Friday, December 1, 2017

[UPDATED] Wages, productivity, progressive policies, and serial correlation: I weigh in on an important, interesting debate. [feedly]

[UPDATED] Wages, productivity, progressive policies, and serial correlation: I weigh in on an important, interesting debate.

UPDATE: See below, at the end of this post, for Anna Stansbury and Larry Summers' thoughtful response to my blog. My goodness, a rational conversation…what a strange and rare event these days!

There's an interesting sort of argument going on between Stansbury/Summers (SS) and Mishel/Bivens (MB). My name has been invoked as well, so I'll weigh in. It's a "sort-of" argument because there's less disagreement than first appears.

It all revolves around this chart, which plots to the real compensation of mid-wage workers against the growth in productivity. For years they grew together, then they grow apart. The levels of both variables almost double, 1948-73, but since then, productivity has outpaced the real comp of blue-collar, non-managerial workers (mid-wage workers) by a factor of 6.

That wedge between productivity and middle-class wage growth has become one of the more important developments in political economy, representing the rise of inequality and the disconnect of paychecks and growth. It's even on a tee-shirt, produced by the group, Fed-Up.

SS agree that the split is real and important, but their motivation appears to be not throwing out the baby with the bathwater. That is, they're concerned—not without reason—that progressives see those two lines and conclude the growth and productivity don't matter. In a follow-up oped, SS cite me as suggesting that in the age of inequality, we can't count on growth to reach middle-class families: "Faster productivity growth would be great. I'm just not at all sure we can count on it to lift middle-class incomes."

That's true and even axiomatic, as per the figure above, but SS's point is that even so, it doesn't mean productivity growth doesn't raise median wages. It just means there's a bunch of other stuff pushing back in the other direction. That's true too.

[Note: it's very important in this debate to distinguish between median and average compensation. As inequality increases, they diverge. Since this debate is about inequality—i.e., the "other stuff" that's driving the gap—it's important to focus on the median as opposed to the average.]

In a series of regressions of median or middle-wage compensation (the latter is for blue-collar and non-managerial workers, for which there's a longer time-series) on productivity growth and unemployment, SS shows that the coefficient on productivity is often close to 1, meaning a 1 percent increase in productivity maps onto a 1 percent increase in wages. Again, this doesn't deny the gap. But it does say growth matters for wages, even at high levels of inequality, weak worker bargaining power, persistently slack labor markets, etc.

I'm sure that's right and it's one reason why so much of my work in recent years has focused on running full employment, full-growth-potential economies. I've written extensively and bemoaning-ly about the productivity slowdown, often trying to push back on those arguing its mostly measurement error. I've called it one of the biggest economic problems we face. Just a few weeks ago, I tried to show the critical role slower productivity growth was playing in current wage growth that's slower than it should be at such low unemployment (see 2nd-to-last figure here).

But, as I wrote in one of the links above: "Faster productivity growth is not by itself sufficient to raise the living standards of all who help to generate it, but it surely is necessary."

SS focus on the "necessary," which is fine and important, but in doing so, they create a bit of a straw man by casting those of us who focus on the "not-sufficient" as inadequately committed to faster productivity growth. That's inconsistent with our writings. FTR, I do think Summers has done a particularly good job in recent years putting equal weight both sides of this equation, lending his heft to not just growth issues, but inequality issues as well (he's also been on a recent and admirable tear against the awful tax cut plan).

I won't belabor this because the MB response is so thorough and I've little more to add. I will, however, underscore one key policy point they make and add a neat econometric point to which I'd like SS to respond.

MB write:

Our contention all along has been that this pay deceleration did not just reflect slower productivity growth, but that it in fact reflected a number of intentional policy decisions that undercut typical workers' ability to demand and achieve higher pay. One such policy decision was exactly over how aggressively the Federal Reserve and other macroeconomic policymakers should target low unemployment. Others included decisions about whether or not to protect workers' rights to organize and bargain collectively (the country obviously chose not to) and whether or not to raise the federal minimum wage in line with inflation or productivity growth (again, we chose not to).

