Forecasting the Jobs Report

MarketWatch tells us that today's selling was concern about the labor market and the resulting economic contribution of consumers.

We sympathize with journalists who need a daily lead to explain modest market moves.  We are amazed that market participants suddenly started to worry about Friday's jobless report.

"Forecasting" the Employment Report

Here at "A Dash"  we have a good model for payroll employment changes, but it is not really a forecast.  We look at concurrent economic data and ask what change in employment would be consistent with the other indicators.

Other people also make forecasts or predictions.  It is a strange game, because the "truth" does not matter.  Everyone is trying to forecast the BLS version of truth.  Here is how we put it in an explanation from two years ago:

  • The Bureau of Labor Statistics (BLS) does not actually measure
    the change in jobs from month to month!  We know this may seem
    confusing.  The change in jobs is what everyone talks about, but it is
    not what the BLS measures.  They try to estimate the total number of jobs, using survey techniques.  They then compare the estimate from one month to the estimate from the next to calculate the change.

The result: 
They can be great at estimating the total, and still have a huge error
band for the change.  If you want more explanation on this point, we covered it here.

  • The original report is revised for two reasons, but not because the government is cooking the books.  The first reason
    is that many of the businesses in the survey do not send their reports
    in on time.  What a surprise!  Some businesses NEVER respond.  The BLS
    does two revisions, based on more complete returns, and then declares
    the result to be final — for a time.  The second reason for
    revision is that the BLS sample for the survey includes only businesses
    that existed at the start of the year.  The dynamic economy is gaining
    and losing businesses all of the time.  The BLS eventually takes actual
    data from state employment offices and compares it to their own count. 
    They adjust the methodology based upon the actual count, using
    something called a birth/death model.

We now believe that there is a recent negative pattern in revisions.  We think it is related to the seasonal adjustment methodology, and we invited other researchers to collaborate with us in investigating this premise.

Current "Predictions"

As we noted, our own approach looks at several other economic variables from the same time frame as the survey.  Our model is linked to the final data series.  There is no point in trying to model the first report, which is known to be either biased or less accurate.

Given the continuing weak picture in initial jobless claims, University of Michigan sentiment, and the ISM report, we expect May job losses of over 600K, greater than the consensus.

Most economists do not reveal the basis for their forecasts, but there are some exceptions.

Today's ADP data draw upon proprietary information about actual job changes, an excellent source.  It is completely possible that ADP could do better on job changes than the BLS, if only because that is exactly what they measure.  Since the initial BLS report is the "official" number, that has become the ADP target.  ADP sees job losses of 536,000 in the private sector.

Wanted Technologies forecasts a job loss of 565,000.  Their method is a proprietary regression model including online advertising for jobs and prior BLS data.  Their methodology is careful and accurate.  Like the rest of us, they are looking at the final revised data.  Interestingly, they show that their "forecasts" are more accurate in calling the final revision than the BLS does itself from initial data.

Good idea.  We should test that on our own results!

Our Take

It is interesting that our approach and those using different methods are so close in the predictions.  Our own estimates have been too bearish, but later revisions have shown us to be very accurate.  And remember, the sampling error alone is more than 100K jobs.

Whatever the exact number, we are a long way from significant improvement in employment.  We expect the old, big-firm employers to add workers only slowly.  While there is vibrant job creation, misunderstood by most, it is fighting a losing battle with job losses.

You may also like

One comment

  • online surveys August 31, 2010  

    Though theres a report and predictions. We really cant tell if our economy will do better this years evaluation.