Monday, February 8, 2010

The Breakdwon of the BLS Models

Last week, when the BLS numbers came out, I highlighted the huge upside swing in the birth/death index:
Most bizarre about the January number is that the notorious and mysterious fudge factor, the birth/death index had a huge downward adjustmnet.. The January Birth/Death adjustment was -427,000 from +25,000 in December. Got that? Without the Birth/Death downward index adjustment (which is generally positive every other month) employment would have been up by an incredible 407,000. This no doubt is really an adjustment for those who lost seasonal Christmas related jobs, rather than a sudden January collapse in new business hires, with a good dose of direct hands on BLS adjusting to get a near flat unemployment number for January.
This huge Birth/Index decline is what was probably behind a HUGE seasonal adjustment in the other direction in January to make the unemployment number come out pretty much flat. But the very nature of a seasonal adjustment number means something that was boosted in January must be taken down at some other time period during the year.

Mish thinks the down swing part of the seasonal adjustment will come between March and April. And because it was a dozy of a seasonal adjustment to the upside in January, it is also going to be a dozy to the downside in March and April. In other words, the unemployment numbers could be real bad in March and April. Here's Mish:

•There is always a big BLS adjustment in January

•There is always a reversion to the mean that overshoots to the downside between March and April

•There is always a secondary rebound back above the 0.0% line in July, followed by a smaller overshoot to the downside in October.

The problem is in the increasing amplitude of these swings, in both directions. It really makes you wonder just what the hell the BLS is doing and why.

I have data charted all the way back to 1999. Prior to January 2009, the biggest January swing was .6 percent, in both 2004 and 2003. In 2008 the January swing was only .5 percent.

The amplitude of January swings in both 2009 and 2010 was .9 percent, way outside the data range for the last 10 years, by a factor of 50 percent (.3/.6).

Likewise, the prior swings in October peaked at -.4 percent on a couple of occasions but hit -.7% in October 2009.

Unless it's different this time (I figure it is not) a reversion to the mean that slightly overshoots in a May-June timeframe will lop off a whopping 1.3 percent off the posted seasonally adjusted rate of 9.7 percent just announced.

In other words, all things being equal (no job gains or losses), we could expect to see the unemployment rate approach 11% by May! Of course we have to factor in actual job growth (or lack thereof). We also have to factor in census bureau hiring.

Heaven knows what census hiring will do to the BLS algorithms. Your guess is as good as mine. However, whatever it does, census hiring will also revert to the mean.

Also remember that it takes 100,000 to 120,000 jobs per month just to hold unemployment rate steady. Think that's going to happen? If so (and again discounting census hiring), then you are living in Bizarro world along with everyone else who thinks the unemployment rate is going to come crashing down.
What's really going on here is that the employment models the BLS has been, well, employing have broken down because of the financial crisis. The seasonal adjustment factors along with employment estimates are not reflecting actual movements in unemployment. Put simply, the algorithms are even more off course than nornal, which means the always edgy numbers are becoming even more edgy, with wild swings making it difficult to determine anywhere close to the true magnitude of the unemployment situation, both on the upside and downside. Just keep all this in mind when the numbers are actually released in coming months.

No comments:

Post a Comment