Subscribe By RSS
The US Employment Report And The Birth/Death Adjustment
By Jeffrey Miller on August 7, 2009 | More Posts By Jeffrey Miller | Author's Website
Nearly everyone interested in the markets pays special attention to the monthly employment situation report. Briefing.com gives it the highest rating for importance. CNBC commentators todaysaid that everyone was breathless, waiting for the announcement. (This article is Part Three in our series on the Birth/Death Adjustment.)
In a way it makes sense. Employment is a crucial measure of the economy. Severe job losses have piled up month after month in the current recession. Everyone cares. Everyone is interested.
It is also seen as a recent data point, partly because it is announced at the beginning of the month.
In a way it does not make sense. There is a wide “confidence interval” of over 100K jobs for sampling error alone. This does not consider the revisions. The sample is only partly complete with the “early returns” for the first announcement. That is the one that gets the market attention. The two monthly revisions include information from late-reporting companies and also “concurrent seasonal adjustments.”
We have not even started with issues other than sampling error.
The Bureau of Labor Statistics (BLS) has created an excellent statistical process for counting every job in one month, counting them all in the next month, and determining the difference. Since there are over 130 million jobs, the error in the result is remarkably small, an envious accomplishment for anyone who understands statistics.
A Universal Error — Impugning the Birth/Death Adjustment
Since the report is so important and the BLS is so open about methods, one would expect everyone to understand the process. Unfortunately, that is not so.
There is a thriving market for pundits who criticize the methodology, take statements out of context, and confuse seasonally adjusted data with non-adjusted data. Since no average reader can be expected to understand the methodology, these criticisms, usually from popular blogs and pundits who are not economists or statisticians, have gained general acceptance among market participants.
The overall result is to discredit the work of the BLS. This criticism has been proven wrong. (See here.) The actual result is known about nine months later when we have information from state employment records. We all wish we had better data sooner, but for now, the scorecard comes later.
In Part One of our series we showed that the estimation of job creation was not just the Birth/Death adjustment. In fact, that is only about 10 or 15 percent of the estimation of new businesses. Most of it comes from the imputation of business births from business deaths. This process, the imputation step, is “concealed” within the sample, so to speak. It captures the cyclical element of the economy very effectively.
In Part Two of our series we noted that most observers seem unaware that there is always significant new job creation, even in recessions. This job creation is on the order of 100K jobs per business day.
Why the Mistake?
Why are so many so wrong in interpreting the data? The answer is easy.
The critics ignore the fact that the estimation of job creation is a process with multiple steps. They ignore the imputation step that captures cyclicality, focusing only on the Birth/Death adjustment, which is merely a residual factor.
The residual factor is a rather stable component. It is a residual, something left over after the cyclical element has been recognized. It is small relative to overall job creation.
In making this mistake, the critics look only at the Birth/Death adjustment and expect it to reflect the economic cycle. Here is a typical example. Our quoted source is seen as a leading authority on the employment report, and most of the information he provides is very accurate and helpful.
A major modification to the NFP measure is the Birth Death adjustment. Changes to the BD were proposed in 2001, and implemented a few years later. This attempts to capture early improvements in employment at the start of a recovery was the goal. The trade off is it wildly overstates strength at the end of a cycle. For example, in 2007, approximately 75% of reported new jobs were due to this adjusatment.
Please note the argument. The writer mistakenly expects the residual to respond to the economy. It does not. It is rather stable. Let us look at the data.
The blue bars represent the actual residual, a series that remains fairly stable during the business cycle through the last recession. The critics complain that the residual becomes a larger portion of net job change. In fact, the process is accurately capturing the residual factor. It should become larger as a percentage of net job changes, and it does.
Briefly put, the critics make the mistake of ignoring the imputation step, the most important part of the process, and focusing on the residual adjustment.
Why This Matters
Here at “A Dash” we have had some very bearish forecasts on employment, including Friday’s report. We have no reason to make things seem better than they are. The employment situation remains very negative.
As always, our motives are to make the best use of available data. The multi-year criticism of the BLS and the methodology used has been shown to be incorrect. Despite this, the myth prevails. This is due partly to the lack of any mainstream media coverage of the BLS viewpoint.
