One of the things we have harped on for quite a long time here at The Big Picture is the flawed BLS Birth Death Model (BDM).
Since 2003, the B/D adjustment has been part and parcel to BLS’ Current Employment Statistics (CES) program, the official measure of US employment.
In brief, the Birth Death adjustment imagines (hypothesizes) how many jobs were created by companies too new and/or too small to participate or be found by CES. The model attempts to create what is perceived as a BLS error at the start of any recovery, when many new jobs are created but missed by BLS.
But in fixing one problem, they created another: At the other end of the cycle — where we are today — the B/D adjustment potentially will hypothesize lots of phantom job creation. That explains the stability in construction and finance jobs in the monthly data. (See Bloomberg’s Why Haven’t Home Construction Jobs Disappeared)
I suspect the misguided attempt to reproduce this modeling error helps explain yesterday’s baffling ADP data.
Ray Stone of Stone & McCarthy Research Associates notes what is the most serious defect of the BDM: “At the same time we recognized the limitations of the BDM. The most serious limitation of the BDM is that it is a time-series model, and as such, does not pick up turning points or inflection points.”
And that is very likely where we are today.
To give you a better idea of how badly the B/D is currently skewing the data, consider these charts below (via Econbrowser). Looking at the changes of the past 3 years, its apparent that the B/D model went from being a modest portion of the CES data to being the increasingly dominant source of reported new jobs over the past 12 months:
>
chart courtesy of Econbrowser
>
Indeed, the actual newly created jobs that are measured — and remember, it is a supposed to be a survey measure of new jobs, not a hypothetical model — has dropped radically.
As the chart below shows, the measured portion of CES was near 70% in 2005-06. Now, it has become so increasingly dominated by the hypothesized B/D adjustment, that a mere 20% of the NFP data is truly a measure of the 400,000 participating firms.
>
chart courtesy of Econbrowser
>
In October 2007, the BLS data on job creation has ballooned up to 80% imagined, and a mere 20% measured. That is not a formula for accuracy or precision.
This suggests several important things to us about the BLS NFP data:
• It has moved from a model highly reliant on measurement to a model highly reliant on more modeling;
• The amount of job overstatement has gone from moderate to very strong;
• If our analysis is correct, than Economic growth is much weaker than reported;
This is consistent with what we have seen from various sentiment surveys. More robust job creation would moderate the general malaise that seems show up in the University of Michigan’s Consumer Survey Center and the Conference Board survey of consumer attitudes on present economic conditions.
~~~
I’ll have more later this week on exactly how much I can deduce the BLS B/D has corrupted the data, and what a more accurate amount of job creation actually might be.
>
Sources:
Current Employment Statistics (CES)
U.S. Department of Labor
Bureau of Labor Statistics
http://www.bls.gov/sae/home.htm
Trusting the birth/death model
James Hamilton
Econbrowser, November 16, 2007 08:17 AM http://www.econbrowser.com/archives/2007/11/trusting_the_bi.html
Why Haven’t Home Construction Jobs Disappeared
John M. Berry
Bloomberg, July 9 2007
http://www.bloomberg.com/apps/news?pid=newsarchive&sid=azttcmXd75qk
Hello,
I’m always going into pawn shops to look for old stereo equipment. About 2 weeks ago I was in one of the larger shops when 4 truck loads of illegal immigrants pulled up loaded with construction equipment; table saws; cement mixers, etc, etc.
They came inside and pawned the whole lot… Looked to me like they were headed home….
My wife is also an RN at the local county Health Department. She also tells me that they don’t have as many immigrants coming in for service anymore….
Construction work is BUST in Panama City Florida!
Barry,
True genius in simplifying the complex. You correctly identified inflation ex-inflation, and now have pointed the finger at model-to-model.
In economics, we have inflation ex-inflation.
In finance, we have mark-to-model.
In employment, we have model-to-model.
Somehow this all seems appropriate considering the fantasy world in which present leadership resides.
