BLS Birth Death Conundrum ?
July Non Farm Payrolls is out tomorrow, making today a good day to review one of our favorite whipping boys: BLS Birth Death model.
We’ve discussed this ad nauseum, but a brief explanation for the new BP readers: The idea behind the B/D model is to make up for the lag between new firms forming and when they eventually get counted into the Labor statistics. B/D has two major impacts over the course of an economic cycle: It makes the reporting of job creation more accurate in the early recovery phases, but at the expense of overstating employment at the end of the cycle.
I’ve noticed that quite a few defenders of the Birth/Death Model have become emboldened by the market rally. I don’t want to name names, but as a group, the B/D defenders were horrifically wrong across the board about nearly everything — about the housing crisis, the credit collapse, the recession, the market crash, and of course, the massive loss of jobs since hiring peaked late 2007-08.
Their track record precludes taking their rhetoric about B/D very seriously.
For those of you who relish an intellectual challenge, consider the following: Since reaching peak employment in January 2008, small-size businesses have shed nearly 2.5 million jobs. Construction employment dropped for the thirty-first consecutive month in August. The total decline in construction jobs since the peak in January 2007 to 1,562,000. ADP counts 2.5 million small biz job losses since January 2008.
Over the same period, the BLS’s Net Birth/Death Model, which is supposed to account for small biz job losses and gains, shows 1.507 million job gains. The difference is over 4 million jobs — about 3% of the total employed.
A quick look at the historical data tracks the changes to the B/D model, proposed in 2001 ans implemented a few years later (try to guess where):
Historical Net Birth/Death Adjustments *
1999: 30k
2000: 193k
2001: 91k
2002: 470k
2003: 695k
2004: 827k
2005: 817k
2006: 1002k
2007: 1130k
2008: 904k
2009: 555k (January – July)____________
* 1999 to 2002 are April thought March measures; 2004 forward is calendar year; 2003 data is for April-December. Data for 2007-09 is preliminary, subject to future benchmarking
>
Is there a legitimate way to defend this? Or, is the B/D model so wildly wrong as to be indefensible?
>
Sources:
Current Employment Statistics – CES (National)
Historical Net Birth/Death Adjustments
http://www.bls.gov/ces/cesbdhst.htm
CES Net Birth/Death Model
Current Employment Statistics – CES (National)
http://www.bls.gov/web/cesbd.htm
Tweet
Facebook
Reddit
Digg this!
September 3rd, 2009 at 9:16 am
It’s not just B&D — we’re inundated with accounting and metrics trickery, designed and implemented solely for the purpose of obfuscation. That’s why we have calls for things such as mark to market, etc. The charts and numbers the government releases are blue smoke and mirrors. Reality isn’t easily cheated, confused, or manipulated.
September 3rd, 2009 at 9:27 am
It has no value. Knowing the actual number is not as important as knowing the trend, and the BD adjustment obscures the measurement of the trend.
September 3rd, 2009 at 9:32 am
A better list would be the adjustments after the fact when the employment numbers themselves are corrected. That would show if the B/D is a useful aid in estimating or not.
Just because it’s positive every year has no meaning. But if the ex poste adjustments were all wildly positive (had statistical significance due to the B/D model) then it would be a big lie. If not, then the B/D deniers would be wrong.
September 3rd, 2009 at 11:37 am
@Venn
I think it being positive all the time does have meaning when it’s positive very month this year except for january, which makes no sense at all.
September 3rd, 2009 at 11:44 am
Hard to believe that any businesses are actually being born in an environment of declining credit to small biz.
BTW, natural gas may be a rare example of an un-gamed un-stimulated market*. It’s not singing a reflation song.
*These were sometimes referred to as “free markets” in the older literature, B.V.S. (Before the Vampire Squid).
September 3rd, 2009 at 11:45 am
Since the B/D model is supposed to capture the creation of new businesses (mostly small businesses and single proprietorships), it’s strange that it seems to only go in one direction no matter what the economy is doing. Could the B/D model have a bias is only picking up the plus side of the quasi-normal curve which one would expect to see happening in a Brownian motion-like process like the creation and destruction of small, individual elements. Right now it exhibits all of the characteristics of a flow which moves to create a current, as time does physics, and which seems to break all of the rules, as time does in physics.
