What Recovery: Reading Notes

My Roosevelt Institute paper on potential output came out last week. (Summary here.) The paper has gotten some more press since Neil Irwin’s Times piece, including Ryan Cooper in The Week and Felix Salmon in Slate. My favorite headline is from Boing Boing: American Wages Are So Low, the Robots Don’t Want Your Jobs.

In the paper I tried to give a fairly comprehensive overview of the evidence and arguments that the US economy is not in any meaningful sense at potential output or full employment. But of course it was just one small piece of a larger conversation. Here are a few things I’ve found interesting recently on the same set of issues. .

Perhaps the most important new academic contribution to this debate is this paper by Olivier Coibion, Yuriy Gorodnichenko, and Mauricio Ulate, on estimates of potential output, which came out too late for me to mention in the Roosevelt report. Their paper rigorously demonstrates that, despite their production-function veneer, the construction of potential output estimates ensures that any persistent change in growth rates will appear as a change in potential. It follows that there is “little value added in estimates of potential GDP relative to simple measures of statistical trends.” (Matthew Klein puts it more bluntly in an Alphaville post discussing the paper: “‘Potential’ output forecasts are actually worthless.”) The paper proposes an alternative measure of potential output, which they suggest can distinguish between transitory demand shocks and permanent shifts in the economy’s productive capacity. This alternative measure gives a very similar estimate for the output gap as simply looking at the pre-2008 forecasts or extrapolating from the pre-2008 trend.  “Our estimates imply that U.S. output remains almost 10 percentage points below potential output, leaving ample room for policymakers to close the gap through demand-side policies if they so chose to.” Personally, I ‘m a little less convinced by their positive conclusions than by their negative ones. But this paper should definitely put to the rest the idea (as in last year’s notorious CEA-chair letter) that it is obviously wrong — absurd and unserious — that a sufficient stimulus could deliver several years of 4 percent real growth, until GDP returned to its pre-recession trend. It may or may not be true, but it isn’t crazy.

Many of the arguments in my paper were also made in this valuable EPI report by Josh Bivens, reviving the old idea of a “high pressure economy”. Like me, Bivens argues that slow productivity growth is largely  attributable to low investment, which in turn is due to weak demand and slow wage growth, which blunts the incentive for business to invest in labor-saving technology. One important point that Bivens makes that I didn’t, is that much past variation in productivity growth has been transitory; forecasts of future productivity growth based on the past couple of years have consistently performed worse than forecasts based on longer previous periods. So historical evidence gives us no reason see the most recent productivity slowdown as permanent. My one quibble is that he only discusses faster productivity growth and higher inflation as possible outcomes of a demand-driven acceleration in wages. This ignores the third possible effect, redistribution from from profits to wages — in fact a rise in the labor share is impossible without a period of “overfull” employment.

Minneapolis Fed president Neel Kashkari wrote a long post last fall on “diagnosing and treating the slow recovery.” Perhaps the most interesting thing here is that he poses the question at all. There’s a widespread view that once you correct for demographics, the exceptional performance of the late 1990s, etc., there’s nothing particularly slow about this recovery — no problem to diagnose or treat.

Another more recent post by Kashkari focuses on the dangers of forcing the Fed to mechanically follow a Taylor rule for setting interest rates. By his estimate, this would have led to an additional 2.5 million unemployed people this year. It’s a good illustration of the dangers of taking the headline measures of economic performance too literally. I also like its frank acknowledgement that the Fed — like all real world forecasters — rejects rational expectations in the models it uses for policymaking.

Kashkari’s predecessor Narayan Kocherlakota — who seems to agree more with the arguments in my paper — has a couple short but useful posts on his personal blog. The first, from a year ago, is probably the best short summary of the economic debate here that I’ve seen. Perhaps the key analytic point is that following a period of depressed investment, the economy may reach full employment given the existing capital stock while it is still well short of potential. So a period of rapid wage growth would not necessarily mean that the limits of expansionary policy have ben reached, even if those wage gains were fully passed through to higher prices. His emphasis:

Because fiscal policy has been too tight, we have too little public capital. … At the same time, physical investment has been too low… Conditional on these state variables, we might well be close to full employment.  … But, even though we’re close to full employment, there’s a lot of room for super-normal growth. Both capital and TFP are well below their [long run level].  The full-employment growth rate is going to be well above its long-run level for several years.  We can’t conclude the economy is overheating just because it is growing quickly.

