Redirecting

Friday, February 05, 2010

Unemployment: Birth/Death Adjustments Overstated Jobs by 902,000

Leading up to the release of today's BLS Employment report, there was a lot of talk about the birth/death adjustment.  The b/d model can be a bit tricky to explain - I think Barry Ritholtz's old piece here does a good job of summarizing what happens.  Basically, the BLS estimates job creation or destruction based on the filings of new company incorporations - because companies which are too new or too small don't get counted in the CES (current employment statistics) survey. In Ritholtz's words:

"Previously, BLS tended to under report new jobs in the beginning of a a cycle turn. What the new B/D Adjustment series did was take new incorporation filings per state, and deduce from them that new jobs were being created. (That took effect around 2003).  This improved somewhat the ability to capture new jobs at the start of the cycle. But the flaw in the adjustment was that the model radically overstated job creation at the end of the cycle. Say a firm goes out of business, or lays off 100s of workers. They form new shops, incorporating these start ups.  According to the BLS, that is job creation.  But in reality, a steady paycheck with benefits has now been transformed into a start up with none of the above. And as we know, 90% of all new businesses eventually fail.  How misleading is the BD adjustment at the end of the cycle? Consider that in 2007, 75% of the BLS newly created jobs were due to the B/D adjustment. That did a nice job masking the actual problems beneath the surface."

Today, when the number came out, the results were even worse than most expected.  From the BLS report:

"The total nonfarm employment level for March 2009 was revised down-ward by 902,000 (930,000 on a seasonally adjusted basis), or 0.7 percent. The previously published level for December 2009 was revised downward 1,390,000 (1,363,000 on a seasonally adjusted basis)."

In plain English, all the employment data that was reported up thru March 2009 understated the number of jobs lost by 902k.    How big a variance is this?  Is it normal?  The CES tells us:

"The March 2009 total nonfarm payroll employment estimate was revised downward by 902,000 or -0.7 percent. Over the past decade, absolute benchmark revisions have averaged 0.3 percent, with a range from 0.1 percent to 0.7 percent.  Benchmark revisions are a standard part of the payroll survey estimation process. The benchmark adjustment represents a once-a-year re-anchoring, based on March data, of sample-based employment estimates to full population counts available through UI tax records filed by nearly all employers with State Employment Security agencies."

In other words, this revision was the largest (at -.7%) since the modern adjustments began 10 years ago.  The interactive Bloomberg graphic I linked to above also highlights this dispersion. 

-KD

5 comments:

Transor Z said...

Hey KD,

I was screwing around on Barry's site today: http://www.ritholtz.com/blog/2010/02/dissecting-the-nonfarm-payroll-data/#comment-253133 and doodling about calculating U3 to factor in declining participation rate. This is not peer-review quality stuff but I wanted to get your take.

Here's the graph of CLF:
http://data.bls.gov/PDQ/servlet/SurveyOutputServlet?request_action=wh&graph_name=LN_cpsbref1

The 1/10 datapoint represents 64.7% participation and the series until 1/09 is a solid ~66%.

If you treat 155 mil as base CLF (instead of the current 153 mil) and just add the missing 2 million to the unemployed number of 14.8 mil you get 16.8 mil unemployed and a U3 rate of 10.8%. Maybe totally coincidentally this is close to the 10.6% NSA.

Imagine if the current methodology had been in place in 1930. What would CLF have been in 1935 after 5 years of decline?

Just a thought.

-TZ

Transor Z said...

http://www.scribd.com/full/26493859?access_key=key-1xt0aottpr19lwva172k

-TZ

Kid Dynamite said...

TZ - did you write that piece? nice.

it's clear to me that SUBJECTIVELY, right now, we have a major major problem, and those who look at statistical recoveries based on a handful of past data points are making massive mistakes.

i think you'll enjoy the piece I have coming Sunday evening.

Transor Z said...

KD, yes I did. Thanks, glad you liked it. If you have any interest in working with the idea and improving on what I did I wouldn't be insulted at all -- in fact I'd like it. Obviously I'm not a statistician but I think there's something important there.

I'll be sure to check out your piece after the Saints win tomorrow.
;-)

Later.

rjs said...

FYI, guys, i linked to this post in my unemployment post this morning, and then pointed out the link to Transor Z's paper in the comments here with my comment below that post...