September 3rd, 2009 at 11:47 am
during construction lay-offs new company creation sky-rockets and competition rises and margins get squeezed, let’s say you are on electrical contractor, a 100k job you make 75k, the guy you layed off will bid it for 80k, cause he works out of his truck and has no overhead, he does work himself and hires a few hands for cash, this will go on for awhile, like any other bizz, most will fail, yet, this phenomena will happen in every building trade………..are the numbers real, imho, you’d need a long time frame to get some good data, most small bizz are gone in 4 years
September 3rd, 2009 at 11:47 am
sorry i meant on 100k job, youre costs are genrally 75k you make 25k,
September 3rd, 2009 at 12:07 pm
What the true numbers are, who know? But the idea that 1.5 million MEANINGFUL small biz jobs have been created during this recession is laughable. A job is only meaningful (and thus worth counting in the job market stats) if it provides a meaningful wage (otherwise it’s nothing more than a hobby or a very poor job). If a person starts a “business” and can only break even (or worse, which is all too often the case), it’s not providing them any personal income and isn’t meaningful in the job market stats. Maybe it will be eventually, but right now it only serves to skew the employment picture.
It’s obviously extremely difficult to try to capture the true and meaningful creation/destruction of very small business jobs. But this model is clearly flawed and needs to be modified. It’s continued existence in this form only serves to harm the credibility of all govt stats.
September 3rd, 2009 at 12:30 pm
Since Barry is on vacation for a few days, here’s a compilation of interesting links to read for the day to make up for no linkfest
http://contraryriches.blogspot.com/2009/09/free-for-all-links-to-interesting.html
September 3rd, 2009 at 12:34 pm
add One Vote for: “the B/D model (is) so wildly wrong as to be indefensible.”
September 3rd, 2009 at 12:41 pm
do they ever really adjust B/D for reality? maybe go back the previous year and redo the numbers based on what really happened. and why do we count only small businesses? since the vast majority don’t employ any one. unless you want to count as small business that have less than a 1000 employees, which sounds more a medium size business to me. not some thing that just pops up and dies quickly as a rule
September 3rd, 2009 at 12:49 pm
As for Barry’s, I suspect, rhetorical question:
“Is there a legitimate way to defend this? Or, is the B/D model so wildly wrong as to be indefensible?”
Of course there is a legitimate way to defend this. The one who defends the B/D-adjustment would have to show that the employment data with the adjustment make the error in the employment statistics smaller compared to the statistics w/o the adjustment, measured against an independent data set about which is known that it has only a small error.
And you, Barry, as an opponent of the B/D model, would have to show that the error in the statistics w/ B/D adjustment is larger than w/o the adjustment. I can’t remember that you have ever delivered on this. I only have seen you claiming that the B/D model distorts the “true” data. That you have done a lot, indeed. Wouldn’t it also be an intellectual challenge to bring some actual evidence relying on data that show that the payroll statistics become worse due to the B/D model, finally?
rc
September 3rd, 2009 at 12:54 pm
@Mark E. Hoffer:
“add One Vote for: “the B/D model (is) so wildly wrong as to be indefensible.””
So, you apparently claim the non-farm payroll statistics would be more exact w/o the B/D model. And how exactly do you know that? Gut feeling?
rc
September 3rd, 2009 at 12:58 pm
@RC
That would be interesting to see but at this point, that the B/D adjustment has estimated that a net 1.5 million jobs have been created through the start of small business in this recession, I find highly improbable in reality.
