How to Deal with Economic Data: The Example of the Employment Situation Report

Each month the market focuses on the Employment Situation Report.  At "A Dash" we have studied this carefully, perhaps more so than anyone else.  Understanding this report requires a combination of methodological skills.  These include research design, survey methodology, time series analysis, and economic forecasting.

Most observers have some, but not all, of this background.  As a result, most observers make various mistakes.

Here are our main conclusions, which you do not see anywhere else:

  1. The basic method is an attempt to estimate the total number of jobs in the US in one month, the total number in another month, and subtract to get the difference.  This method, even if done very well, has a built-in error range.
  2. The sampling error from the survey is about +/- 100K jobs.  Most market watchers infer false meaning from changes less than this.  Check this out at our Payroll Employment Game site.
  3. The initial report has only partial information from businesses, leading to later revisions.
  4. The final results are benchmarked against actual data from state reports.  These are much more accurate, since no one pays taxes unless required.  These reports pick up the new businesses, but are available only after a delay of six months or more.
  5. There is non-sampling error, because new businesses are not sampled.  Most observers have an inaccurate fixation on this part of the process — the birth/death adjustment.  This has actually been very accurate, but the conventional Wall Street Truthiness holds the job creation number to be wrong.  These observers have not studied the data.

Our "Forecast"

Here is a brief explanation of how we estimate the monthly change, as reported Monday on RealMoney:

My monthly employment model is a pretty good fit as these things
go, but there is plenty of luck involved in each specific month. The
90% confidence interval is +/- 100K, and that is on the data as
ultimately revised. I look at three pieces of data collected at the
same time as the jobs surveys (the week including the 12th of the
month). These are concurrent economic indicators including the ISM
manufacturing survey, (terrible at 32.4) University of Michigan
Sentiment Index (awful at 60.1), and the 4-week moving average of
initial claims as of the relevant week (disgusting at 556,000).
The job loss consistent with those other data points is around 580K,
about 100K greater than I am seeing from other sources. With the error
band, it could have a "6" handle. A positive surprise seems unlikely. I
suspect that some economists have a similar method, since the whisper
numbers have moved more negative (and closer to mine) in each of the
last few months.


The data represent a snapshot from three weeks ago.  It should be no surprise that things were very bad during the fourth quarter.  Today's ADP report had an estimate of job loss even greater than ours, by more than 100K jobs.

The question for investors is not whether things were bad in the fourth quarter, since we all know that.  Normal business activity drew to a halt in October, when the commercial paper market and other short-term lending hit a complete freeze.

Instead, the question is whether the various government programs have improved normal lending and helped conditions.  Employment data, like most other economic statistics, are coincident indicators at best.

When will the market look forward?  We do not know, but we suspect that a bit of evidence will be needed first.

You may also like

One comment

  • David Merkel January 7, 2009  

    Jeff, I think one thing that could put the nail in the coffin of critics comes from point four, which is new to me. Could you summarize the data showing how the final raw figures (prior to seasonal adjustment) track the state results? That would give me a lot of confidence in the results.