With this in mind, those who would close that gap, which include SS, MB, and JB (that's me), would, I suspect, readily admit that we have a lot more confidence in our knowledge of gap-closing policies in the space MB reference above versus ideas to boost productivity growth. The latter, I regret to say, remains largely a mystery to economists.

So, pushing for higher minimum wages, full employment (direct job creation), progressive taxation, collective bargaining, overtime rules, gender equity, a robust safety net, more balanced trade, financial market regulation (a complement to full employment—we can't have them blowing up the economy every cycle), and so on are gap-closing ideas that we know will help. What's more—and this part is important given SS's findings—these measures are not anti-growth.

The punchline is thus twofold. Of course, 1) faster productivity growth is a necessary component of faster wage growth. But so is 2) re-linking wage growth with the ongoing productivity growth we have. And we know more about how to achieve #2 through progressive policies than #1.

Two econometric points:

First, Biven shared the data with me and I replicated one of SS's main findings: a coefficient of 1 on the productivity variable in a simple specification of mid-wage comp on productivity growth and unemployment. But they neglect to mention that the residuals are serially correlated in that regression (DW=0.54), implying an omitted variable bias. That's not surprising: you can't explain a complex variable like mid-wage comp with just productivity and unemployment. But it is a problem for their trade-off conclusions if what's missing correlates—shares explanatory power—with productivity growth.

Which appears to be the case: when I add an AR(1) term to whiten the residuals, the productivity coefficient falls by half (to 0.5) and DW=1.5. To be clear, that doesn't undermine their point. The half-a-percent is an elasticity very much worth tapping! But it's important recognize an omitted variable bias that likely has to do with the "other stuff" in that big gap in figure 1.

Second, MB show that much of the juice in SS's findings come from a period I'm highly obsessed with: the latter 1990s, when true full employment ultimately prevailed, productivity growth was strong, and real pay throughout the wage scale rose quickly, in step with productivity (which is why the correlation drops when you take out that period). This observation does not at all disprove SS's findings, but it did lead me to think about the role of full employment in these dynamics.

I and others have argued that in persistently weak labor markets, employers do not need to uncover efficiency gains to maintain profitability. But in truly tight labor markets, where pressure on labor costs cuts into profit margins, that calculus changes, and in order to avoid higher unit labor costs (comp relative to productivity) and lower unit profits, you've got to find efficiency gains. This is the full-employment-productivity-multiplier about which I've hypothesized.

In this story, the causality goes from full employment to both faster productivity growth and faster wage growth. It also creates non-linearities in the data that complement the missing variable point above. Going for 7 to 6 percent unemployment will do less in the model I'm thinking about here than going from 4 to 3 percent.

BTW, this would imply that the current low unemployment rate should be juicing productivity growth a bit.  In fact, averaging the past six months, year-over-year productivity growth has accelerated to 1.4 percent, a nice bump, though with these jumpy numbers, nothing you'd want to read much into. And, as noted, wage growth still hasn't caught as much of a buzz as I'd like to see.

It will take both faster growth and much more progressive policy to close the big gap in the figure. On that, I suspect we all agree.


We very much appreciate Jared Bernstein's comments on our recent paper on the link between productivity and pay. As he noted, there's a lot more agreement than disagreement between our perspectives: on the importance of raising productivity growth, on the problem of the stagnation of typical workers' pay, as well as on policy approaches towards both of these – all of which he has written about extensively and thoughtfully (e.g. on recent slow wage growth, the productivity slowdownboosting productivity growth, and in his book the Reconnection Agenda).

He raises a couple of points on our paper, on which we'd like to respond.