Then a question:
If the bias of the model were known and believed by the wider market – would the recent bullish activity have been tempered?
Data based on models instead of reality–where have I heard that before? LOL
The first chart shows that we are approaching the point where the jobs report is 100 percent fiction. It is now about 80% fiction, it appears.
Barry,
Intuitively, it seems that the B/D component of the BLS employment statistic is a lagging indicator, perhaps with a very long lag, and by nature, will miss inflection points in small company employment changes.
Further, think about what these small businesses are: mainly entrepreneurial ventures started by a single individual or small group. In my experience, many of these entrepreneurial start ups are formed when someone leaves the corporate world, thinking that they have a reasonable chance of success in building out their business idea. However, with consumer confidence falling, the prevailing view changes to that of the risk becoming too great to leave the comfortable corporate job. Thus, potential entrepreneurs postpone or cancel their plans.
Second, and perhaps more importantly, consider the source of capital funding for new small company formation. With the credit markets freezing up, and lenders unwilling to accept new risk, new ventures face poor prospects for obtaining the necessary start up capital.
Furthermore, some recent start-ups may be throwing in the towel and shutting down their new ventures, particularly those involved in residential and retail development, due to credit issues (foreclosure/banruptcy) and sagging consumer spending. Thus, not only is the B/D adjustment overstating job growth, it may also be missing small company job destruction (meaning a potentially negative B/D component).
Whoever stated this:
“Just when the administration was about to see job creation drop off the measurable map….a new model using guesses and hope was introduced.”
Can’t remember who posted that but it’s a good summary of what the current BLS data is worth.
The second graph tells you all you need to know , even if you can’t read.
Love the brokers jerking each other off this morning…….”all those shares we told you to buy? just hold them now”
Ciao
MS
Welcome to America’s DAILY Family fun Program of:
“Oversubscribe the Repo.”
brought to you by those fabulous people, the banks, who can’t tell the truth and want YOU to help keep all those foreclosed homes off there balance sheet. Come on and “Hope with Us”
http://www.newyorkfed.org/markets/omo/dmm/temp.cfm
Amount Offered: $12.25 billion
Subbed over by: $87.95 billion
WHy in the holy hell is Treasurey doing the “plan”…should’nt it be done through Commerce????
Getting a little hot in Paulsen’s seat methinks.
Just like MLEC it will fail and be seen for what it truly is……yet another vehicle for postponement by and for the banks.
Ciao
MS
Barry, good post and good charts!
I strongly suggest you to submit this article to those nut heads in BLS. Your brief yet convincing post should reach the minds of the model makers in BLS.
This fake BDM thing has been repeated many times in the past few months by many many bloggers, but I am surprised BLS seemed never realize this.
Seriously, I think you should submit this post to BLS.
The CES Birth/Death plug is a problem in general, but is unlikely to have been the source of the ADP jolt for November.
ADP admits it tries to incorporate the effect of the CES plug into its estimate of private hiring (see the FAQ section at the ADP website), so the plug does creep into ADP data through the back door. However, over the past few years, the CES plug for November has been less than 30k. The magnitude of the plug for particular months tends to be fairly stable over time. Since the ADP private hiring tally was 189k, odds are ADP counted (and seasonally adjusted) roughly 160k jobs or more and plugged 30k or less. It is also notable that ADP showed gains in mid-sized and large firm hiring that is larger than the recent norm. “Births” end up in the small firm tally. Small firm hiring has been the strongest part of ADP figures for most of this year, so there is your plug problem, but strong hiring outside the small firm tally strongly suggests the big gain in ADP hiring figures for November is not due to the plug. ADP also showed better than usual hiring for the goods sector, ex-factory – that’s construction and mining. The CES plug for construction hiring in November is generally negative, mining near flat. Again, that suggests the pop in ADP’s hiring tally is not the result of emulating the CES birth/death plug.
