The Mystery of Birth and Death Unravelled

May 5th, 2008 8:46 am | by John Jansen |

 I have never published an entire bit of commentary from a Street economist. What follows is in its entirety a note to customers from JPMorgan economist Michael Feroli. It is a clear explanation of the Labor Department’s methodology and debunks the conspiracy theories of those who see manipulation of the numbers. It begins here:

The adjustment to the April payroll figures to account for business births and deaths has stirred up more than the usual amount of commentary. One frequently heard claim is: “The BLS birth/death model added 267,000 jobs to April payrolls, but since this model is not cyclically sensitive it is overestimating employment, without this adjustment private payrolls would have fallen by 296,000.”  

There are several problems with this line of argument, but the most important point to remember is that the BLS birth/death adjustment is a two-stage process, and only the magnitude of the second-stage adjustment is reported every month.  Any public commentary regarding the size of the overall birth/death adjustment is inherently flawed because the public only sees part of that adjustment.

The second stage of the adjustment process, the net birth/death model, is reported each month. However, the first stage of the adjustment process is not publically reported and is usually quantitatively more important.  BLS reports that on average the first-stage adjustment process accounts for about 75% of the overall adjustment.  Moreover, this first-stage process *is* cyclically sensitive. Given that, we can said a few things about the rumors floating around:

  • The BLS birth/death adjustment did not add 267,000 to April payrolls.  The second stage of BLS’ two-stage estimation process, the net birth/death model, did add 267,000 to non-seasonally adjusted payrolls.  But as BLS relates, it is perfectly feasible that the first-stage adjustment subtracted more than 267,000 jobs, leaving an overall negative birth/death adjustment.
  • The BLS birth/death model is cyclically sensitive.  The first stage of BLS’ birth/death adjustment (the more important stage) uses information on the growth of payrolls at ongoing businesses (in this case, those in business in both March and April) in estimating employment growth from business births.  When ongoing business expand faster, the birth adjustment is larger, and the reverse when ongoing businesses slow or contract.  (A numerical example is given at the end of this email).
  • The second-stage birth/death adjustment is added to non-seasonally adjusted payrolls, adding this adjustment to the seasonally adjusted number has no meaning. Given that, the only thing that can be said is: excluding the second stage adjustment, nsa private payrolls would have expanded by 439,000 in April. But this is not particularly informative as nsa private payrolls almost always expand in April.


The JPMorgan forecast for (seasonally adjusted) April payrolls, -90,000, missed by 70,000. Blaming this miss on a faulty BLS birth/death model which doesn’t account for the business cycle is not only bad sportsmanship, it is bad analysis.

*********************

A (mildly tedious) numerical example .

The first stage of the BLS birth/death model excludes employment losses from business deaths in order to offset the missing employment gains from new business births. This is achieved by calculating the over-the-month change in the sample solely from continuing businesses.

Consider the following example: total employment in the last month is 5,000. The sample size in the last month is 1,000. Firms that account for 100 jobs in the sample go out of business. Firms employing the other 900 stay in business.

First, assume the ongoing firms expand 5%.  Using the BLS first-stage method, the current month employment is grown out at 5% to a level of 5,250. If instead the entire sample were used rather than just the continuing establishments, employment would contract by 5.5% (total jobs of 945 in the current month versus 1000 in the last month) to a level of 4,725.  The difference between the current period employment levels in the two different estimation methods, 525, represents the imputed job gains from business births.

Now, assume instead ongoing firms contract 5%.  Once again, using the BLS method, total employment is contracted 5% to a level of 4,750.  If instead the entire sample were used rather than just the continuing establishments, employment would contract by 15.5% (total jobs of 855 in the current month versus 1000 in the last month) to a level of 4,275.  Again, the difference between the two current month employment levels using the different methods, 475, represents the imputed job gains from business births.

As can be seen by comparing the first to the second example, imputed job gains for business births are greater in an economy where ongoing firms are growing (525 jobs in our example) compared to one where ongoing firms are contracting (475 jobs).

Finally, BLS does not publicly report the magnitude of the first stage adjustment because in real time it is generally impossible to distinguish firms going out of business from non-respondents that are going concerns.  Both are treated the same in the sample and the difference is only know with a lag of about a year when the complete unemployment insurance tabulation is available.
————————————————————–
Michael Feroli
US Economist, JPMorgan Economics
tel: 212-834-5523
e-mail: michael.e.feroli@jpmorgan.com
JPMorgan’s Full Research Disclaimer ->
https://mm.jpmorgan.com/legal/credit_research_disclaimer.html

Be Sociable, Share!
  1. 4 Responses to “The Mystery of Birth and Death Unravelled”

  2. By vex on May 5, 2008 | Reply

    I don’t get this. The publically-reported adjustment is an described as an ARIMA model, which includes a periodic (seasonal or possibly cyclical component). That’s one stage, unless you count differencing as a separate stage. Or is the 1st stage something else, another method?

  3. By S on May 5, 2008 | Reply

    If the model was so good why not publish tboth stages. This way people can make up their mind as to what exactly is going on. This is a pathetic defense of the model. You don’t see the first half so don’t second guess the final result. Worse than useless.

  4. By molten_tofu on May 5, 2008 | Reply

    The only concern I have is that job gains and losses are well correlated with firm deaths as well as births. That is, your assumption that the same number of firms go out of business whether the existing firms add or shed jobs isn’t necessarily true.

    For instance, say that in bad economic times firms representing 200 employees go out of business, instead of just the standard 100. Then, it looks like the calcs work out to 760 jobs out of 1000, or a 24% decline, resulting in 3800 jobs, or an imputed addition of 950 jobs due to births (instead of 475).

    Do what you will with that implication. I can barely claim to know the whole story.

    I also really wonder what the curvature of firm birth/deaths against existing firm employment is like, as well the the curvature of firm birth/deaths against industry-specific supply/demand.

  5. By husk on Jul 7, 2008 | Reply

    1) By not publishing the first stage the theory seems to “just trust us”.

    2) “Finally, BLS does not publicly report the magnitude of the first stage adjustment because in real time it is generally impossible to distinguish firms going out of business from non-respondents that are going concerns. Both are treated the same in the sample and the difference is only know with a lag of about a year when the complete unemployment insurance tabulation is available.” This says the stated employment rate is a myth that can be corrected after a year or in short useless except for historical review.

Sorry, comments for this entry are closed at this time.