Wall Street took this as a weak jobs report, and it may be worse than people think.
The Bureau of Labor Statistics reported today that U.S. nonfarm payrolls, as measured by their survey of establishments, increased by a seasonally adjusted 92,000 workers in July, less than many analysts had been anticipating. I have long urged looking at these numbers together with other employment indicators, which at the moment give a considerably more pessimistic reading.
One good alternative source is Automatic Data Processing, Inc., who use the data for the 23 million Americans whose payrolls they process to come up with an alternative estimate of the number that the BLS is trying to infer. ADP’s estimate, reported on Wednesday, was that the economy added only 48,000 private sector jobs. The discrepancy between this and the BLS number is even bigger than it might appear, because ADP does not include government employees. According to BLS, the public sector shed 28,000 jobs in July, so to convert the ADP number into units directly comparable to the BLS number, one would subtract 28,000 from ADP’s 48,000, arriving at a mere 20,000 increase in total July employment implied by the ADP private-sector estimates together with the BLS government-worker estimates.
A third source to which I pay attention is the separate BLS household survey, according to which U.S. employment fell by a seasonally adjusted 30,000 workers in July. One benefit of watching those BLS household numbers is that they’re not subject to the birth-death adjustments that can make it difficult to recognize turning points with real-time establishment data.
I’ve been advocating using a weighted average of the three estimates (BLS establishment, ADP, and BLS household) to get the best impression of what is going on, recently recommending weights of 0.7, 0.1, and 0.2, respectively. That calculation comes to
(0.7)(92) + (0.1)(20) – (0.2)(30) = 60
or a net total employment gain of only 60,000 jobs for July.
Some would argue, and quite legitimately, that we should not put too much stock in one month’s numbers, and the poor July numbers look better when you average them in with the strength reported for June. Even so, things at the moment don’t look so great to me.
Technorati Tags: BLS,
employment
Jim, I agree. There is some slowing in the July data, but probably the story is worse than the headline numbers suggest.
Question: Why aren’t the employment numbers plotted with their 95% confidence intervals or some other measure of the uncertainty? If you plot the series that way, it naturally falls out that the newest monthly data point is the least certain and should be weighted accordingly. Then do a weighted moving average to look at the trend over time.
Is this just a remnant of the press and policy makers only liking to look at a single number? My impression is the parsing of the monthly employment numbers always seems like a lot of overinterpretation of a single number that is relatively uncertain.
That’s a valid point, Anonymous, though it’s not clear exactly how to construct your 95% confidence intervals. One could base them on the difference between the initial and final reported value, though that would amount to treating the final estimate as if it is the “truth”, which it in fact is far from being.
Well, certainly construction and residence-related finance are two big job engines in this country and both are being hard hit right now. Tens of thousands lost their jobs in Q2 in homebuilding and now we are seeing tens of thousands more losing their jobs as some of the home lenders shut down. I would anticipate continued erosion over the next few months. The loan businesses are being run for cash and not for growth or income right now.
If the ADP numbers are only private sector employees then to make them strictly comparable to the BLS payroll numbers, wouldn’t you compare the ADP numbers to the change in private payroll employment from BLS? In that case you’d want to adjust the payroll numbers up by 28,000. Unless I’m missing something….
Professor,
Thanks for your analysis. I tend to agree but I do not believe it is as bad as some thought it might be.
Could this give Bernanke ammunition to lower rates? Sadly, I think it just holds him back from raising rates. The FED probably will not make any change though this is a good time to make the move a slower cyclical period entering the fall, slow employment, low housing numbers. Sadly, he will probably weigh the weak dollar and take action that will weaken it even more.
I know there is some uncertainty around the BLS numbers and their legitimacy but what about the unemployment numbers? They would seem to be much better documented and reported. One would think that we should have clearly seen some incremental increase in both weekly numbers and the 4 week average. So far not so much.
Awoolf, one can either (a) compare the ADP number as reported (48,000) with the BLS for private-sector employment (92,000 + 28,000 = 120,000), in which case the difference between the two is 72,000; or (b) compare the BLS number as reported (92,000) with the implied ADP for all employment (48,000 – 28,000 = 20,000), in which case the difference between the two is 72,000. Either way you do it, the difference between the two is exactly the same number, and it is the difference between the two that I am highlighting. I do it here in terms of option (b) rather than option (a) because that is how we’ve always done it here at Econbrowser, and because people are more used to thinking in terms of the number as reported by BLS rather than the number as reported by ADP.
The only way unemployment was kept this low was by morphing the economy into a giant residential development, construction and debt pumping machine…finally supported by lending $500000.+ blocks of money to those who recently swam the Rio Grande or “overstayed” their visas. It should not come as a shock to anyone’s sensibilities that the “ruling elite” became desperate for a “grand compromise” on immigration…a deal that would deliver lots of warm bodies to borrow lots of hot dollars.
If ever a nation was in need of a “cleansing” and a return to a highest and best allocation of capital it is the USA today. It will require a higher level of unemployment and a lot of pain, particularly for those who hoped to “borrow themselves rich.”
Looks like the markets are wary of someone holding a “warehouse” full of toilet paper loans blowing up over the weekend.
There will be several as we move ahead with the cleansing.