September 3rd, 2009 at 1:20 pm
I’m in the torrie-amos camp at 11:47 am “during construction … competition rises and margins get squeezed … a 100k job you make 25k” … buy 75k work of stuff which is a write off on taxes (those numbers sounds rosie) to the point of zero profit to pay taxes on
point being – the number that should matter in real time is the deposits to the Kansas City IRS in payroll with-holding … those folks in traditional tax income for the government matter most .. because remember the chart a couple weeks back on who is hiring – the government
small business jobs are break even for the most part – thats why their small .. so who pays for the MIC then .. feeds itself NOT .. a couple days ago a poster here at TBP mentioned (paraphrased) “markets are creating profit” NOT .. making things from base materials creates profit … I think it was in a thread on the market economy and the real world economy
September 3rd, 2009 at 1:24 pm
ps – “75k work of stuff ” made by big corporations who do their best to offset those profits as to not pay taxes to Uncle Sam because thats how we do it
we are to believe that investing in markets will make up for lost wages .. ya right .. NOT
September 3rd, 2009 at 1:26 pm
@Greg
I agree, real time payroll withholding should be what’s important and the number is afwul as documented by TrimTabs.
September 3rd, 2009 at 1:28 pm
“Is there a legitimate way to defend this? Or, is the B/D model so wildly wrong as to be indefensible?”
I think it’s time for B/D to D. Shouldn’t the government collect data and outside sources analyze it? Otherwise it seems like we should call it the BLP, Bureau of Labor Propaganda.
September 3rd, 2009 at 1:31 pm
r_c,
given the ability we, now, possess, in the field of Data Collection/Harvesting, Data Warehousing, and Data Mining, we, really, could Know, easily w/in a %-age point, the true level of Employment of those ‘on Payrolls’ of whatever sort..
September 3rd, 2009 at 1:40 pm
somebody help me understand this…
a long backdated monthly list of (after the fact) adjustments to the monthly employment number will not do anything to verify the veracity of the birth death adjustment factor?
If that is correct, what would….might? The IRS data would miss all the cash economy which is what has collapsed along with traditional industries. And while those who pay taxes helps us all they still contribute to the velocity of money in the system. Will the census next year offer anything to correct the assumptions?
Mark…I agree with your last point in principle but how?
September 3rd, 2009 at 1:43 pm
Cohen,
that would be counter-intuitive, indeed, but this doesn’t prove that the payroll statistics have a larger error w/ B/D model than w/o B/D model. It’s possible that the payroll statistics w/o the B/D adjustment have overstated the job destruction and the B/D model corrects this so that the final statistics w/ the B/D model have a smaller error. How I understand the approach, the B/D model is an additive statistical correction to the numbers from the employment survey. It’s not a model that is supposed to simulate real job creation or destruction.
I have looked through some of Barry’s older postings. Usually his argument goes like this: He shows the numbers or the graphics for the payroll statistics w/o the B/D model. And he shows how the numbers are increased by applying the B/D model. Then he claims the B/D model paints a rosy, distorted picture of the “true” employment situation. The logical fallacy in his argument is that he already assumes what he would have to prove first. He assumes the numbers w/o the B/D model are more accurate than w/ the B/D model.
rc
September 3rd, 2009 at 1:59 pm
DM RTA,
from a different perspective, see: http://epic.org/privacy/carnivore/foia_documents.html
and, similiarly, though more broadly– http://clusty.com/search?input-form=clusty-simple&v%3Asources=webplus&query=PROMIS
as two, of many, examples of the type of ‘webs’ that have been fashioned to collect the vibrations of the ‘Economy’, writ large, and its multi-variate Actors.
remember, again from “Wall Street”: “Information is the most valuable Commodity I know of..”–GG
others have long shared that POV. “No Strings Attached.”
~~
more prosaically, tera-FLOPS and peta-Bytes make for some awesome combinatorial equations that can answer many, seeming, riddles..
September 3rd, 2009 at 2:28 pm
rc
quote from the BLS website: “These models do not attempt to correct for any other potential error sources in the CES estimates such as sampling error or design limitations.”
Also, am I understanding your argument correctly as: maybe the error in the B/D model just happens to be offset by error in other parts of the employment statistics that actually ends up making the overall stats more accurate, but really just by “accident.” Nice model, if so.
quote from the BLS website: “If at some future time, there is a substantial departure from historical patterns of employment changes associated with the residual of net business births and deaths, the model’s contribution to error reduction could erode.”
