Everything about the Birth/Death model you wanted to know but were afraid to ask

  Via Mike Shedlock or MISH. This is some really good stuff. Some I Knew, but some I did not. It should help understand what goes into the cocktail that is the employment report.

BLS Birth/Death Model Yet Again

The BLS Birth/Death model is an endless source of confusion. It's actually no wonder.

This is what we know about the Birth/Death Model.

1. The BLS does not adequately disclose how they determine business births or deaths, nor does it disclose precisely how their seasonal adjustments work.
2. The BLS does admit the Birth/Death model is wrong at economic turns.
3. The BLS recently changed from annual model revisions to quarterly model revisions which makes historical comparisons of Birth/Death numbers invalid. For example, January and July used to be the only negative months, but in 2011, January, September, November, and December were all negative.
4. The BLS applies non-seasonally adjusted Birth/Death numbers to non-seasonally adjusted nonfarm employment numbers, then seasonally adjusts the result and spits out seasonally adjusted numbers widely used in mainstream media reports on the first Friday of every month.

For the purpose of this discussion, the key point is number 4. Failure to understand that point leads to huge mathematical errors.

Month in and month out people subtract seasonally non-adjusted numbers from seasonally adjusted headline numbers even though that is statistically invalid. 

Please consider a table I put together that shows the influence of  the Birth/Death numbers on the non-seasonally adjusted numbers.

Not Seasonally-Adjusted Nonfarm Numbers and Birth/Death Adjustments January 2011 Thru August 2012 (in Thousands)

DATE	Nonfarm Employees	Birth Death Factor	Monthly Change in Nonfarm	B/D Percentage of Monthly Change
2011-01-01	128327	-339
2011-02-01	129148	112	821	13.64
2011-03-01	130061	117	913	12.81
2011-04-01	131240	172	1179	14.59
2011-05-01	131889	211	649	32.51
2011-06-01	132340	141	451	31.26
2011-07-01	131038	5	-1302	-0.38
2011-08-01	131278	89	240	37.08
2011-09-01	131975	-26	697	-3.73
2011-10-01	132870	116	895	12.96
2011-11-01	133172	-30	302	-9.93
2011-12-01	132965	-1	-207	0.48
2012-01-01	130297	-397	-2668	14.88
2012-02-01	131210	91	913	9.97
2012-03-01	132081	90	871	10.33
2012-04-01	132945	206	864	23.84
2012-05-01	133723	204	778	26.22
2012-06-01	134057	124	334	37.13
2012-07-01	132840	52	-1217	-4.27
2012-08-01	133092	87	252	34.52

In percentage terms, over the past two years, the influence of the birth/death adjustments on the month-to-month change in non-seasonally-adjusted nonfarm numbers varies from zero to as high as 37%. That is quite a range and clearly the influence can be significant.

In August, the reported seasonally-adjusted job report showed 96,000 more jobs than in July. The average person makes a huge mistake assuming 87,000 of those 96,000 jobs came from the Birth/Death model.

Last month the Birth/Death factor was 52,000 jobs but the month-over-month non-seasonally-adjusted change was a whopping 1,217,000 jobs. Thus, the Birth/Death effect for the month (on the seasonally-non-adjusted jobs number) was only about 4%.

The Birth/Death numbers I plugged in for January, February, and March of 2011 are from archives and may not be valid. The latest Birth/Death benchmark was March 2012.

Except at economic turns, I suspect but cannot prove, that BLS Birth/Death errors are reasonably small.

However, at economic turns, the Birth/Death model can potentially be way off in terms of both magnitude and direction. Such errors may be considerable, and those errors are cumulative! The result is the huge backward revisions in reported job numbers from time to time.

I believe the economy is indeed at an economic turning point. If so, expect more huge revisions in the Birth/Death adjustments as well as the seasonally and not-seasonally-adjusted job numbers, at some point in the future.

In the meantime, if you want to make any sense of the Birth/Death numbers, you have to start with the non-seasonally-adjusted reports, not the widely-reported seasonally adjusted headline number.