And that is a good not a bad thing.
Maybe,
it might be a prospective healthy situation,
if the pace of increase of the labor market NOW is slowing gradually a bit,
just to balance the hourly productivity of american workers.
Just to decrease some upside risks in 2H core PCE deflator.
In my opinion, the 2 HALF risk still remains into core consumer inflation dynamics.
Will it really moderate???
‘Cleansing?’ Why stop there? Doom, doom, the fire, purify. They shall be punished…
One thought:
Change in employment divided by work force or 92K/145 mil = very small number. But full of deep significance no doubt. Just ask an economist.
A Rat in the cage…
As Robert Auuman “proved” long ago.. in a Bayesian framework.. rational people can’t disagree for long …
Step back a minute…read your forecast.. and then look at today’s changes in the probabilities of the FED easing.. the market is actually pricing in close to 50% probability that the FED eases as early as September.. we’re back essentially to where we were
months ago.. before the weather clouded our vision so to speak..
Here’s the question I pose to you…if the market is clearly pricing all the scenarios’ that you consider probable.. already..
By the way the market is actually pricing and trading options that are bets that that the FED will actually EASE twice by December..
these are trading at approximately 35% “probability”.. i.e pay 350 dollars and get back 1000 if the FED eases twice by December….
I like to make a couple of points:
If you look at ADP numbers since its first release June last year, it probably has the worst track record as compared to all those economists included in survey. Not sure why people are still paying that much attention on it.
A much better payroll predictor may be claims. Most of the times payrolls high when claims lower. One example is this May, when the survey week claims number was below 300k, the payrolls first came out 157k and then revised up to 190k. This month claims also strong, which should have consist with above consensus number. The surprise was the -28k goverment loss, particularly the teacher job loss probably due to bad seasonality.
Professor,
Without getting involved in the birth/death debate I find it useful to analyze Table 3 published by BLS.
You can compare the latest data with previous 3/6/12 month averages. The data gets smoothed out and trend spotting is probably more meaningful.
What Table 3 tells me is that overall picture is OK especially in private service sector.
1. Numbers were 132K/145K/134K/157K for 1/3/6/12 month averages. One can argue that last 6 months are softer that the 6 months prior.
2. The type of these jobs (from earnings point of view) is another matter.
a] Education/Health was the largest category at 39K/56k/49K/47K or roughly 30% of total private service category.
b] Leisure & Hospitality was second at 22K/40K/33K/40K or roughly 25%.
c] Professional & Technioal Services are third at 26K/30K/24K/24K or roughly 15%. These are the GOOD jobs and growth on the admin services was actually negative for most of the periods.
d] Finally Trade/Transport/Utilities were at 19K/11K/21K/23K also roughly 15%. The Durable Goods sector has been consistently solid (driven by exports I presume). Retail trade has deteriorated significantly (correlated to softening consumer spending).
Some surprises (of course subject to revisions). Local gov.went negative at -16K (they were consistently at +20K). Another was Real Estate and Credit intermediation were strong. Perhaps this will change by next month.
Bottom line is that hiring is still happening (this recovery has always been tepid) but for low wages jobs.
regards,
While I have done some work on all of these topics, I would like to focus on the confidence interval.
The results are adjusted as more companies in the sampling frame report. Even after all have reported, the 90% confidence interval is +/- 99K jobs. This is not bad when trying to estimate all of the jobs in the economy, but not so good when trying to infer monthly changes.
The focus on the birth/death adjustment is, I think mistaken. There are millions of “imputed jobs” every month, since the BLS has learned that some companies will never respond to the survey.
The birth/death adjustment is a minor part of this imputation, but one that has improved results.
Too much weight is placed upon monthly results, especially the first report. The non-sampling error might also be large, and is more difficult to estimate.
My understanding of the ADP numbers is that since it is pvt sector only, one should normally add about 25k to the number to get the comparable BLS number. Since the BLS came out showing a decline in govt jobs, incidentally almost all at the local education level, on should just compare the 48k ADP number with the BLS pvt sector number of 120K. However, since the govt decline was almost all in teachers and we are in the middle of the summer, it sure raises the possibility of a funky seasonal adjustment factor to me.
Yes if the focus is to forecast BLS release then ADP is useless. This month it gave a correct direction for some companies to lower their forecasts, but for a wrong reason.
Jeff,
I’m not even an amateur economist, so forgive my ignorance of the key terms. The final estimate is reported how long after the initial estimate? Do you have an idea of the magnitude 90% confidence intervals for the initial estimate?
I believe I read somewhere that the initial estimates have uncertainties on the order of 200k jobs. If that is the case, an uncertainty weighted analysis would 1/4 the weight on the newer numbers relative to the older numbers.
Are these uncertainty estimates publicly available anywhere? I do like plotting data, and would like to see how the jobs report look with their uncertainty intervals added.
You just reiterate a fundamental empirical fact – in the long run low growth rate of labor force defines low unemployment and low inflation
http://inflationusa.blogspot.com/2007/07/what-is-driving-force-of-inflation-and.html
Fluctuations around these long term equilibrium relationships are completely related to measurement errors:
http://inflationusa.blogspot.com/2007/07/how-accurate-are-labor-force.html