The Atlanta Fed GDPNow Forecast declined from 3.4% on February 1 to 2.7% on February 7.

Roughly 0.3 percentage points of the decline was due to a slowdown in auto sales. The rest was split between the non-manufacturing ISM report and last Friday’s jobs report. But how?

I asked Patrick Higgins, creator of GDPNow “Is it possible for you to break out the employment report from non-manufacturing ISM for February 3?

His answer follows:

Hi Mish

Using the data as of February 3rd except for the ISM Nonmanufacturing Index, the model forecasts first-quarter real GDP growth of 2.653%. This is only slightly below the published February 3rd forecast with the ISM Nonmanufacturing Index of 2.666%.

The composite ISM Nonmanufacturing Index is the only series from that report used to estimate the model’s dynamic factor while there are just over 30 series from the employment situation report used to estimate the dynamic factor. So a lot more data from the employment report gets fed into the dynamic factor model.

For reference, here are the January 2017 values of the dynamic factors for the last 3 days in which there was a GDPNow posting.

Values of January 2017 dynamic factor in GDPNow [mean = 0, standard deviation = 1]
Jan 30th (forecast): 0.05
Feb 1st: 0.87
Feb 7th: 0.37

Best regards,

Employment Analysis

I expected that response as noted earlier today in GDPNow Forecast Sinks to 2.7% from 3.4%: What Happened?

The non-manufacturing ISM report posted a reading of 56.5, inline with consensus and generally thought to be strong. The employment component rose from 52.7 to 54.7. Business activity was a solid 60.3.

If we rule out ISM, the 0.4 percentage point decline on February 3 had to have come from the employment report. However, the employment report was also widely believed to be strong.

The headline job number came in at beat-the-street estimate of +227,000 jobs.

My assessment was much different: Shocking Fact in Today’s Job Report: Employment Stalls

What Else Was Bad in the Jobs Report?

I do not have an itemized breakdown of the 30 items in the jobs report that led to the 0.4 percentage point drop. But let’s consider 11 numbers from my Employment Stalls article.

BLS Jobs Statistics at a Glance

  1. Nonfarm Payroll: +227,000 – Establishment Survey
  2. Employment: -30,000 – Household Survey
  3. Unemployment: +106,000 – Household Survey
  4. Involuntary Part-Time Work: +242,000 – Household Survey
  5. Voluntary Part-Time Work: +236,000 – Household Survey
  6. Baseline Unemployment Rate: +0.1 to 4.8% – Household Survey
  7. U-6 unemployment: +0.2 to 9.4% – Household Survey
  8. Civilian Non-institutional Population: -660,000
  9. Civilian Labor Force: +76,000 – Household Survey
  10. Not in Labor Force: +264,000 – Household Survey
  11. Participation Rate: +0.2 to 62.9 – Household Survey


  • Points 2 through 10 were an absolute disaster. Point 1 stood alone in excellence. Point 11 seems favorable (more people are looking for jobs).
  • Note that employment actually fell by 30,000 even though job purportedly rose by 227,000.
  • 242,000 more people want full-time jobs and can’t get them.
  • The unemployment rate rose slightly, with U-6 unemployment rising even more.
  • Many of the Household Survey numbers vary substantially from report to report. But this set was universally bad across the board.

To be fair, we have seen these kinds of divergences before, and they resolved in accordance with jobs. But this divergence has gone on since March of 2016.

Employment Stalls


Employment growth was +129,000 on  average from a year ago, +79,000 per month since March, and only +11,000 per month for the last three months.

Employment has stalled. Those who went gaga over the reported 227,000 increase in jobs while ignoring the decline in employment may wish to reconsider.

Then again, let’s look at adjustments.


Population adjustments skewed the numbers badly.


In 2016, the BLS notes that it over-estimated population by 831,000; the labor force by 508,000; and employment by 487,000.

Revision Comments

Instead of revising all the numbers month-by-month, the BLS just lumped them into January.

Purportedly, employment in January would have risen by 457,000 if only the BLS did not previously overestimate employment by 487,000. Got that?

Thus, the jobs-employment divergence could once again resolve with employment rising to catch up to jobs, only to see another adjustment in January of 2018 of a half-million or so, negating a huge portion of the climb.

Looking Ahead

Those focused on the bright side may wish to simply reply on the reported 227,000 increase in jobs.

Everyone else may wish to consider the fact that employment growth is somewhere between lagging and badly-lagging purported jobs growth.

Looming trade wars may not matter. Auto sales might also pick up, and consumers might go on a buying spree even though home sales are down.

Here’s one final point to consider: Huge revisions are the norm just after the economy turns. Anyone foresee a negative 1,000,000 employment adjustment in January of 2018?

Mike “Mish” Shedlock