Saturday, March 1, 2014

What If Labor Markets Are Tight?

Maybe the unemployment rate isn't as misleading as we might think. Maybe it doesn't massively overstate the amount of progress and healing in the labor market. I'm working on a piece for next week, and I thought I'd post this chart and some comments as a preview.

It's a scatterplot of the U3 unemployment rate on the horizontal axis and the quit rate on the vertical axis. In tighter labor markets, people tend to quit their jobs and find better ones. In looser labor markets, people cling to their jobs because unemployment is terrible. The data goes from January 2001 to December 2013.

What the chart shows is that we're seeing the expected number of people quitting their jobs given the current unemployment rate. I've put the most recent data point in red.

This is new and, in my view, compelling evidence that labor markets are pretty tight. Why? Think about the decision to quit. It's a function of your confidence that you'll find a better job quickly -- which embodies some unobserved but holistic measure of labor-market tightness. The fact that the unemployment rate appears to be a robust predictor of the quit rate both before and after the 2007-2009 recession suggests that the unemployment rate, our imperfect observation of labor-market tightness, is a close proxy for the tightness people think is important when they decide to quit or not to quit.

That the unemployment rate is not such a good proxy for an unobservable measure of labor market tightness is exactly the empirical claim underlying current monetary policy and, quite frankly, much of what I've spent the last two years writing about.

It's worth explaining why this particular way of looking at labor-market tightness is pretty clearly superior to the alternatives.

Previously, I've looked at the relationship between unemployment and broader measures of underemployment, and Paul Krugman has done some similar stuff, if a bit less formally. The point that comes out of those analyses is that the U.S. has a lot of slack labor capacity sitting around.

But that slack might not matter. If the long-term unemployed are disconnected from the labor market, they really can't matter much in a macroeconomic sense -- that is, their unemployment no longer has the power to restrain wage growth or discourage the employed from quitting and switching jobs. It's worth pointing the reader here to ideas like the insider-outsider theory of unemployment and labor-market segmentation.

That's why you can't look at underemployment directly to conclude that the unemployment rate is misleading -- because it assumes that underemployment is macroeconomically relevant, which is what you're trying to prove.

What the graph above suggests is plainly that job-switchers don't think the long-term unemployed are any competition at all. If they were -- if the unemployment rate overstated the amount of labor-market tightness -- then we should have seen the relationship break down. In particular, the curve should have shifted downwards and to the left. There should be fewer quits for any given rate of unemployment.

It's also the kind of evidence you should take seriously if you take the Diamond-Mortensen-Pissarides model of labor-market search seriously. The DMP model, to explain it informally, says that labor markets are messy in that the jobless can't find vacancies instantly -- that is, there are "frictions" in the labor-market that put the search problem at the center. If search matters, then quits matter, and then the unemployment rate can't really be understating labor-market tightness in the meaningful sense.

There is also a close intellectual connection in my chart to the Beveridge curve. It seems about as important. The Beveridge curve broke down during the recession because of long-term unemployment, as economist Rand Ghayad has shown. Mine could too. But if it did, that would be exactly the kind of evidence we'd need to say that labor markets are looser than they look.


  1. If the population was aging into retirement, so there was a well predicted decline in labor force participation (independent of a part that might be due to hysteresis), wouldn't that push the quit rate lower?

    All Ghayad shows in that paper is that the proportion of the unemployed who are long-term unemployed is high as a historical matter, which we know from summary statistics. Given that job hires aren't broken down by any kind of duration measure, I think his invocation of the Beveridge curve confuses more than it enlightens. If you look at the job finding rate of the short-term unemployed, it's still low compared to pre-crisis.

  2. Playing with the data, note that if you are positing a curve here akin to the Beveridge curve, the recovery features quit rates higher during late 2009 to now than a fitted line would predict (you can eyeball it in your graph).

    Which means quit rates were higher than predicted at 9 percent unemployment, which means labor markets were more "tight" than expected throughout 2010. That doesn't seem right.

  3. * "push the quit rate higher" in first comment.

    1. This is a response to all 3 comments.

      On the demography point: Yes. If the LFPR decline is mostly demographic -- and I think the evidence on this is really unclear -- then that's another reason to think labor markets are tighter.

      On Ghayad, the reason the analysis is phrased in the Beveridge curve is b/c an apparent outshift was being used as evidence for a higher NAIRU. I think his study is very useful in explaining job matching there. But your work on job finding rates has been super cool too!

      On the second comment: Not sure if you saw my follow-up piece with the regression results. It looks like the quit rate also incorporates the change in unemployment, not just its level. People don't want to quit their jobs right before the economy stinks, but once we hit bottom in 2010, the momentum effect reversed. Which makes some sense as a forward-looking behavior.

