Thursday, October 11, 2012

Jobs Report Statistically Impossible

Jobs Report Statistically Impossible
Let's do a statistical analysis of that infamous September jobs report that Rick Santelli tore apart the other day in his epic on-air rant. The report showed initial unemployment claims of 339K. That's nearly 6--six--standard deviations away from the mean consensus estimate of 368K. For those unfamiliar with statistics, that's a lot. It's so much, in fact, that the probability of such a statistical anomaly occurring is well below 1%. (6 standard deviations is about as close as you can get to 0% without reaching it.)

Jobs Report Statistically Impossible - chart 1

If that didn't knock you over the head, there's more: The divergence between the reported ("headline") weekly numbers and the revised numbers keeps getting larger. As of the last report, the BLS has added 200K initial claims to its revisions. This would suggest that when the September jobs report is revised, several thousand initial claims will be added to the reported 339K.

Jobs Report Statistically Impossible - chart 2
 
But there's still more: The last time we saw such a large month-to-month drop in initial claims was February 2006, when layoffs were unusually low (because we were at top of the housing bubble) following typical seasonal layoffs in January.

Jobs Report Statistically Impossible - chart 3

Add all this together, and you get a September jobs report that's not a statistical improbability; it's a statistical impossibility.