Forecasting the Payroll Employment Number
Each month credit and equity market observers alike pounce on the Employment Situation Report from the Bureau of Labor Statistics. Markets gyrate wildly when the actual result is 40,000 or 50,000 payroll jobs away from the economic forecasts.
This is a big mistake. Here is the problem. The level of change that the market believes to be significant, is not great enough to be measured by the BLS tools. It is like measuring a hair with a grade-school ruler.
And that is just the measurement issue. It is even more difficult to forecast the results. We always enjoy reading Bob McTeer, whose blog is one of our featured sites. His recent commentary notes that economists are not very good at forecasting.
Unfortunately for the reputation of economists with the general public
is that they think the job of economists is forecasting. That's
unfortunate because economists can't forecast very well. It just can't
be done. Economists know that, and most will admit it, but they are
stuck with it, especially if they wish to be employed as an economist
His main point is that these predictions are difficult. People making forecasts should provide probability figures and ranges. (For a similar argument, check out Barry Ritholtz on The Folly of Forecasting). The problem is that when the forecaster is honest about the error range, the consumer becomes less interested. Well, so be it!
McTeer has a number of excellent points on using models and some Greenspan stories, so enjoy reading the entire article. After noting that most everyone is a "closet extrapolator" he writes as follows:
I've always been amazed at economists who presumed not only to see
around a corner, but to see around more than one corner, as in "the
economy will pick up through year-end, then slow for several months,
and then take off again stronger than ever." To me that sounds as
silly as what they would say to the waiter after sniffing a wine cork.
It is something to think about the next time you hear some elaborate, multiple-turn economic or market forecast.
So with this in mind, here is our monthly update on the employment report, (also appearing in the Columnist Conversation on RealMoney).
The Employment Report
Each month we take a look at the likely change in payroll employment
jobs, given other economic data. This is not really a forecast, since
the data are all contemporaneous indicators of current economic
conditions. We simply use the already reported data to ask what we
expect from the payroll number.
Our model takes Michigan Sentiment (hitting new lows), the ISM
index (hanging in there at just under 50), and the four-week moving
average of initial jobless claims (weak, but not recessionary). Be
careful not to consider today's jobless claims numbers. The non-farm
payroll report is based on a survey where respondents are asked to tell
about employment during the week including the 12th of the month.
Revisions to the report come because many employers do not submit the
information on time. Some never report, since for
many there is no legal requirement to do so. The BLS tries hard to make
it easy, and get a complete response.
Our approach suggests a loss of 82,000 jobs, weaker than
consensus guesses of -60K or so, and much weaker than the ADP estimate.
It may surprise readers to learn that losses in this range are still
consistent with modest GDP growth of 1% to 1.5%, as are the other
economic numbers. I don't recommend any big bets, since the 90%
confidence interval on the survey is +/- 100,000 jobs! And that is
after all of the reports are in and revisions done. The excessive
attention to this report creates a dangerous trading situation. You can
make your own (risk-free!) guesses at our Payroll Employment Game site. This shows you that even if you know the answer in advance, the
reported number can be far away. The "official" answers will eventually
converge on the "truth" but there is plenty of error along the way. In
addition to this resources, you can check out some other little-known
facts in this prior article which has links to other references.