Friday, April 3, 2009

Today's Real Payroll Number: 750,000, Total Unemployment Rate at 19.8%

John Williams, who runs the great Shadow Government Statistics site, presents the true numbers behind today's "seriously flawed BLS payroll reporting." The upside bias coming out of the BLS is almost scandalous: one wonders how much "push from above" there is to concoct these numbers.

From John's distribution earlier today:
BLS Jobs Reporting Is Seriously Flawed, at Best. This morning’s (April 3rd) reported March jobs loss of 663,000 again was close to consensus expectations, but, as has been common in recent releases, major downward revisions to prior reporting helped to mute the current headline jobs loss significantly. In each of the six most recent monthly payroll reports, the prior month’s payroll level was revised lower. For October 2008 to March 2009 reporting, the downward revisions to the prior month’s seasonally-adjusted payroll level were respectively: 179,000, 199,000, 154,000, 311,000 (still significant net of benchmark revisions), 161,000 and 86,000. Five of the six revisions exceeded the Bureau of Labor Statistics’ (BLS) 95% confidence interval of +/- 129,000 jobs for monthly change.

Net of revisions, the March jobs loss would have been 749,000. Net of the Concurrent Seasonal Factor Bias (CSFB), which reflects the reporting problems, the loss would have been 750,000, in line with my estimate in the March 29th Flash Update.

Payroll Survey. The BLS reported a statistically-significant, seasonally-adjusted jobs loss of 663,000 (down 749,000 net of revisions) +/- 129,000 (95% confidence interval) for March 2009, following an unrevised 651,000 jobs loss in February, but January’s jobs loss was revised from 655,000 to 741,000. Annual contraction (unadjusted) in total nonfarm payrolls continued to deepen, down 3.56% in March, versus a revised 3.10% (was 3.12%) in February. The annual decline in March was the deepest since July 1958. The seasonally-adjusted series also continued contracting year-to-year, down by 3.48% in March versus a revised 3.08% (was 3.02%) contraction in February.
During the Clinton Administration, "discouraged workers" — those who had given
up looking for a job because there were no jobs to be had — were redefined so as
to be counted only if they had been "discouraged" for less than a year. This time qualification defined away the bulk of the discouraged workers. Adding them back into the total unemployed, unemployment in line with common experience, as estimated by the SGS-Alternate Unemployment Measure, rose to about 19.8% in March, from 19.1% in February.
ZH is now taking bets what the backward revisions will be for February and March, when April and May numbers are announced (assuming the U.S. hasn't defaulted by then).
Sphere: Related Content
Print this post


Broken said...

Employment is now lower than it was in Clinton's last year.

Broken said...

My own quick and dirty calculation for real employment is to use the BLS Historical B tables - not seasonally adjusted - and subtract 100,000 per month past January for the birth/death model. Subtract an additional 70-100K from the current month for revisions.

Birth/Death model works OK during a recovery.

steve said...

I'm sorry to veer off-topic here, but this is an administrative issue and I didn't see any more suitable post to append this comment to. I'm presuming to speak here on behalf of your subscribers who are high-volume blog readers.

I love and immediately began to rely on your blog upon discovery. Then, a few days ago, tragedy struck...It is a huge setback to have to click through from my RSS reader to see each entire post. Since my subscriptions push 800-1200 post a day to me from around the web, by sheer necessity I end up triaging which one's to read in full *now* vs. *later* partly on the basis of the extra click and load time to get to the site. Experience tells me that a meaningfully high percentage of the partial-feed blog posts I save for *later* often end up expiring from my reader unread. I was particularly disappointed when I saw that your writing was going to end up being subjected to that blunt filter. The backlog from just these past few days confirms my initial expectations.

If the change in your feed settings from full to partial text was inadvertent, please please change the setting back. If deliberate, please at least consider making your reasoning public (did I miss that post), so that those of us who miss the full text feed can think about and suggest other ways for you to meet your objectives.

What the hell, I'll even try to anticipate a couple:

If you are trying to force click-throughs as prep toward monetizing your site with advertising, for instance, I think your RSS subscribers would certainly be willing to live with in-feed ads.

If your goal is usage stats and you are not seeing the number of RSS subscribers you have, there are ways to do so which the community can get back to you about.

The key take-away though (assuming you are still reading this) is that most of the bloggers with big followings send a full feed unless they are anchored to dead media (e.g. NYT), and not always even then...Ridholtz, Salmon, Yves, CR, Hempton (ok, not so big yet). Really, Roubini's RGE is the only major exception, and that backfires for them, as I ended up just subscribing directly to the bloggers who cross-post on RGE, so RGE gets no click-throughs anyway on those posts -- worst of all I often end up reading what the other bloggers have to say about Roubini's posts before or even instead of reading him directly. I can't be the only one managing my info-firehose this way.

Okay, I'm done obsessing self-pityingly about this for now. Just please put it back.

ts said...

I like John Williams work, but I get really annoyed when he trashes the B/D models, and not just because I helped develop them.

The proof in the pudding is how far off BLS estimates are when the data is benchmarked in January. If B/D was a big issue, there'd be much larger revisions to the data as the estimates are replaced with UI rolls.

The revisions since B/D was implemented have been -0.1, -0.2, -0.2, 0.2, -0.1, 0.6, 0.2 and -0.1 (percent). This includes three turning points in the business cycle where one would expect B/D estimates to have the biggest issues. This is a much better performance than in the years prior to implementation of B/D.

I do hate the concurrent seasonal adjustment (readjusting each month). This adds unnecessary volatility to the data (particularly the revisions) and doesn't improve the results.

Incidentally, if BLS was able to revise the B/D numbers more frequently (operationally, it is impossible to do), it would only add to the problems, as there would be an B/D revision added to the seasonal factor revision and sample estimate revision.

Anyone with a political axe to grind always hops on any change in the number, and this would just provide more ammo.

Anonymous said...

There will have to be a downgrade of USA's debt. It's amazing it hasn't happened yet. I think Moody's, s&p and Fitch are all bribed. Bribed or just plain scared as to what will happen if they do. Don't think US will default though. But downgrade that country NOW please. You know it makes sense