    2. Retirees aren't included in quits (they are apparently "other separations"), so I walk back my first comment.

      Let me put it another way - this argument assumes that the quit rate is solely a function of people quitting to take a new job or find a new job. But if people are quitting to leave the labor force, perhaps because it's a slack labor market, then it's not an actual measure of labor tightness in the sense you mean it. It's not clear how you square that with the measurements available in JOLTS.

    3. Now that's very interesting, about other separations.

      What other motives for quitting are there that would have changed meaningfully since 2000? Retirement was the only candidate that made sense to me. SSDI? Why would people quit because the labor market was slack?

    4. I would love to see the evidence supporting the claim that demographic shifts explain little or none of the 2007-to-2014 decline in the labor force participation. A weak economy surely accounts for part of the decline, but the huge share of 50-to-67 year-olds in the population accounts IMHO for at least half the decline.

    5. One reason would be the huge wave of people returning to school during the crisis, which drives at least a part of the labor force participation rate.

      To put it another another way, there's a good, or at least good enough, reason why we look at a Beveridge curve to understand these things, because we assume the hiring process for a job opening rate is well founded in economic logic. (And even then there's massive problems with the endogenity of search effort in a weak economy and the circularity Peter Diamond emphasizes.) It's not clear to me how quit rates evolve at 10% unemployment.

    6. There have been many more young "non-starters" (high school/college graduates) rather than older "quitters" (retiring Baby Boomers) that's driving down the labor force participation rate. And the LFPR has been in decline since April 2000 — almost 8 years before the first Boomer took an early retirement in 2008. FAct is, there simple are not enough jobs being created. Decades of offshoring is mostly the biggest reason for this.

      Boomers turned 62 in 2008

      Year-------Social Security Retirees-----------High School Grads

      2008------ 748,047----------------------------------3,001,337

      Boomers turned 65 in 2011

      Year-------Social Security Retirees------------High School Grads

      2011------1,006,724---------------------------------- 3,103,540

      And we're not even counting high school drop-outs. All data and links in the link below:

  4. "the long-term unemployed ... can't matter much in a macroeconomic sense -- that is, their unemployment no longer has the power to restrain wage growth or discourage the employed from quitting and switching jobs"

    I think you're saying they might not matter in this particular macroeconomic sense. They still might matter in their effect on aggregate demand, for instance. I don't think there's anything to suggest here that wage increases due to the presumed tight employment market will be equivalent, in aggregate, to what would arise from a higher participation rate.

  5. The unemployment rate has been understating the strength of the recovery:

    1. Responded in a comment on your post.

    2. cross-posted comment from my blog:
      Thanks for your response, and the references.

      Regarding (1), there are many studies that give a range of relationships, and some are much higher than the Farber Valletta study. Here is a study by David Grubb at OECD:

      Here's a GAO survey that shows behavior of workers who exhausted EUI. Less than 1/5 left the labor force. Also, they skew very old:

      Regarding (2), have you seen this report from the Boston Fed?

      It shows different behavior by age and reason for losing a job. For older workers, unemployment rates for job leavers have been normal, but unemployment rates for job losers (eligilble for EUI) have been high. This pattern is not clear in middle-aged workers, and reverses among younger workers (young job losers look normal, but job leavers, reentrants, and new entrants have higher unemployment).

      I would suggest that this bifurcated behavior of the disaggregated Beveridge Curve is a sign of the distortions from EUI, and regular looking ST unemployment together with high LT unemployment is exactly what we would expect to see if we thought EUI would distort labor markets.

      Also, the job opening behavior is what you would see in a labor market with less intense available labor supply, as described in my oil analogy above. If there are job openings available in jobs where highly qualified and experienced workers are available, but are being less aggressive about taking new employment due to a temporary policy, then we should expect employers to be less aggressive about filling jobs with applicants from the willing pool of workers - much like the oil driller would be ambivalent about drilling $60 oil in a $100 market when they know that $40 oil is being kept under the ground by political issues.

      If I was looking for evidence that the unemployment rate was overstating slack in the labor market caused in part by EUI, your evidence from point (2) is the kind of thing I would expect to find. (Although, I'm sure there are issues in the recruiting intensity paper that would challenge my interpretation. I haven't gone through the whole thing.)

      In fact, my broad estimate of the extra unemployment coming from EUI is based on comparing unemployment durations under 26 weeks with unemployment durations over 26 weeks. There has always been a very tight relationship between these measures, which broke down in the recent recession.

  6. Interesting that you mention the vacancy rate so much. Have you seen the Shimer 2005 paper about predicting unemployment rates in the US with job finding rates and quit rates? He concludes that the finding rate is much more indicative of unemployment than quit rates - and by my extrapolation, labor market health. The finding rate still hasn't recovered to its pre-recession rates, which would indicate that there is still slack in the market.

    And through the DMP model lens, the above could be explained by a high matching rate for employers, but not for workers because of the large pool of long-term unemployed.