The mistake will become more important when an economic recovery begins. Those who blindly ignore the actual process of estimating job creation will be left behind.
For some months we have sent private emails to professional economists who misinterpret the data. Most of these sources have backed off after giving further study. In future articles we will be more explicit in highlighting those who do not understand the results.
Subscribe by RSS 
Subscribe by Email
Email This Post To A Friend
3 Reasons Why ETNs Are Shrinking - AIG And Its Counterparties
- Positioning For Year-End
Why Pimco’s Fleeing From Mortgage Debt Into Government Debt - US Bonds Are Blasting A Warning
- *Sweden Nov. Manufacturing Confidence Indicator Rises To Minus 10 From Minus 13 In October - 5 mins ago
- Japanese Stock Market Ends Higher On Bargain Hunting, Dip-buying - 6 mins ago
- *Sweden Nov. Economic Sentiment Indicator Rises To 98.8 From 94.9 In October - 7 mins ago
- *Sweden Nov. Consumer Confidence Rises To 11.4 From 7.5 In October - 8 mins ago
- *Czech Nov. Consumer Confidence Rises To 94.1 From 87.7 In October - 20 mins ago
I’ve been trying to wrap my brain around the B/D adjustment for some time now. The whole translation from non-seasonal to seasonal, the monthly divergence for the series, the flying-in-the-face-of-logic “growth” of small businesses in 2009 (leisure and hospitality has been a large driver this year), etc.
While I think I understand your arguments above and those on your website, I believe that the B/D adjustment will prove to be misleading this time around - primarily for the reason that this is a STRUCTURAL recession (IMHO) and not a CYCLICAL recession like those in the past. While using past data series to attempt to construe relationships make sense, if these relationships change or break down, then the data series will not be as predictive as before.
To make my point, the American Bankruptcy Institute has reported that 2009 1H business Chapter 7 filings (business deaths) have jumped 57 percent from 2008 1H to 20,375. Chapter 11 filings are up 113 percent to 7,396.
Conversely, the CEO of Intuit has been reported as commenting that business formations (business births) are running at 50% of normal levels (http://blogs.barrons.com/techtraderdaily/2009/08/20/intuit-ceo-learning-to-dance-in-the-rain/). Admittedly, this is an anecdotal data point, but Intuit does have the leading small business accounting package, and therefore some thumb on the pulse of this market.
So, business deaths up, births down. This seems to jibe with the credit contraction, deleveraging by banks and consumers, and soft loan demand and lending.
Yet for 2009 the B/D monthly adjustment has been running HIGHER in 2009 than 2008 - about 150,000 cumulative jobs from April - August 2008 v. April - August 2009.
More jobs created in 2009 v. 2008 even though other data would indicate the opposite?
Also, one more thing - the January 2009 B/D adjustment was a -356, meaning the B/D adjustment reduced jobs for the month. The rest of the monthly changes have been positive. Query - how much of the second derivative trade in the trend of payroll numbers is due to the quoting of this -700’s print in January largely due to the negative B/D number? People keep talking about how the jobs number has improved from the -700’s, but this number seems to be taken dramatically out of context and perhaps the job market really deteriorated the most in the February - March period and the improvement since has not been as dramatic (the B/D adjustment peaked in April - June).
Just some thoughts…
Trader Joe — The key — and very important — is not to look first to the B/D adjustment. Most of the job creation is measured through the imputation step, which I have carefully described in this series.
If you look only at the B/D adjustment, which represents a residual, your intuition will lead you astray.
You are not alone. That is what nearly everyone does.
I have a lonely task in trying to explain the method. Please take another look at the complete series.
And thanks for commenting — a viewpoint that many share.
Jeff
Jeff - thanks for responding. I think I understand your point on the imputation step, but I’m going to stick with my original assertion (that the B/D model is overstating business creation and under-estimating business deaths in this credit-driven, structural recession).
I spent a bit more time on the BLS website. The imputation stage you mention: “Earlier research indicated that while both the business birth and death portions of total employment are generally significant, the net contribution is relatively small and stable. To account for this net birth/death portion of total employment, BLS uses an estimation procedure with two components: the first component excludes employment losses from business deaths from sample-based estimation in order to offset the missing employment gains from business births. This is incorporated into the sample-based estimate procedure by simply not reflecting sample units going out of business, but imputing to them the same trend as the other firms in the sample. This step accounts for most of the net birth/death employment.” BLS website.