Gee, you think there’s any difference in employment changes through this recession which is more extreme than any we’ve seen since at least WWII, if not the GD? But instead of making some adjustments to adjust for this, we continue to add absurd numbers of net job creations to the employment numbers. And then some wonder why all the skepticism.
Once again, by including this component that is so clearly flawed (lacking any evidence that it’s accurately portraying the employment situation) during this economic contraction, they reduce the credibility of the entire data set.
September 3rd, 2009 at 2:57 pm
I know this lacks intellectual rigor, but…
What if I start my own business but have no work/income? Should be counted as a new business? I could start five businesses over as many months. Whether or not they succeed they could be counted regardless of their value to the greater economy. I don’t know of any qualifiers or filters, just a simple statistical “guess” at the total number.
I agree, I’m not certain there is much value in knowing how many folks called the phone numbers on posterboard signs stapled to utility poles promsing “CEO level salaries for working from home 4 hours a week”. Those could be “small businesses”.
I can’t even count how many freelance artists (had photoshop on their PC), independent record producers (only self-released their now defunct band’s CD), small recording studio owners (have garage band installed on their macbook, and one condenser mic), show promoters (invited all their myspace friends to a rented VFW hall, no PA or extension cords) and the like I came across when I played in a band. I was never so bold as to call these folks complete frauds, they were trying at least.
For years, one of my relatives was constantly engaged in starting his own business, otherwise he’s what you would call a “marginally attached worker”.
I guess the point is, each was doing *something* similar to “work”. Maybe there were 1.5 million jobs created, but none may not be a potential 40 career ending with a lavish retirement. Most people, I think, envision a job as a (nearly) guaranteed present and future source of income. But that is not a definition of a job, it’s a defintion of success, I may be inclined to interpret B/D as a ‘success adjustment’.
Perhaps B/D is ‘inflationary’ with respect to jobs. 1.5M jobs, but with a value less than prior jobs created.
September 3rd, 2009 at 3:17 pm
Onlooker,
“quote from the BLS website: “These models do not attempt to correct for any other potential error sources in the CES estimates such as sampling error or design limitations.””
I think the emphasis here is on “other potential error sources”. The paragraphs before says for what error sources in the sample survey the B/D model is supposed to account.
“Also, am I understanding your argument correctly as: maybe the error in the B/D model just happens to be offset by error in other parts of the employment statistics that actually ends up making the overall stats more accurate, but really just by “accident.” Nice model, if so.”
That’s a possibility. Another possibility is that the B/D model isn’t so wrong after all and is able to correct a systematic error that comes in with the survey. Or at least that is has before the GD.
“Gee, you think there’s any difference in employment changes through this recession which is more extreme than any we’ve seen since at least WWII, if not the GD? But instead of making some adjustments to adjust for this, we continue to add absurd numbers of net job creations to the employment numbers. And then some wonder why all the skepticism.”
It’s certainly a valid hypothesis that the B/D model has recently increased the error in the statistics instead of decreased it during the GD. But you ask here for making quick adjustments based on a hypothesis only. One would need to have independent data first against which the performance of the model currently applied can be tested. If this is done it probably won’t until after the GD. And one doesn’t even need to assume malice to expect that it will take time. Considering the inertia of a bureaucracy with respect to changes is sufficient.
rc
September 3rd, 2009 at 3:20 pm
RC,
I’m glad you at least admit that the B/D adjustments appear counterintuitive. In theory, you are correct. If there was some problem in the survey that routinely undercounted employment, then one would anticipate some adjustment to correct for that. However, here, the ostensible reason for the B/D adjustment is that the survey will miss both deaths and births of businesses at turning points. If that was the case, we would expect to see death (negative) adjustments at tops and birth (positive) adjustments at bottoms. Has that been the case? Not at all. We have seen a constant stream of births regardless of the business environment. Not only that, but the B/D adjustment was introduced by one of the most politicized/partisan administrations in recent history. I think we can all be forgiven for looking askance at this measure based on what we have seen about it without comparing its results to some unknowable true measure. Do you have any reason to believe that the B/D adjustment reduces the error?!?!?!? I think we have enough reasons to believe that it doesn’t.