His second post focuses on the straightforward but often overlooked point that policy should take into account not just our best estimates but our uncertainty about them, and the relative risks of erring on each side. And if there is even a modest chance that more expansionary policy could permanently raise productivity, then the risks are much greater on the over-contractionary side. [1] In particular, if we are talking about fiscal stimulus, it’s not clear that there are any costs at all. “Crowding out” is normally understood to involve a rise in interest rates and a shift from private investment to public spending. In the current setting, there’s a strong case that higher interest rates  at full employment would be a good thing (at least as long as we still rely on as the main tool of countercyclical policy). And it’s not obvious, to say the least, that the marginal dollar of private investment is more socially useful than many plausible forms of public spending. [2] Kashkari has a post making a similar argument in defense of his minority vote not to raise rates at the most recent FOMC meeting. (Incidentally, FOMC members blogging about their decisions is a trend to be encouraged.)

In a post from March which I missed at the time, Ryan Avent tries to square the circle of job-destroying automation and slow productivity growth. One half of the argument seems clearly right to me: Abundant labor and low wages discourage investment in productivity-raising technologies. As Avent notes, early British and even more American industrialization owe a lot to scarce labor and high wages. The second half of the argument is that labor is abundant today precisely because so much has been displaced by technology. His claim is that “robots taking the jobs” is consistent with low measured productivity growth if the people whose jobs are taken end up in a part of the economy with a much lower output per worker. I’m not sure if this works; this seems like the rare case in economics where an eloquent story would benefit from being re-presented with math.

Along somewhat similar lines, Simon Wren-Lewis points out that unemployment may fall because workers “price themselves into jobs” by accepting lower-wage (and presumably lower-productivity) jobs. But this doesn’t mean that the aggregate demand problem has been solved — instead, we’ve simply replaced open unemployment with what Joan Robinson called “disguised unemployment,” as some of people’s capacity for work continues to go to waste even while they are formally employed. “But there is a danger that central bankers would look at unemployment, … and conclude that we no longer have inadequate aggregate demand…. If demand deficiency is still a problem, this would be a huge and very costly mistake.”

Karl Smith at the Niskanen Center links this debate to the older one over the neutrality of money. Central bank interventions — and aggregate demand in general — are understood to be changes in the flow of money spending in the economy. But a lon-standing tradition in economic theory says that money should be neutral in the long run. As we are look at longer periods, changes in output and employment should depend more and more on real resources and technological capacities, and less and less on spending decisions — in the limit not at all. If you want to know why GDP fell in one quarter but rose in the next (this is something I always tell my undergraduates) you need to ask who chose to reduce their spending in the first period and who chose to increase it in the first. But if you want to know why we are materially richer than our grandparents, it would be silly to say it’s because we choose to spend more money. This is the reason why I’m a bit impatient with people who respond to the fact that, relative to the pre-2008 trend, output today has not recovered from the bottom of the recession, by saying “the trend doesn’t matter, deviations in output are always persistent.” This might be true but it’s a radical claim. It means you either take the real business cycle view that there’s no such thing as aggregate demand, even recessions are due to declines in the economy’s productive potential; or you must accept that in some substantial sense we really are richer than our grandparents because we spend more money. You can’t assert that GDP is not trend-stationary to argue against an output gap today unless you’re ready to accept these larger implications.

The invaluable Tom Walker has a fascinating post going back to even older debates, among 19th century anti-union and pro-union pamphleters, about whether there was a fixed quantity of labor to be performed and whether, in that case, machines were replacing human workers. The back and forth (more forth than back: there seem to be a lot more anti-labor voices in the archives) is fun to read, but what’s the payoff for todays’ debates?

The contemporary relevance of this excursion into the archives is that economic policy and economic thought walks on two legs. Conservative economists hypocritically but strategically embrace both the crowding out arguments for austerity and the projected lump-of-labor fallacy claims against pensions and shorter working time. They are for a “fixed amount” assumption when it suits their objectives and against it when it doesn’t. There is ideological method to their methodological madness. That consistency resolves itself into the “self-evidence” that nothing can be done.

That’s exactly right. When we ask why labor’s share has fallen so much over the past generation, we’re told it’s because of supply and demand — an increased supply of labor from China and elsewhere, and a decreased demand thanks to technology. But if it someone says that it might be a good idea then to limit the supply of labor (by lowering the retirement age, let’s say) and to discourage capital-intensive production, the response is “are you crazy? that will only make everyone poorer, including workers.” Somehow distribution is endogenous when it’s a question of shifts in favor of capital, but becomes exogenously fixed when it’s a question of reversing them.

