, without considering population expansion.
@Ken: You mean first derivative. The benchmark revision that the Sandwichman mentions imply, for someone in a glass-half-full mood like me (today, anyhow) that the second derivative improvement was stronger than prevously suggested.
That said, I don’t disagree at all with the ‘drop in the ocean’ comment. I think I have good cause to be happier with this month’s report than with last month’s, but clearly we need a long string of very strong employment reports just to get out of the hole.
The birth/death model is typically a counterweight to new payroll employment trends, so the point made here has a good chance of being correct. Not, however, because of the big downward adjustment to the plug for January. The economists at BLS explain that they have adjusted their model to reflect to great optimism since Q4 of 2008. That means the net B/D adjustment for this year is going to be more negative than it would have been without revisions based on that miss. However, the big change in the January B/D adjustment actually reduces absorbs a good bit of the revision for the year. The guys at BLS note that the miss in January of last year was extraordinarily large, leading to a very large adjustment to this year’s January plug. Given any sized shift in the model toward more negative plugs for the full year, the larger the negative shift in January, the less there is for the remaining months.
BLS also recognizes that the big misses in the plug through the period of the recession disproves one of their assumptions. They had assumed that hiring from new firms and layoffs from failed firms were not very sensitive to turns in the economy. Don’t know why they assumped that, but they did. There is probably a rewrite coming to make the model more sensitive to turns. That may be accomplished by giving the current or recent payroll count a bigger role in the B/D model.
I sympathize with whoever at BLS got to write the FAQ line to the effect of, “No we can’t measure the contribution of the net b/d adjustment to SA payrolls so for pete’s sake don’t ask us.” I gather that at least one source of complexity is that the effective SA factors aren’t independent of the net b/d adjustment. I’d like to think that the ‘normal seasonal variation’ is not that sensitive to the net b/d plug(s), not least because I wouldn’t expect the b/d adjustment to have small sampling variability even if its performance at turns of the cycle improved greatly.
If that’s so, then variability of b/d can be viewed as being influential on the SA headline figure (since the -20K is equivalently the NSA payroll change being ~0.7 percent lower than expected; that’s small relative to past variability of the total January b/d plug). Iif the b/d adjustment “learned” a bigger downward January adjustment from last year’s experience, then that seems to reinforce the claim that it performs poorly at the turns.
My bottom line is there is far more statistical noise than signal in the January report. So wait for next month… but don’t hold your breath.
didnt we expect a bounce in employment with 1.2 million census hires? anyone have a number for those?
Sandwichman:
agreed.
Census employment is included, but BLS data show that it only accounts for +9K jobs over December ’09.