Graph of the Day: Just how tight is the relationship between inflation and wage growth? (Answer: not very)

August 19th, 2014 at 3:09 pm

There’s been of late an interesting outpouring of quality research on one of my favorite topics: the relationship between labor market slack and wage growth. I’ve got a longer piece on this coming out later, but the punchline is, as you probably know if you’re drawing a paycheck: considerable slack remains in the job market and it’s still a drag on the real wage growth of most workers.

I’ve got links below to the pieces I cite in my forthcoming analysis, but here’s a figure I think deserves more attention than I gave it, from this new paper by Knotek II and Zaman, economists at the Cleveland Fed.

It takes a bit of explaining. The lines in the figure show the errors from a model designed to predict inflation. (A BVAR is a Bayesian Vector Auto Regression, which sounds a lot weirder than it is: it’s just a way to tease out the ways in which a bunch of economic variables effect each other, with few restrictions placed on the model). As you go down the X-axis, you’re forecasting further out in time, so as you’d expect, the errors grow.

But what the researchers were really asking here is whether adding wages to the model improved its accuracy. After all, there’s a lot of folks running around claiming that unless they want to face uncontrollable (or un-anchored, in Fed jargon) inflation, the Fed better take pre-emptive action against incipient wage growth. Yet, as this and other analysis in the paper reveal, linkages between  wage and price growth are but weakly correlated. Adding various different wage series (that’s what those acronyms are: AHE, CPG, ECI) to a forecast model for inflation hardly improves its accuracy at all.

What’s the implication? It’s that Fed policy makers should not assume that wage growth is de facto inflationary. Go ahead and keep your powder dry, FOMC, but in the words of my monetary policy mentor, Dr. Robert Marley, when it comes to wage growth, don’t “kill it before it grows!”*

infl_errs

Source: Knotek and Zaman

*You know the original lyrics, right?

I raised the funds rate!
But I did not raise the discount rate.

Oh no, oohhh

I raised the funds rate!
And they say it will make capital less tense.

Relevant links:

https://www.clevelandfed.org/research/commentary/2014/2014-14.pdf

http://chicagofed.org/digital_assets/publications/chicago_fed_letter/2014/cfloctober2014_327.pdf

http://eml.berkeley.edu/~jrothst/workingpapers/rothstein_labormarket_update201406.pdf

http://www.federalreserve.gov/econresdata/notes/feds-notes/2014/effect-of-labor-slack-on-wages-evidence-from-state-level-relationships-20140602.html

 

Has the Beveridge Curve Really Shifted?

August 13th, 2014 at 10:58 am

The Beveridge Curve (BC) is a favorite tool of labor economists showing the inverse relationship between job openings and unemployment. It’s thus a kind of index of strength of labor demand: when the job market is tight, there’s low unemployment and more unfilled openings/job vacancies, and vice versa.

As shown below, using the BLS monthly version of the BC, openings are on the y-axis and unemployment is on the x-axis, so the curve slopes down from the upper left (tight job market…yay!) to the bottom right (weak market…boo!).

bev1

Source: BLS

So far, so good, but the BLS’s BC shows a marked shift in the curve toward the northeast. Whussup with that?!

Such shifts in the BC are thought to represent changes in the efficiency of the matching process in the job market. That is, if, as the recovery progresses, employers have openings but they can’t match the applicants they’re seeing to the jobs that need filling, the BC moves as it does in the figure. It could be a skills mismatch, where applicants’ skill supplies are notably weaker than employers’ skill demands. Or it could be geographical, such that the employers who are hiring are in different physical places from job seekers.

If you follow this part of the debate, you know that there are many who’ve made precisely this skills mismatch argument, often based on these data. I’ve been on the other side of that debate, arguing that the problem has been weak demand. What’s missing, I’ve argued, are jobs, not skills.

[insert defensive caveat paragraph here] To be clear, I’m not saying such skill mismatches don’t exist. Of course they do. Nor am I saying every applicant has the skills they need—surely many would be better off with more training/education.

I’m saying that right now, these skills issue are secondary to the demand shortfall, a view that carries the policy implication that we should focus first and foremost on reducing slack (though of course slack reduction and better training are by no means mutually exclusive).

