Here’s another salvo in the great labor force slack debate.
Summarizing what is now a large and growing canon, a lot of digital ink has been spilled over three measly percentage points. That’s the amount the labor force participation rate (LFPR) is down from its pre-recession peak of about 66%, and the big question is how much of it is cyclical and how much structural. The intuition is that the cyclical part—a function of continued job market slack—can be restored while the structural part is gone.
Most work suggests the cyclical part of the gap is falling over time and is now quite minimal, as both the recovery takes hold and “scarring” effects make it tougher for labor force exiters to find their way back into the job market. However, a look at the relationship between the LFPR and wage trends suggests this conclusion may be premature.
A few months ago I reported on some important work by Blanchflower and Posen (B&P) showing that the decline in labor force helped to explain wage trends. Why should this matter? As I wrote back then:
B&P’s finding that including labor force inactivity in their wage analysis helps explain some of the variation…has a very important implication. If the depressed labor force is a statistically identifiable contributor to slack, then some of the current labor force inactivity can be reversed, much the same way unemployment comes down in the face of strengthening labor demand. Or, put differently, part of the decline in the labor force is just slack that’s not measured by the unemployment rate.
Here, I’d like to expand a bit on that insight.
First, updating some analysis from a few months ago, see the first figure below. In an analysis designed to show that the unemployment rate still retains some validity as a measure of labor market conditions, Goldman Sachs economists showed how it correlates with recent wage trends (the wage trends are a pooled version of five different series). However, as the red line shows towards the end of the figure, using just unemployment leads to an upwardly biased forecast in recent data. This is not unexpected, as the jobless rate has fallen faster than the underlying slack, a point Chair Yellen frequently makes and a reason she uses her dashboard and other data mashups.
Source: see text
Following B&P, the green line adds the labor force participation rate (LFPR) to the model, which tracks the wage trends more closely and underscores B&P’s point about the cyclical component of today’s LFPR. Even in this toy model, it significantly improves the fit (B&P create a state panel that generates much more data and variance; that’s a much better approach to what I’m showing here—though even this bit of blog-o-metrics makes the point, IMHO).
The next two figures add to the analysis and raise some new points. The first figure shows the coefficient on the LFPR from a “rolling regression,” where starting in 2009q1, I add one quarter at a time. So the first data point in the graph is the coefficient from a model running from 1995q1-2009q1; the second data point is from a model that runs from 1995q1-2009q2, and so on.
Note how the LFPR coefficient climbs as observations are added over the past few years, suggesting that at least in terms of its correlation with wage growth, the cyclical component of the LFPR is growing over time, a finding that stands in contrast with much more detailed decompositions of the LFPR.
Source: see text
Of course, statistical significance matters and the next figure shows the t-statistics from the LFPR coefficients in the prior figure. Importantly, they start out insignificant but gain significance as the weak recovery proceeds, again suggesting that the LFPR is increasingly correlated with the weak wage trends. (Note: As a “falsification test” I ran this model using data back to the early 1980s and found the LFPR was insignificant and “wrong-signed.”)
Source: see text
While such simple modelling can only be called suggestive, both the increase in magnitude of the LFPR coefficients and its t-stats suggests that our depressed labor force embodies identifiable slack and that Chair Yellen is correct when she argues that a stronger labor market could likely pull some of those sideliners back in.
Data note: the model regresses the pooled wage trend on a constant, the unemployment rate, the LFPR, and an AR(1) term needed to whiten the residuals. I’ll post the data later so others can play with them.