A number of heterdox writers have identified the claim that productivity growth depends on demand as Verdoorn’s law (or the Kaldor-Verdoorn Law). For example, the Post Keynesian blogger Ramanan mentions it here and here. I admit I’m a bit dissatisfied with this “law”. It’s regularly asserted by heterodox people but you’ll scour our literature in vain looking for either a systematic account of how it is supposed to operate or quantitative evidence of how and how much (or whether) it does.

Adam Ozimek argues that the recent rise in employment should be seen as an argument for continued expansionary policy, not a shift away from it. After all, a few years ago many policymakers believed such a rise was impossible, since the decline in employment was supposed to be almost entirely structural.

Finally, Reihan Salam wants to enlist me for the socialist flank of a genuinely populist Trumpism. This is the flipside of criticism I’ve sometimes gotten for making this argument — doesn’t it just provide intellectual ammunition for the Bannon wing of the administration and its calls for vast infrastructure spending,  which is also supposed to boost demand and generate much faster growth? Personally I think you need to make the arguments for what you think is true regardless of their political valence. But I might worry about this more if I believed there was even a slight chance that Trump might try to deliver for his working-class supporters.


[1] Kocherlakota talks about total factor productivity. I prefer to focus on labor productivity because it is based on directly observable quantities, whereas TFP depends on estimates not only of the capital stock but of various unobservable parameters. The logic of the argument is the same either way.

[2] I made similar arguments here.


EDIT: My comments on the heterodox literature on the Kaldor-Verdoorn Law were too harsh. I do feel this set of ideas is underdeveloped, but there is more there than my original post implied. I will try to do a proper post on this work at some point.

Can We Blame Low Labor Participation on Past High Unemployment?

Fifth post in a series. Posts onetwothree and four.

We know that US GDP fell sharply in 2008-2009. We know that none of that decline has been made up by faster growth since the recession: GDP today is about 14 percent below the pre-2008 trend, a gap that shows no sign of closing. We also know that one-third of that shortfall is accounted for by slower productivity growth, and the remaining two-thirds by slower employment growth.

To put numbers on it: Over the past decade, US employment rose by a total of 6 percent, or about 0.5 percent per year. This is about half the rate of employment growth over the last ten years before the recession, and less a quarter the average rate for the postwar period as a whole. 2000-2010 was the first decade since the Depression in which US employment actually fell. Since the unemployment rate today is very close to that of ten years ago, this whole slowdown is accounted for by a decline in laborforce participation.

Employment growth, unlike productivity growth, was already slowing prior to the recession, and  pre-recession forecasts predicted a further slowdown comparable to what actually occurred. This is consistent with a widely-held view that the slowdown in employment is the result of demographic and other structural factors, not of the recession or demand weakness in general. In the next couple posts, I want to take a critical look at this claim. How confident should we be that employment would be the same today in a counterfactual world where the 2008-2009 didn’t happen? How responsive might employment be to stronger demand going forward? And more broadly, how much do changes in laborforce participation seem to be explained by more or less exogenous factors like demographics, and how much by demand and labor-market conditions?

The rest of this post is about an approach to this question that did not produce the results I was hoping for. So I probably won’t include this material in whatever paper comes out of these posts. But as we feel our way into reality it’s important to note down the dead ends as well as the routes that seem promising. And even though this exercise didn’t help much in answering the big questions posed in the previous paragraph, it’s still interesting in its own right.


Can the fall in laborforce participation be explained as a direct, predictable effect of the rise in unemployment during the recession? It seems like maybe it can. The starting point is the observation that unemployed workers are much more likely to drop out of the laborforce than people with jobs are. You can see this clearly in the BLS tables on employment transitions. As the figure below shows, about 3 percent of employed people exit the laborforce each month, a fraction that has been remarkably stable since the data begins in 1990. Meanwhile, about 20 percent of unemployed people drop out of the laborforce each month.


On the face of it, this 17-point difference suggests an important role for the unemployment rate in changes in labor force participation. All else equal, each year-point of additional unemployment should reduce the labor-force participation rate by two points. (0.17 x 12 = 2.) So you would think that much of the recent fall in laborforce participation could be explained simply by the rise in unemployment during the recession.

When I thought of this it seemed very logical. It would be easy to do a counterfactual exercise, I thought, showing how laborforce participation would have evolved simply based on the historical transition rates between employment, unemployment and out of the laborforce, and the actual evolution of employment and unemployment. If you could show that something like the actual fall in laborforce participation was a predictable result of the rise in unemployment during the recession, that would support the idea that demand rather than “structural” factors are at work. And even if it wasn’t that strong positive evidence, it would suggest skepticism about similar counterfactual exercises using historical participation rates by age and so on.

