Friday, August 22, 2014

The Great Unskewing

One way to think about the dynamics of jobs and firms is to examine the distribution of firm growth rates. The high pace of job reallocation we observe implies that firms vary considerably in employment growth rates; some firms are shrinking (or closing their doors), some are operating year after year without changing the size of their workforce, and some are rapidly expanding. We basically know about the first moment of this distribution (the mean or median), which must be slightly positive to accommodate growth of the workforce over time. The second and third moments--the width of the distribution and the degree to which it is skewed up by high-growth firms--are measures of dynamism.

Suppose we sliced the distribution of firms by firm age. The figure below is from our recent JEP paper (click for larger image):

Figure from Decker, Haltiwanger, Jarmin, and Miranda (2014a)

Here you can see the Geico "everybody knows that" fact about young firms "accounting for" aggregate job growth. As a cohort of firms ages, its growth rate distribution converges on zero--so old firms as a group do not create jobs in aggregate (more on cohort dynamics here and here). The second moment can be seen here too: young firms vary widely, but firm cohorts see their growth rate distribution narrow as they age (but note that the distribution never becomes degenerate--there is still lots of churning among old firms). In this sense, old firms are less dynamic than young ones. Finally, observe that young firms are skewed toward high growth. The gap between the median firm and the firm at the 90th percentile is larger than the gap between the median firm and the firm at the 10th percentile.

As part of the general aging of firms (though not entirely because of it), the aggregate firm distribution increasingly looks like the distribution of old firms. The next figure (from a current working paper) plots the gap between the 90th percentile and the 10th percentile firm, for all firms and for only continuing firms (i.e., no entry or exit):

Click for larger image
Figure from Decker, Haltiwanger, Jarmin, and Miranda (2014b)

Observe that this measure of the second moment--the dispersion of firm outcomes--has been falling, so the firm distribution has been tightening up. This might seem like a good thing, particularly if the tightening is happening because fewer firms are shrinking. This is basically the reason that some people are skeptical about the need to be concerned about declining dynamism. Who cares if dispersion is falling? A lot of that churning is unnecessary anyway. The next three figures examine the third moment--skew--by comparing the gap between the 90th and 50th percentiles to the gap between the 50th and 10th percentiles.

Figures from Decker, Haltiwanger, Jarmin, and Miranda (2014b)

While it is true that the bottom of the firm distribution is pulling up, the top is falling down more quickly. The US economy is unskewing, beginning to look more and more like the old timers. When we look at all firms in aggregate or just at the high-tech sector, it looks like the distribution could be totally symmetric soon. Among young firms, there is still a high degree of skew but it is falling (keep in mind that age-0 firms necessarily have positive growth rates, which gives the young firm distribution a boost).

The unskewing of the firm distribution means that we are seeing fewer high-growth firms (or, equivalently, the bar for what constitutes "high growth" has fallen). In partial equilibrium, at least, it's very hard to interpret that result as good news. It is likely to affect how the economy responds to the business cycle, and over time it could have consequences for productivity. For now these are speculative claims, but this is a growing literature. A new paper by Davis and Haltiwanger finds evidence that lower dynamism causes lower employment, not to mention slower wage growth. In my view this evidence greatly complicates the recession stories that dominate the econ blogosphere. It may not be a simple Econ 101 aggregate demand story.


  1. Okay, so let me swoop in and try to defend the "Econ 101 aggregate demand story", or at least some version of it. (I think I'm on record calling the AD/AS model an "abomination", so I won't defend everything from Econ 101. But "aggregate demand" is like my bat signal.)

    I find your JEP fascinating, but I have a hard time seeing how it maps onto our recent cyclical episode (as opposed to more general long-term stuff). In particular, how can we explain the massive drop in employment, which for most people was the most important aspect of the recession? Even if we're in for slower productivity growth, etc., due partly to reasons identified in this data, why does that mean lower employment?

    Let's ignore non-labor sources of income for a moment. Then, in a very simple static model of labor supply, unless substitution effects are much stronger than income effects (which would imply a counterfactual secular rise in labor hours), a fall in the level of real wages does not imply a decline in labor hours. If we adopt cancellation between substitution and income effects as a baseline, then hours should be constant regardless of the real wage. Not clear why a fall in productivity, say, should translate into a decline in hours.

