What the Recent Job Openings Revisions Tell Us About Structural Unemployment

I just wrote a short paper titled Dramatic Job Revisions Bust Structural Unemployment Myths (pdf). Here is the abridged version:

“When the facts change, I change my mind. What do you do, sir?” – John Maynard Keynes

• The debate over structural unemployment that has taken place surrounding high levels of job openings and the Beveridge curve in 2010 needs to be reconsidered. Due to a pre-recession calibration of its birth/death model, the Bureau of Labor Statistics dramatically overestimated the number of job openings throughout 2010. It corrected the numbers for 2009 through 2010 in March 2011.
• On average, there were 172,000 fewer job openings per month in 2009 and 235,000 fewer job opening per month in 2010, reducing the job opening rate by an average of 0.18% over 2010 than had previously been reported. A quick and simple analysis shows that this BLS correction would have dropped estimates of increases in the NAIRU at the time from 1.3% to 1.0%.
• Since April 2010 when concerns about structural unemployment started to enter the debate, the Beveridge curve has almost entirely shifted leftward, with unemployment down roughly 1% and job openings up a meager 0.1%. Runaway unfilled job openings haven’t shown up in the data.

The Bureau of Labor Statistics reported that the number of job openings spiked in the summer of 2010. While this might normally be a cause for celebration, it actually worried a number of economists and policymakers. If an increase in the number of job openings co-exists with high unemployment, it means firms’ ability to find workers has decreased. If this signals that an economic model called a Beveridge curve has shifted outward, it means structural unemployment has increased and the ability of the economy to match workers and firms has decreased. In general, demand-based stimulative fiscal and monetary policy will do little to help the economy when it faces these kinds of problems.

That this increase in the number of job openings the BLS was reporting changed the nature of the debate on the job crisis is an understatement. The president of the Federal Reserve Bank of Minneapolis, Narayana Kocherlakota, used the numbers to state that unemployment would be at 6.9% without structural problems and that “it is hard to see how the Fed can do much to cure this problem.” Former President Bill Clinton told NPR and the David Letterman Show that “first time new job postings are opening up twice as fast as job hires….because of skills mismatch….[Without these problems unemployment would drop from] 9.6 to 6.9,″ echoing Kocherlakota’s analysis. David Brooks used Kocherlakota’s analysis in a 2010 editorial about structural unemployment, noting, “One of the perversities of this recession is that as the unemployment rate has risen, the job vacancy rate has risen, too.”

A large debate broke out among economists, with one side saying that the rise in job openings showed deeper problems in the American workforce and another side saying that the rise in job openings was a natural part of a cyclical downturn and that we’d see circular motions in the Beveridge curve. Conferences were held, panels convened; Peter Diamond spent part of his Nobel Laureate lecture discussing these issues. Both sides took for granted that the 2010 job opening rates had increased to the extent that the BLS reported they had.

In March of 2011, the Bureau of Labor Statistics revised its numbers for job openings from 2006 through 2010, dramatically reducing the number of job openings for the last months of 2008 and the entirety of 2009 and 2010. The revisions are in Figure One below - click charts to enlarge .

(Full revisions, 2009 revisions, 2010 revisions.)

On average, there were 172,000 fewer job openings per month in 2009 and 235,000 fewer job opening per month in 2010 than had previously been reported. This reduced the job opening rate by an average of 0.18% over 2010. The average number of job openings in the pre-correction data was over 3 million during 2010; in the new data, the number of job openings only reaches 3 million once. This is a massive shift downward from the previously reported number of job openings.

Why Did this Shift?

Why were the revisions so dramatic and entirely downward? As people from BLS explained, JOLTS samples from data in their sampling frame, and there’s a year lag between establishments entering their sample. As such, they can’t sample from businesses in their first year while they are outside their sampling frame. A lot of hiring and job activities happen in the first year, so it is important for BLS to estimate this figure.

In order to compensate for this, the BLS JOLTS team created a birth/death model (for births and deaths of business establishments) in 2009 to help gather numbers related to jobs. This model forecasts hires and separations of businesses in their first year. It was inserted into BLS numbers in 2009 and used values from 2000 to 2007 to populate the model. It then projected those values forward. Given that the 2000 to 2007 calibration didn’t feature a recession as deep as the one that occurred post-2008, this biased its birth-death model toward higher numbers.

So the birth model didn’t include data from the recession. The model wasn’t updated in 2010. In March 2011, it updated this birth/death model with additional data from the recession of 2008-2009 that then became available. Before it had used data through 2007 forecasted into the future, and now it uses data through 2009 forecasted into the future.

Now the birth model takes the recession into account. The difference is obviously dramatic, and when it went back and did a revision it drove down all the rates from 2009 and 2010. Since unemployment was much higher and business activity more depressed in 2008 and 2009 than in the seven years before, the previous projections forward were too optimistic. This is why the job opening rates were revised downward.

As Figure 2 shows, this has important implications for an economic model called a Beveridge curve, shifting the curve downward during the recession and subsequent period.

What Does This Downgrade Mean For Structural Unemployment?

In order to estimate how much this downward shift would change our estimates of an increase in structural unemployment, I created a quick and simple regression on the Beveridge curve using data from JOLTS for 2001 through 2009 (see Figure 3). For the year 2009, I used the average between the corrected and uncorrected data.

I then projected how far out the unemployment curve has moved for the values of the year 2010. To take an example, in October 2010 the unemployment rate was 9.7%. The uncorrected job opening rate was 2.5, which corresponds to an unemployment rate of 6.29%. The corrected job opening rate was 2.2, which corresponds to an unemployment rate of 7.25%. With the corrected job opening rate we see less of a shift of the unemployment curve.

The Federal Reserve Bank of San Francisco argues (pdf) that “Credible estimates of the natural rate over these earlier periods suggest that it may have changed about half as much as the horizontal shift in the Beveridge curve.” Using this rule of thumb, I found that my estimate gives an increase in the NAIRU of 1.30% when I use uncorrected job openings, findings similar to other estimates, including the Federal Reserve Bank of San Francisco, which estimates that this uptick is largely temporary. This leaves a lot of cyclical slack with unemployment at 9%.

With the corrected job openings data, we see the estimate of an increase in NAIRU drops to 1%, which means using the corrected numbers should drop the estimate of NAIRU 0.30%.

What has Happened Since Summer 2010?

Policymakers have worried that structural unemployment would stifle the recovery. The implication is that job openings would continue to rise while unemployment stayed high.

In both the corrected and uncorrected version of the job openings data, job openings peaked in April of 2010. What has happened to the curve since then? Figure 4 shows what has happened to the curve over the past two years (it uses corrected data).

There has been a straight decrease in the unemployment rate with virtually no increase in the job opening rate. The number of jobs available has remained flat and unemployment has fallen roughly one full percentage point. Those who were worried that any decrease in unemployment would happen alongside a runaway rise in the job opening rate should shift their concerns back toward low job growth.

The numbers guiding policymakers and experts are subject to revision, and job openings, a key component of any recovery, have turned out to be much lower than anyone had expected during 2010 when the debate over structural unemployment started.

Even before this, there was considerable evidence that there is tremendous slack in the economy and room for monetary and fiscal stimulus. With this correction, the story of an economy held in check by a weak labor force rather than weak jobs is harder to justify, and the argument for doing more is stronger.