I mean, it makes sense, right? Unemployed people are much more likely to leave the workforce than employed people, so a rise in unemployment should naturally lead to a decline in laborforce participation. But as the figure below shows, the numbers don’t work.

What I did was start with the populations of employe, unemployed and not-in-the-laborforce people at the end of the recession in December 2009. Then I created a counterfactual scenario for the remaining period using the actual transition rates between employment and unemployment but the pre-recession average rates for transitions between not in the workforce and unemployment and employment. In other words, just knowing the average rates that people move between employment, unemployment and out of the workforce, and the actual shifts between employment and unemployment, what path would you have predicted for laborforce participation over 2010-2016?

transitions2The heavy gray line shows the historical fraction of the population aged 16 and over who are not in the laborforce. The black line shows the results of the counterfactual exercise. Not very close.

There turn out to be two reasons why the counterfactual exercise gives such a poor fit. Both are interesting and neither was obvious before doing the exercise. The first reason is that there are  surprisingly large flows from out of the labor force back into it. Per the BLS, about 7 percent of people who report being out of the labor force in a given month are either employed or unemployed (i.e. actively seeking work) the following month. This implies that the typical duration of being out of the workforce is less than a year — though of course this is a mix of people who leave the workforce for just a month or two and people who leave for good. For present purposes, the important thing is that exogenous changes to the employment-population ratio decline quickly, with a half-life of only about a year. So while the historical data suggests that a rise in unemployment like we saw in 2008-2009 should have been associated with a large rise in the share of the population not in the laborforce, it also suggests that this effect should have been transitory — a couple years after unemployment rates returned to normal, participation rates should have as well. This is not what we’ve seen.

The large gross movements in and out of the laborforce mean that sustained lower participation rates can’t be straightforwardly understood as the “echo” of high unemployment in the past. But they do also tend to undermine the structural story — if the typical stint outside the laborforce lasts less than a year it can hardly be due to something immutable.

The second reason why the counterfactual doesn’t fit the data was even more surprising, at least to me. I constructed my series using the historical average transition rates into and out of the workforce. But transition rates during the recession and early recovery departed from the historical average in an important way: unemployed workers were significantly less likely to exit the workforce. This turns out to be the normal pattern, at least over the previous two business cycles — if you look back to that first figure, you can see dips in the transition rate from unemployed to out of the workforce in the early 1990s and early 2000s downturns as well. The relationship is clearer in the next figure, a scatter of the unemployment rate and the share of unemployed workers leaving the workforce each month.



As you can see, there is a strong negative relationship — when unemployment was around 4 percent in 1999-2000 and again in 2006-2007, about a quarter of the unemployed exited the laborforce each month. But at the peak of the past recession when unemployment reached 10 percent, only 18 percent of the unemployed left the laborforce each month. That might not seem like a huge difference, but it’s enough to produce quite different dynamics. It’s also a bit surprising, since you would think that people would be more likely to give up searching for work when unemployment is high than when when it is low. The obvious explanation would be that the people who are out of work when the unemployment rate is low are not simply a smaller set of the same people who are out of work when the rate is high, but are different in some way. The same factors that keep them at the back of the hiring queue may make also be likely to push them out of the laborforce altogether. Extended unemployment insurance might also play a role.

It would be possible to explore this further using CPS data, which is the source for the BLS tables I’m working with. No doubt there are papers out there describing the different characteristics of the unemployed in periods of high versus low unemployment. (Not being a labor economist, I don’t know this literature.) But I am going to leave it here.

Summary: The fact that unemployed people are much more likely to leave the laborforce than employed people are, suggests that some part of the fall in laborforce particiaption since 2008 might be explained by the lingering effects of high unemployment in the recession and early recovery. But this story turns out not tow work, for two reasons. First, the rapid turnover of the not in the laborforce population means that this direct effect of high unemployment on participation is fairly shortlived. Second, the rate at which unemployed people exit the laborforce turns out to be lower when unemployment is high. Together, these two factors produce the results shown in the second figure — the fall in participation you would predict based simply on high unemployment is steeper but shorter-lived than what actually occurred. The first factor — the large flows in and out of the laborforce — while it vitiates the simple story I proposed here, is consistent with a broader focus on demand rather than demographics as an explanation for slow employment growth. If people are frequently moving in and out of the laborforce, it’s likely that their decisions are influenced by their employment prospects, and it means they’re not determined by fixed characteristics like age. The second factor — that unemployed people were less likely to give up looking for jobs during 2009-2011, as in previous periods of high unemployment — is, to me, more surprising, and harder to fit into a demand-side story.

Employment, Productivity and the Business Cycle

Fourth post in a series. Posts one, two and three.