    Dynamically, there is a similar invariance of labor hours to shifts in the long-run level of productivity, but there may be some intertemporal substitution of labor depending on the (1) real interest rate vs. the (2) rate of expected real wage growth. Specifically, labor supply today is increasing in (1) - (2), though the actual amount of intertemporal substitution is determined by the Frisch elasticity, which does not seem to be that high in practice. The recent recession involved declines in both (1) and (2), maybe with an overall decrease in (1) - (2), but (I don't think) quantitatively enough to get a large decline in labor supply. Intertemporal substitution is pretty problematic as a mechanism in general, since the people whose employment fell the most tended to be the ones who don't have much ability to smooth wages over time and thus can't engage in much intertemporal substitution.

    Now making this model a little less cartoonish, what about non-labor sources of income? Mostly this goes in the wrong direction, because the income effects from the collapsing values of securities should have induced more labor supply. But at the low end, means-tested or unemployment-contingent government transfers could have pushed down labor supply through both substitution and income effects. This is Casey Mulligan's thesis, which might have something to it - but overall, it's pretty lacking in empirical support, and the best evidence on some of the key questions (like unemployment insurance) suggests that the effects are real but small.

    Anyway, I could go on and on about the ways in which a classical labor supply model is not a good description of reality, and we'd probably agree there. My point is to emphasize that since it's hard to justify the decline in employment in a frictionless model, there are probably some substantial labor market imperfections at work here. And a lot of the most plausible imperfections suggest that we should pursue policies to boost aggregate demand, if only as an application of the theory of the second best.

    For instance, one parsimonious hypothesis that can explain a lot about labor markets is downward nominal wage rigidity. Most macroeconomists today prefer search-based models, but most such models that can explain the strong procyclicality of employment end up bringing in some form of wage rigidity anyway. And once they do that, there's a strong basis for Keynesian-ish policies.

    I guess I'd summarize it like this: no account of the recent recession is perfect, but I haven't seen many models without substantial rigidity-like labor market imperfections even being able to get the basic stuff right. Econ 101 aggregate demand still looks pretty good.

    1. Thanks Matt (and sorry for the delay).

      I definitely buy the argument that productivity growth is not necessary for maintaining high employment. Also I admit that I don't have a good model in my head for this stuff yet--so far my intuition is pretty partial equilibrium. And there is no doubt that nominal rigidities matter a lot. I'm not sure how to think of them over the long term though--the recession started in 2007, and 7 years later we're not back to full employment. By any assumption, nominal rigidities don't last that long. Do we have good models that rely on them that will give us such prolonged labor market problems? I think usually we have to add other stuff to generate that kind of persistence, no?

      As for my complaints about aggregate demand in light of firm dynamics data: Again, my intuition is very partial equilibrium, and that's being generous. I don't have a model in my head for how firm dynamics affect the business cycle, and it's hard to come up with one when we're so clueless about what is driving these trends. I'm not focusing so much on an employment effect via a productivity channel; I'm thinking more directly about firm hiring behavior over the cycle. My basic thought is this: how can the firm growth rate distribution unskew this way and still be able to accommodate growth in the labor supply? If young firm activity, which "accounts" for all net job creation, is on the decline, how will the slack be picked up in general equilibrium?

      I do think that any model that will do what I'm thinking will need some sort of nominal rigidity and/or labor market frictions. For example, in a standard firm dynamics model I could unskew things by changing the idiosyncratic TFP process, but wages would adjust so full employment endures. Incidentally, though, we can't find evidence that the TFP process has changed in the right way (at least in the manufacturing data).

      So obviously I'm being very non-rigorous. That's why I'm blogging it instead of trying to write a paper (in our papers we're much more cautious about what potential implications are). Basically my approach is to say, how could it NOT matter that firm dynamics are changing? In partial equilibrium, it seems like a very bad thing. Either this stuff matters for the business cycle--including this one--or it doesn't; if it doesn't, I'm very curious about WHY it doesn't.