September 3rd, 2009 at 3:41 pm
What I understand of the BDA methodology at BLS and Census exemplifies the old adage that you have to ask the right questions to get good answers.
Apparently, they have an awesome database that’s not quite comprehensive but does capture the majority of US businesses. The database is populated with entries based on business UI filing/EIN across the 50 states. I believe they track 6+ million businesses, large and small.
The major problem, as I see it, is that they query the database numbers using a statistical model based on a 5 year MA. So when you’re at the leading edge of falling off a cliff like we did in late 2008 . . . well, most people here know what moving averages do to smooth trend lines.
Another problem is that there is a basis for believing that there has been a rising trend over the past 10 years in businesses failing across the US while new incorporations have remained relatively flat over the same span. I have no idea why there have been more failures Y/Y, but that’s what the data for several large states indicate.
September 3rd, 2009 at 4:10 pm
past discussing the minutiae of a system that is nothing more than a Complex Failure (BLS ‘Methodolgy’), we might do well by viewing this proposition (how to count?) the lens of the title of this weblog: “The Big Picture”.
in addition to my, above posts, see this: http://www.eff.org/deeplinks/2009/09/cybersecurity-act-returns-with-a-fresh-coat-of-paint
“..Like the original bill, the new version appears to give the President carte blanche to decide which networks and systems, private or public, count as “critical infrastructure information systems or networks.” And alongside that authority, there still appears to be murky language that would permit the President to shut down the Internet. Note the troubling provision in the original bill, which said:
The President [...] may order the disconnection of any Federal Government or United States critical infrastructure information systems or networks in the interest of national security;…”
as a negative instance..
or, IOW: “”Find out just what the people will submit to, and you have found out the exact amount of injustice and wrong which will be imposed upon them; and these will continue until they are resisted with either words or blows, or with both. The limits of tyrants are prescribed by the endurance of those whom they oppress.”– Frederick Douglas (1857)
September 3rd, 2009 at 4:50 pm
Good grief.
First of all, it looks like few of you have read the BLS methodology for birth/death. If you had, you would know that the published b/d adjustment that gives everybody heartburn is the second step of the adjustment. The first step is implicit in how they use sample data to estimate employment.
Second. the current monthly estimates are based on a sample of employers plus the b/d adjustment. These preliminary estimates are then benchmarked a year later with data from every single employer in the U.S. that pays unemployment insurance. So we can easily see how well the system works, in terms of sample error and b/d adjustment error.
The last benchmark comparison from March 2008 found that BLS had overestimated employment by 89,000 jobs (see http://www.bls.gov/ces/cesbmart08.pdf). Starting from March 2007, using a sample of employers and the b/d adjustment, after twelve months of estimating, they were off by 0.1 percent. This is a pretty small error. Without the b/d adjustment, they would have been off by about 800,000 jobs. We’ll see soon how well the March 2009 number holds up.
For a great overview and, yes, defense of the b/d adjustment, see http://oldprof.typepad.com/a_dash_of_insight/2009/07/the-bls-responds-to-birthdeath-adjustment-critics.html.
I have analyzed firm-level data for every employer in the state where I work (200,000+ employers). What surprises a lot of people is that even in the worst of times, there are still a lot of new jobs being created through births and expansions. In the fourth quarter of last year, when there were big net job losses, there were still 116,000 jobs created through expansions or births. The losses from contractions and deaths were way larger.
I respect a lot of what Barry writes, but in this case, he doesn’t understand the statistics and he is just plain wrong. And too many of his readers jump on the bandwagon without understanding the numbers.
September 3rd, 2009 at 5:18 pm
[...] called the birth/death model. Barry Ritholtz does steady job of commenting on this data. Here’s his latest. The issue with the B/D model is that it seems to count non-existent jobs and pump-up the overall [...]
September 3rd, 2009 at 6:09 pm
@Sbailey:
I’m going to go way out on a limb and GUARANTEE that the March 09 employment figure was off by a multiple of 80,000.