Jared argues that faster productivity growth is necessary but not by itself sufficient to raise living standards, and that by focusing only on the "necessary" part of this argument we risk straw-manning others' arguments. This certainly hasn't been our intention, so we think it's useful to clarify which mechanisms we were looking to distinguish between with our paper.

We've encountered three broad classes of arguments about the link between productivity and typical pay:

  • Arguments that productivity growth does not affect typical workers' wages at all. Instead other factors determine typical wages – and if workers become more productive, the benefits flow to higher-income workers or capital owners. The implication here would be that no matter what reforms are made to rectify inequality, marginal increases in productivity growth will not translate to increases in typical workers' wages.
  • Arguments that productivity growth is necessary for typical workers' wages to rise, but productivity isn't currently able to affect typical workers' pay because other factors are blocking that transmission mechanism (perhaps factors to do with wage inequality, such as low worker bargaining power). The implication here would be that given the current structure of the economy, marginal increases in productivity growth will not translate to increases in typical workers' wages, but if certain reforms were enacted, productivity growth would once again affect pay. This is how we interpret the argument that productivity growth is necessary but not sufficient to raise pay growth.
  • Arguments that productivity growth exerts pressure to push typical workers' wages upwards, but other orthogonal factors have simultaneously been exerting pressure to pull typical workers' wages downwards. The implication here would be that even given the current structure of the economy and no reforms to address inequality, marginal increases in productivity growth will translate into increases in typical workers' wages.

Few economists subscribe to argument (1). But having seen the chart displaying the divergence between productivity and typical pay since 1973, reasonable economists could differ as to whether they subscribe to argument (2) or argument (3). It's this distinction that motivated our research and the right answer ex ante wasn't clear. Having investigated the question in the postwar and post-73 data, we believe there is much more evidence for argument (3) than argument (2) (or indeed argument (1)). The evidence does more than just make the case that productivity growth is necessary for pay growth: it suggests that higher productivity growth even without changes in policy will have substantial benefits for workers (as Jared notes in his blog). Also of course this conclusion does not make the case against policies to promote fairness or redistribution. It's our view that strategies that focus both on productivity growth and on labor market or redistributive policies are likely to have the greatest impact on living standards for typical workers.

Jared also raised a concern with our econometric strategy, to do with serial correlation. Our baseline estimation method uses moving averages, which mechanically introduces serial correlation into the regression. There's a trade-off here: a moving average strategy is one of the most appropriate ways to get at the low frequency relationship we're interested in, but by its nature introduces serial correlation. On the other hand removing the serial correlation removes much of the useful variation. We settled for the former and used Newey-West HAC standard errors so that our standard errors appropriately accounted for the serial correlation. Our second set of baseline regressions (Table 2 in the paper) use distributed lags, which are more robust to the concern about mechanically containing serial correlation. We used Newey-West HAC standard errors for these regressions in our paper. We've re-done these distributed lag regressions, correcting for serial correlation by assuming an AR(1) error process as Jared suggests. This doesn't change the coefficient estimates importantly.

Whether using these corrections or Jared's serial correlation-corrected estimate of 0.5 for middle-wage workers, the coefficient estimates we end up with are within the ranges that we concluded were most reasonable in our paper: that a one percentage point increase in the rate of productivity growth is associated with a two thirds to one percentage point increase in median compensation growth and a 0.5 to 0.7 percentage point increase in production/nonsupervisory worker compensation growth ("middle-wage" compensation).

Finally Jared discusses the benefits of a high-pressure economy (Larry Mishel and Josh Bivens raised a similar point in their response to our paper). We agree that this is a very important subject, worthy of further investigation – and indeed is closely related to hysteresis issues that one of us has pursued (e.g.  19882014).

Jared closes with the comment that it will take both faster growth and much more progressive policy to close the big gap between productivity and typical pay. And he notes that policy-wise we know more about how to reduce inequality than about how to boost productivity. On both points, we agree.

 -- via my feedly newsfeed

No comments: