The Atlanta Fed's macroblog provides commentary and analysis on economic topics including monetary policy, macroeconomic developments, inflation, labor economics, and financial issues.

Authors for macroblog are Dave Altig, John Robertson, and other Atlanta Fed economists and researchers.

« How close to deflation are we? Perhaps just a little closer than you thought | Main | Gauging the inflation expectations of business »

July 16, 2010

A curious unemployment picture gets more curious

UPDATE: One of our eagle-eyed macroblog readers thought something was fishy-looking in the second chart of yesterday's (July 15) post. He was right—the chart was in error. This post is an updated, edited version with the erroneous chart replaced. There have also been some text revisions to better reflect the revised chart. The new text is bolded in this post.

At first blush, the second quarter statistics from the Job Openings and Labor Turnover Survey (commonly referred to as JOLTS and released Tuesday by the U.S. Bureau of Labor Statistics) suggest little has changed recently in U.S. labor markets:

"There were 3.2 million job openings on the last business day of May 2010, the U.S. Bureau of Labor Statistics reported today. The job openings rate was little changed over the month at 2.4 percent. The hires rate (3.4 percent) was little changed and the separations rate (3.1 percent) was unchanged."

Despite a slight step backward in May, the overall trend in job openings has been positive—Calculated Risk has the picture—but in a sense this fact has just deepened the puzzle of why the unemployment rate is so darn high. As we wrote in the first quarter issue of the Atlanta Fed's EconSouth:

"The disconnect between the supply of and demand for workers that is reflected in statistics such as the unemployment rate, the hiring rate, and the layoff rate can be dynamically expressed by the Beveridge curve. Named after British economist William Beveridge, the curve is a graphical representation of the relationship between unemployment (from the BLS's household survey) and job vacancies, reflected here through the JOLTS."

Since the second quarter of last year, the unemployment rate has far exceeded the level that would be predicted by the average correlation between unemployment and job vacancies over the past decade. Tuesday's report indicates that the anomaly only deepened in the first two months of the second quarter.


The dashed line in the chart above, which is estimated from the data from 2000–08, represents the predicted relationship between the number of unemployed persons in the United States and the number of job openings. That simple relationship would suggest that, given the average number of job openings in April and May, the unemployed would be expected to number about 10.4 million—not the nearly 15 million we actually saw.

Some analysts have suggested the unemployment benefits policies of the last couple of years may be responsible for abnormally high unemployment rates. Estimates generated by several researchers in the Federal Reserve—here and here, for example—suggest that extended unemployment benefits may have increased the unemployment rate by somewhere between 0.4 and 1.7 percentage points. But even if we accept those numbers and adjust the Beveridge curve by assuming that the number of unemployed would be correspondingly lower without the benefits policy, it's not clear that the puzzle is resolved:


If you tend to believe the higher end of the benefits-bias estimates, no puzzle emerges until the second quarter of 2010. And, of course, some estimates apparently deliver an even larger impact of the extended benefits policy. Let's call the question unsettled at this point.

The most tempting explanation for the seeming shift in the Beveridge curve relationship (to me, anyway) is a problem with the mismatch between skills required in the jobs that are available and skills possessed by the pool of workers available to take those jobs. The problem with this tempting explanation is that it is not so clear that the usual sort of structural shifts we might point to—for example, only nursing jobs being available to laid-off construction workers—are so obviously an explanation (an issue we explored in a previous macroblog post).

But these sorts of subplots may miss the truly big part of the story. I have noticed a recent spate of articles repeating a theme we hear anecdotally from many sources, in many industries. For example, this from a June USA Today article

"…the [auto] industry is poised to add up to 15,000 this year and could need up to 100,000 new workers a year from 2011 through 2013.

"…Automakers need workers with more and different skills than in the past on the factory floor.… Among priorities: computer skills and the ability to work with less supervision than their predecessors. That likely means education beyond high school."

… or more recently, this one from the New York Times:

"Factory owners have been adding jobs slowly but steadily since the beginning of the year, giving a lift to the fragile economic recovery…

"Yet some of these employers complain that they cannot fill their openings.

"Plenty of people are applying for the jobs. The problem, the companies say, is a mismatch between the kind of skilled workers needed and the ranks of the unemployed."

Now I realize that a few anecdotes don't make facts, but I have been in more than a few conversations with businesspeople who have claimed that the productivity gains realized in the United States throughout the recession and early recovery reflect upgrades in business processes—bundled with a necessary upgrade in the skill set of the workers who will implement those processes. This dynamic suggests that the shift in required skills has been concentrated within individual industries and businesses, not across sectors or geographic areas that would be captured by our most straightforward measures of structural change.

The data necessary to test this proposition are not easy to come by. That challenge is unfortunate, because the return on figuring out what is beneath those Beveridge curve graphs is very high.

By Dave Altig, senior vice president and research director at the Atlanta Fed

July 16, 2010 in Data Releases , Employment , Labor Markets , Productivity | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference A curious unemployment picture gets more curious :


Could you describe a bit how you created the second figure? Here's how I would have thought the chart would be constructed. If the unemployment is currently about 10%, and is overstated by 2.5 percentage points, then the number of unemployed workers should actually be lower by 1/4. It looks like there are currently 15000 unemployed, so the number for 2010q2 should fall to 11,250, in which case it would be on the beveridge curve. Is it possible that you cut the number of unemployed by 2.5 percent, instead of the 2.5 percentage points, or am I just misunderstanding?

Posted by: Ian Dew-Becker | July 16, 2010 at 08:52 AM

1. Manufacturing employment has been falling since at least 2000 (http://data.bls.gov/PDQ/servlet/SurveyOutputServlet?series_id=CES3000000001&data_tool=XGtable ). Compared to that lose of 6 million workers, hiring another 100,000 over 3 years is a tiny number; it seems unlikely that the auto industry really can't find enough qualified people in that pool. Worst case, if none of the skilled workers are unemployed, they'd have to pay a bit more to get people to switch from existing jobs (which could then perhaps be filled by the unemployed).