So, from earlier research (when? was it a time of contraction, expansion, etc.?) it is assumed births = deaths and that while the gross numbers are high, the net numbers are “small and stable”. Therefore, it is assumed births and deaths net each other out, and the current trend is applied to the deaths instead of shutting them down.
Ok, but, what if the net between births and deaths is not small, and not stable in unusual periods? What if you have a massive credit shock, and a dramatic reduction in GDP, closing lots of small and medium-sized businesses (increasing deaths), and the lack of credit prevents many small and medium-sized business from starting (declining births). [The AT&T and Verizon CEO's were quoted today as discussing the far lower numbers of new businesses.]
The analogy would be using historical correlation figures for asset returns to build your portfolio allocation model, only to find that correlations all increase in a de-leveraging, panic driven sell-off when every asset class goes down at once. The data series works in calm and relatively stable times, but not in chaotic periods.
For the second component: “The second component is an ARIMA time series model designed to estimate the residual net birth/death employment not accounted for by the imputation. The historical time series used to create and test the ARIMA model was derived from the UI universe micro level database, and reflects the actual residual net of births and deaths over the past five years.”
So, they take the experience of the past five years and use it to estimate how many jobs were created by new businesses in a current period. The past five years were marked by easy credit, good GDP growth, and coming off a “jobless recovery” where some entrepreneurial activity was driven by corporate employees taking buy-outs to start their own businesses (this last point is speculation and a bit weaker to make).
Again, fitting the past five year’s data to the 2008/2009 Great Recession, I believe, will lead to an over-estimation of the jobs created by business births, and an under-estimation of the jobs lost through business deaths.
Finally, even though the numbers listed in the B/D tables are the ‘residuals’ - they still show the bias of the data - for positive job creation through business births v. business deaths.
I believe - and I don’t know to what order of magnitude - that the data series will prove to be wrong during this period of more-rapid business destruction and less-frequent business creation.
Thanks again for the discussion and your original post(s)…
Joe — The fact that we have actual data on job creation (with a lag of about nine months) is fortunate. We do not need to speculate. We just look at the data from state employment reports.
So far, these reports have proven the BLS critics to be wrong — year after year. They continue because they claim that this period is different. You seem to agree, but you are looking at data with an open mind.
I suggest this. We now have actual results for the worst period in history — right in the heart of your credit crisis thesis. Please look at these actual data, and then tell me what you think.
http://oldprof.typepad.com/a_dash_of_insight/2009/08/the-most-important-data-you-missed-last-week.html
Thanks,
Jeff
Today’s BLS release showed a benchmark revision of -824,000, with -855,000 for the private sector.
From the BLS Commissioner’s statement:
“In keeping with standard practice, we are announcing the
preliminary estimate for the next benchmark revision to payroll employment. The benchmark process annually revises the payroll survey’s sample-based employment estimates to incorporate universe employment counts derived primarily from unemployment insurance tax reports.
Preliminary tabulations indicate that the estimate of total nonfarm payroll employment for March 2009 will require a downward revision of approximately 824,000, or six-tenths of one percent. The historical average for the benchmark revision over the prior 10 years has been plus or minus two-tenths of one percent. Most of the additional job loss occurred in the first quarter of 2009,
when payroll employment was declining most steeply, and appears to be due in part to an increase in the number of business closings.”
Your primary argument seems to be that the B/D model improves the preliminary estimates. Fair enough.
But, I think the benchmark revisions and the Commissioner’s comments also demonstrate my point: using past results to model and impute current estimates can be problematic if the nature of the situation changes. More businesses are closing in a deleveraging, recessionary environment.
http://www.bloomberg.com/apps/news?pid=20601068&sid=aXoQJ14iSlWg
traderjoe — Thanks for the pointers. I’ll do another piece on this, of course. As I have often noted, the BLS checks their results against actual data, but there is a lag of about nine months. If there was going to be a break down in a process that has worked for many years, I would have expected it to show up in Q408.
More later.
Thanks,
Jeff