How then, do I explain this shift in the BC? A few years ago, those on my side of this aisle thought that perhaps the shift was temporary but we’re now five years into the recovery. So what say we now?

As economist Gary Burtless reminded me, a recent paper by Krueger et al shows this to be at least partly a function of long-term unemployment (jobless for at least 26 weeks), a particularly serious problem in this recession/recovery. The following graphs, updated from Krueger et al, confirm this point by plotting the same BLS graph on the top (without the formatting bells and whistles) and the same figure but with short-term unemployment (jobless less than 26 weeks) replacing total unemployment on the bottom. The shift pretty much disappears (a statistical test confirms the significance of the shift using overall unemployment and its insignificance using short-term rates).

bev2

Source: my analysis of BLS data.

So what the heck does that mean?

One interpretation is that the long-term unemployed are not really in the labor market; they’re not really looking for work and not contributing much to slack. In that case, it’s the short-term unemployment rate that really matters. Re the BC, this implies no change in the efficiency of job matching, just a measurement problem with the total jobless rate.

But there’s a problem, one that the aforementioned Burtless underscores in his analysis of the alleged skills mismatch: the absence of any wage pressures. If much of the slack has been squeezed out of the job market, as the short-term unemployment rate suggests—it’s back down to its historical average of around 4%–and even more so if employers can’t find the workers they need, then we ought to see them bidding wages up to get and keep the workers they need.

The fact that we don’t see such wage pressures poses a stiff challenge to both of these arguments: the shift in the BC and the long-term-unemployed-don’t-add-to-slack. Simply put, it’s awfully hard to square a bunch of job vacancies and flat wage growth. If employers really wanted to fill those vacancies, we should see evidence on the wage side. And please don’t invoke lags, as in “wage pressure is right around the corner!” That may be true—I think and hope it is.  But as I’ve stressed, the shift in the BC is years old by now.

Economist Jesse Rothstein has also cast doubt on the BC’s outward shift, arguing that as measured by the data source in the figures shown throughout–the BLS JOLTS–we shouldn’t assume that more job openings mean more labor demand, especially, once again, absent faster wage growth.

The problem, Jesse argues, is that job vacancies as measured by the JOLTS and vacancies that underlie the theory of the BC are quite different in ways that make it hard to interpret that shift. Interestingly, the problem is the weak labor market itself.

The matching theory behind the BC is simple: an employer has an opening, someone shows up who meets the basic qualifications and the firm hires her right away at the market wage. At full employment, something closer to that dynamic may hold. But in slack labor markets, not so much. Back to Jesse:

If employers are indeed taking advantage of the weak labor market to reduce offered wages or to hire more qualified workers, one would expect this to reduce the rate at which posted vacancies are filled and therefore to raise the job openings rate. This limits our ability to diagnose labor market tightness based solely on the aggregate Beveridge Curve.

[See Jesse's updated paper on this and related questions, including a deep dive on the absence of wage pressures which is key to accurate diagnosis and prescription in this space right now.]

Here’s what I think is going on. The job market has in fact been tightening, but in a somewhat unusual way: through diminished layoffs more so than through robust hiring (see figure here and this important related work as well).  The former—fewer layoffs—is keeping the short-term unemployment rate nice and low. The latter—tepid hires—is keeping the long-term jobless rate high. That’s creating the illusion of decreased matching efficiency but it’s really just the result of the persistent slack and the unusually high share of long-term unemployment with which we’ve been stuck for years now.

So yes, the BC has shifted but that shift is likely saying more about slack than matching problems.

This dynamic is improving (thus my hope re future wage growth, assuming the Fed doesn’t over-react and kill it before it grows). The labor force has stabilized in recent months, the pace of job growth, including new hires, has picked up, and the long-term unemployment rate has been coming down as well.

But we’re still far from full-employment and it will take more than diminished layoffs to get there. We need more robust job growth to both close the gap between vacancies and hires and to generate some long awaited wage growth. Like I said, I’m all in for more worker training, but at a time like this there’s nothing like a really tight job market to boost matching efficiency.