2. Alternatively, if the mismatch in skills covers a sufficiently large set of workers, and companies are that desperate to hire people, one would presume that they would invest in training. From the companies' point of view this would be equivalent to a higher compensation expense, except that some of the money would end up enhancing worker skills instead of going into their pockets.

So, either companies are irrationally unwilling to pay enough to get the workers they need, or they feel there is not enough demand for their products for them to be able to pay more for workers. I'd take a little from column A, a lot from column B.

Posted by: Itamar Turner-Trauring | July 16, 2010 at 10:56 AM

Maybe a better explanation is that the JOLTS will never go to zero (unless an asteroid hits) :). The numbers of unemployed is a function of the number of hires, but also a function of the number of terminations. During much of 2009, the economy was shedding a lot of jobs from some sectors but not others. The JOLTS was reflecting the recession proof areas of the economy ONLY in 2009 and could not go lower because companies that are shedding jobs cannot have negative hires. Job loss has not returned to its pre-recession level of 300-350K per week and is at a much higher range of 420-460 per week. If one is looking for a shift in the Beveridge curve, then a higher rate of job loss will require a higher rate of new hires to arrive at a level of employment- a shift in the curve. A shift in the rate of job turnover will shift the Beveridge curve.

It has been noted that the Beveridge curve has been subject to shifts in the past and the cause is debated. (See Dickens:


and references therein. Dickens has a nice figure showing previous shifts in the curve.

Also note in using the JOLTS data, the correlation is more linear if the JOLTS and unemployment numbers are offset by 1 quarter. (JOLTS is a leading indicator of unemployment). That is, unemployment number more closely tracks the JOLTS of the previous quarter than the current quarter. The linear relationship obviously breaks down at higher levels of unemployment. JOLTS will asymptote at a non-zero level. A better fit occurs if the JOLTS is plotted against the log of the unemployment number. The need for an explanation of the departure from the linear relationship may have nothing to do with a shift in the curve. It is likely that unemployment in this recession is so high that we have dropped below the linear portion of the Beveridge curve and onto the tail.

Posted by: jonny bakho | July 16, 2010 at 10:24 PM

I agree that there has to be a (Kurzweil) acceleration in the changes in technology that are affecting employment.

But, could the really big story be that employers still prefer China?

800K employees, 62Bln in revenues at Foxconn makes one wonder.

Posted by: Joe Rotger | July 17, 2010 at 10:03 AM

This anomaly might just come under the heading of Beveridge curve dynamics. (I wouldn't want to declare a shift in the curve based on just one observation.) Presumably it takes time to fill vacancies. (Even in a weak economy, it takes time, because employers have more applicants to process.) So an exogenous increase in vacancies should only gradually be reflected in the unemployment rate.

The size of the jump is pretty striking, though, and it's consistent with what I've seen in other indicators (e.g., the Monster Employment Index rising 21% over 12 months). Assuming this represents a real Beveridge curve shift, it would seem to be a reversal of the trend of the past 25 years, where the Beveridge curve (as best we can tell from available data) has been shifting inward. It might mean an increase in the NAIRU (which under the circumstances may be good news, given the below-target inflation rate and the constraints on macro policy).

I've been studying the Beveridge curve off and on for the past 20 years, and I still haven't found or heard any really convincing explanations for the past shifts. But they do seem to correlate with shifts in the Phillips curve.

Posted by: Andy Harless | July 17, 2010 at 10:35 AM

Wondering if expectations of the future have anything to do with it?
Next year, we get a massive tax increase. I can't help but think that business is figuring in a slow down in activity. http://www.bls.gov/news.release/empsit.t18.htm
This chart shows hours worked is up slightly. Instead of new hires, which are expensive, it's cheaper to pay a little overtime.

I think that unemployment will be over 10% by December.

Posted by: Jeff | July 18, 2010 at 10:03 AM

fifteen million unemployed, hmm, must have graduated from the enron school of economics if u beleve that.the only reason employers can't match skills is simply because they expect u to know it all, walk on water and work like you've been there twenty five years. guess what, never happen.

Posted by: gangsta | July 18, 2010 at 10:27 PM

Not mentioned above, but frequently included in discussions of this sort, is the issue of labor mobility.

Given the state of the housing market, many skilled laborers with homes ... and underwater mortgages ... may be unwilling to relocate to where the jobs are.

Also, dual-income families may be either more abundant than formerly, or less inclined to move for one spouse's job than they were in the past. Given the uncertainty in the economy, it may be that unless both breadwinners can relocate and find jobs, families with one stable income may find it more difficult to uproot themselves and hope for a better life elsewhere.

Posted by: Wisdom Seeker | July 19, 2010 at 01:21 PM

Two other comments:

(1) The unemployment rate is clearly pocketed based on age and skill sets. But perhaps, instead of "needing" more skilled workers (what a B.S. waste of language), businesses ought to be thinking about how to make better use of the available workers?

(2) Employers also appear to be unwilling to take chances on older workers, regardless of skill sets. Alternatively, the older unemployed are holding out for the best possible job offers. (Or maybe they are just the most rooted and unable to move?)

Calculated Risk had a nice guest post on this a week or two ago. The older unemployed have the longest durations of unemployment.

Posted by: Wisdom Seeker | July 19, 2010 at 01:22 PM

Yes, that is quite curious. It is hard to know to what extent the various factors are affecting unemployment.

Posted by: Mark B. | July 19, 2010 at 02:10 PM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

Google Search

Recent Posts



Powered by TypePad