September 3rd, 2009 at 9:12 pm
Sorry to shout but: THE PUBLISHED B/D DATA IS NOT SEASONALLY ADJUSTED
ok got that out of my system. go to this post by calculated risk which shows NON-SEASONALLY ADJUSTED non-farm payroll data and you will instantly see why this is important. the chart is rather startling:
http://www.calculatedriskblog.com/2009/07/comment-on-seasonal-adjustments.html
September 3rd, 2009 at 11:51 pm
I’m always amazed to see people adamantly arguing without any facts or theory to back them up. Such as Transor Z saying “I’m going to go way out on a limb and GUARANTEE that the March 09 employment figure was off by a multiple of 80,000.”
That’s not science, that’s religion. Is your guarantee worth what AIG’s guarantee to Goldman Sachs was? Do you have any idea what you are talking about?
Read the article SBailey recommended. It explains how the birth/death model actually works. And it explains that the government’s estimates ARE tested each year against unemployment tax records (and it’s been shown time and again that the government’s “imputations” make for better real-time estimates).
And the article explains that even during recessions, there is a tremendous amount of job creation going on. Lots of people start their own business during recessions for obvious reasons: Companies lay off very talented and ambitious people, who realize that they can win any competition against their old boss, because he or she is such an IDIOT!
Now, maybe the birth-death model is crap. But before you say so, you should understand what it is. If you don’t, then please shut up. The world doesn’t need any more random noise; it’s chaotic enough as it is.
September 4th, 2009 at 7:39 am
First, for those of you interested, here is a link to the BLS’ page on the birth/death model:
http://www.bls.gov/search/?cx=011405714443654768953%3Abtgxl8qv780&cof=FORID%3A10%3BNB%3A1&ie=ISO-8859-1&prefix=&q=birth+death&filter=0&sa=Search#967
If that doesn’t work, type in birth/death in the upper right hand corner’s search box.
For those interested, go to http://www.ssrn.com — the social science research library. Type in “birth death”. You’ll get one paper related to economics from the NBER titled “Slow Recoveries.”
This database has over 200,000 papers and over 120,000 authors contributing material. Yet there is no mention of the birth/death model.
In short, the people who should be presenting information challenging the credibility of this model — academics etc.. — have yet to issue a paper outlining their objections in order for their peers to review them.
If anybody knows of such a paper — written by a person with appropriate credentials — please let me know because I’d be happy to consider it.
September 4th, 2009 at 8:21 am
@rexn:
What part of the BDA model using a 5 year MA don’t you understand? I have state data supporting the view that Y/Y growth in corporate dissolutions are significantly outpacing incorporations. The guarantee is data-supported and I stand by it.
In short, the people who should be presenting information challenging the credibility of this model — academics etc.. — have yet to issue a paper outlining their objections in order for their peers to review them.
Indeed.
September 4th, 2009 at 8:37 am
I passionately typed at 1:20pm & copy pasted at 1:24pm .. “work of stuff” should have been “worth of stuff” on this here/gone/here again thread
in short the point is employment is a newspaper number for election time .. Kansas City IRS income #s matter to the USA balance sheet to pay its bills .. the number of people working (or not) – taking care of themselves in free market capitalism but not paying taxes into the higher causes of government – well it just doesn’t matter . really … (except for the burden expected on those higher cause services) .. the payroll deduction deposits and the allowable deductions for business are the #s to focus on .. simply the USA account balance ….. and focus on the power of business over country
September 4th, 2009 at 9:06 am
and before I see “free market capitalism” thats what I’m talkin about .. BooYa
here in our system there is corporate welfare as much if not more than social welfare .. and we call that side of the coin R&D for job creation … oh . and security for the world
September 4th, 2009 at 2:31 pm
The benchmarking process involves retroactive changes going back 5 years.
http://www.bls.gov/ces/cesregrevtec.htm
Revised business birth/death estimates also are incorporated into the post benchmark months. Additionally, seasonal adjustment models are rerun, and seasonally adjusted estimates are replaced for 5 years back.