Yesterday’s ADP figures may be wrong, whatever wrong might mean, but they are not wrong due to emulating the CES birth/death plug.
Another marked to model scam. It’s becoming a trend isn’t it? But then again why shouldn’t it be this way? The financial people all go to the same schools, take the same classes based on normal distributions(a fallacy but easy to apply if you’re a professor),learn the same Keynesian economics, learn the same philosophy, apply their incorrect reasoning and obtain the same results(simultaneously).
Phil-Las Vegas
The CES plug for construction hiring in November is generally negative, mining near flat
Wrong.
Ditto on the big/medium sized firms as well.
It was all B/D.
What is even more scary, is the fact that
Service Employment represents nearly 84% of Total Employment.
Service Consumption as percent of Personal Consumption has risen in a similar manner in the last 30 years. The attraction of a favorable Service price-inflation has promoted competition and job creation.
But as I posted in “The Service Economy Peril”, the relative price inflation favoring the Service Sector is subsiding. This will lead to job losses in this important secotr.
Cherry,
You have declared me “wrong” but I’m working from data straight off of the BLS tables. Until you can provide something more convincing than mere assertion, I’m gonna have to stick to what can be discovered on the BLS tables.
“In economics, we have inflation ex-inflation.
In finance, we have mark-to-model.
In employment, we have model-to-model.”
Well said. I’d add to this list:
In politics, we have reality ex-facts.
Francois
What is the formula governing how much of the CES data is from the B/D model? If we are at the inflection point of a normal business cycle, would the percentage attributed to the actual sample increase over the next few months? Is the effect of this method of calculating CES that it delays the visibility of recession-like unemployment? Or does it permanently understate unemployment in all phases of the cycle?
BLS Hypothetical Job Growth is 2500% of Total!!
The attention-getting headline for today’s article is true, but misleading. It parallels the monthly assault on the BLS Birth/Death model from Barry Ritholtz and others. In the most recent article, Barry incorrectly writes that the BLS has substituted …
“In politics, we have reality ex-facts”
Stephen Colbert calls this “truthiness”.
Hocus Pocus…
Comments from Macroeconomic Advisers, the fine folks who compute the ADP National Employment Report: interesting discussion, but here are some clarifiers….
(1) Barry is incorrect in saying that we attempt to reproduce the BLS Net Birth/Death modeling in the ADP National Employment Report.
(2) Sean is correct in saying that the acceleration in employment for November shown in the ADP National Employment Report is unrelated to our treatment of births / deaths.
(3) The chart is irrelevant before March of 2006 (that is, the last benchmark), since none of the BLS data before then is based on the BLS modeling of net births and deaths.
(4) Since BLS already has announced the March 2007 benchmark estimate for total employment, we know how much in error the BLS birth-death adjustment was over the period from March 2006 through March 2007. It was slightly high on average.
(5)Since March of 2007, the chart becomes more interesting.
Awesome post Barry (and this comes from a long time reader).
If/when you have time to reflect on this weeks data, could you also talk a bit about how revisions are so dramatically different from the original numbers that get reported?
Thanks.
Brief Notes
It’s been a busy few months. I’ve been dating, working and considering new career options. In between I’ve been writing very little. The Casey Saga came back and then went away again, as did fellow blogger Aspeth. I switched bars…
? Juan or santa Claus?
The birth/death adjustment process to reflect creation and shutdowns of businesses not counted in the survey has contributed substantially to the overall job creation reported. In the twelve months ended November, this procedure accounted for 78% of the jobs created where as it made up only 47% of payroll employment created in the twelve months ended November 2006. The upshot is that hiring is probably noticeably slower than implied by the headline payroll numbers because of this procedure.
More NFP: Worse than Reported
This weekend, we noted that in terms of job creation, the 2002-07 post-recession recovery was the very worst in the post-WWII era. This was in response to the pundit commentary that job losses aren’t nearly as bad today as the 2001 mild consumer-led re…