Tomorrow's employment report

Private sector payroll employment rose by 151,000 in October, well ahead of forecasts. Now the consensus forecast is for another gain of 145,000-168,000 in November. I think we're at another one of those points when the forecasters are a step or two behind the curve; for reasons laid out in a previous post I think we're back on track for strong employment gains in the months ahead.

I got an accurate forecast of employment in October by looking at the ISM reports, but the ISM non-manufacturing report for November has not been released yet. The manufacturing report was positive but not quite as strong as in October, but then again manufacturing is relatively small fraction of total employment.

We do have information on initial unemployment insurance claims and the ADP figure for November. ADP says private sector employment growth was 93,000 in November. All year long the ADP has been underestimating private sector payroll employment by 50,000 to 200,000. The average understimation is somewhere around 70,000, so that would imply +163,000 for private sector employment. Because of underestimation, ADP employment is now about .98 of BLS private sector employment, so if I inflate up from the level of ADP employment I also get something in the +150,000 to +250,000 range (the numbers are very sensitive to my assumption of the scaling factor).

Initial claims for unemployment insurance have been falling for the last month or two. If I run a regression of the change in private employment on its own lag and the current and two lags of initial claims and then use that regression to forecast the change in private employment in November, I get +175,000.

So let me express my forecast as a probability distribution, based on the above computations, my budding optimism of the economy, my respect for the uncertainty involved in these things, and the seat of my pants:

Growth in private sector employment in November:
100,000 < x < 150,000 (prob = .2)
150,000 < x < 175,000 (prob = .2)
175,000 < x < 200,000 (prob = .2)
200,000 < x < 225,000 (prob = .2)
225,000 < x < 250,000 (prob = .2)

Prob (x>250,000 or x < 100,000) = close to zero.

I guess that makes my mean estimate somewhere around +185,000.