Jobs Day Tomorrow and A D(isappointing) P(redictor)

January 5th, 2012 at 3:08 pm

Another first Friday of the month, another jobs day tomorrow.

At 8:30AM tomorrow we learn the BLS estimates for job growth, unemployment, and other goodies for last month.  With tomorrow’s release, we’ll also have data for full year 2011, which always provides some interesting comparisons.  I’ll be manning the battle stations first thing and will link to Chad Stone’s always trenchant analysis as soon as it’s ready.

A short note on what to expect…short because forecasting these numbers is a fool’s game.  There’s a pretty wide confidence interval around them—for example, a change in payroll employment of less than 100,000 is statistically indistinguishable from no change at all.

But fools rush in, and especially with the large pop in the ADP number released today (up 325K), along with the recent, consistent declines in UI claims, there’s a potential for an upside surprise tomorrow–here’s a nice review of all of this stuff.  Markets expect about 155K on jobs and the unemployment rate to tick up to 8.7%, giving back a tenth from last month’s larger than expect decline in the jobless rate.

Note the caveats about the ADP report.  It’s tempting to get excited when the payroll agency reports a net gain of 325,000 private sector jobs last month, but you shouldn’t.  At least you shouldn’t expect the BLS data to necessarily show a pop of that magnitude.

It’s not because movements in the two series aren’t highly correlated: the changes in the series have a correlation of 0.95.  It’s because when we’re looking at changes in big numbers—there are over 100 million jobs in the private sector—even small differences will be large.

For example, if you use the change in the ADP to predict change in the BLS private payroll number, you get a very high R-squared (0.90; it’s a measure between 0 and 1 of how well the model fits the data) but a RMSE of 75K (root mean squared error—basically, how much, on average, your forecast differs for the actual BLS number*).

EG, if I used the ADP to predict November’s result, I would have come up with 211K net new private sector jobs.  If I believed it, I could have written that it’s going to be a pretty big jobs day.  But the actual estimate turned out to be just an OK 140K.  Note that I would have been off by around the amount predicted by the RMSE (211-140=71), but you still would have been disappointed, no?

That’s why I always look at the ADP closely, but re the BLS number, I consider it a A D(isappointing) P(redictor).

*An added complication here is that both the BLS and ADP numbers undergo significant revisions each month, so what may look like a bad forecast can turn out to be good, and visa versa.

Print Friendly, PDF & Email