Empirically-oriented macroeconomists have recognized since the early 20th century that output, employment and productivity move together over the business cycle. The fact that productivity falls during recessions means that employment varies less over the cycle than output does. This behavior is quite stable over time, giving rise to Okun’s law. In the US, Okun’s law says that the unemployment rate will rise by one point in each 2.5 point shortfall of GDP growth over trend — a ratio that doesn’t seem to have declined much since Arthur Okun first described it in the mid-1960s. [1]

It’s not obvious that potential should show this procyclical behavior. As I noted in the previous post, a naive prediction from a production function framework would be that a negative demand shock should reduce employment more than output, since business can lay off workers immediately but can’t reduce their capital stock right away. In other words, productivity should rise in recessions, since the labor of each still-employed worker is being combined with more capital.

There are various explanations for why labor productivity behaves procyclically instead. The most common focus on the transition costs of changing employment. Since hiring and firing is costly for businesses, they don’t adjust their laborforce to every change in demand. So when sales fall in recessions, they will keep extra workers on payroll — paying them now is cheaper than hiring them back later. Similarly, when sales rise businesses will initially try to get more work out of their existing employees. This shows up as rising labor productivity, and as the repeated phenomenon of “jobless recoveries.”

Understood in these terms, the positive relationship between output, employment and productivity should be a strictly short-term phenomenon. If a change in demand (or in other constraints on output) is sustained, we’d expect labor to fully adjust to the new level of production sooner or later. So over horizons of more than a year or two, we’d expect output and employment to change in proportion. If there are other limits on production (such as non-produced inputs like land) we’d expect output and labor productivity to move inversely, with faster productivity growth associated with slower employment growth or vice versa. (This is the logic of “robots taking the jobs.”) A short-term positive, medium term negative, long-term flat or negative relationship between employment growth and productivity growth is one of the main predictions that comes out of a production function. But it doesn’t require one. You can get there lots of other ways too.

And in fact, it is what we see.

prod-emp correl

The figure shows the simple correlation of employment growth and productivity growth over various periods, from one quarter out to 50 quarters. (This is based on postwar US data.) As you can see, over periods of a year or less, the correlation is (weakly) positive. Six-month periods in which employment growth was unusually weak are somewhat more likely to have seen weak productivity growth as well. This is the cyclical effect presumably due to transition costs — employers don’t always hire or fire in response to short-run changes in demand, allowing productivity to vary instead. But if sales remain high or low for an extended period, employers will eventually bring their laborforce into line, eliminating this relationship. And over longer periods, autonomous variation in productivity and labor supply are more important. Both of these tend to produce a negative relationship between employment and productivity. And that’s exactly what we see — a ten-year period in which productivity grew unusually quickly is likely to be one in which employment grew slowly. (Admittedly postwar US data doesn’t give you that many ten-year periods to look at.)

Another way of doing this is to plot an “Okun coefficient” for each horizon. Here we are looking at the relationship between changes in employment and output. Okun’s law is usually expressed in terms of the relatiojship between unemployment and output, but here we will look at it in terms of employment instead. We write

(1)    %ΔE = a (g – c)

where %ΔE is the percentage change in employment, g is the percentage growth in GDP, is a constant (the long-run average rate of productivity growth) and a is the Okun coefficient. The value of a says how much additional growth in employment we’d expect from a one percentage-point increase in GDP growth over the given period. When the equation is estimated in terms of unemployment and the period is one, year, a is generally on the order of 0.4 or so, meaning that to reduce unemployment by one point over a year normally requires GDP growth around 2.5 points above trend. We’d expect the coefficient for employment to be greater, but over short periods at least it should still be less than one.

Here is what we see if the estimate the equation for changes in output and employment for various periods, again ranging from one quarter up to ten years. (Again, postwar US data. The circles are the point estimates of the coefficients; the dotted lines are two standard errors above and below, corresponding to a standard 95% confidence interval.)

emp on output

What’s this show? If we estimate Equation (1) looking at changes over one quarter, we find that one percentage point of additional GDP growth is associated with just half a point of additional employment growth. But if we estimate the same equation looking at changes over two years, we find that one point of additional GDP growth is associated with 0.75 points of additional employment growth.

The fact that the coefficient is smallest for the shorter periods is, again, consistent witht he conventional understanding of Okun’s law. Because hiring and firing is costly, employers don’t fully adjust staffing unless a change in sales is sustained for a while. If you were thinking in terms of a production function, the peak around 2 years represents a “medium-term” position where labor has adjusted to a change in demand but the capital stock has not.

While it’s not really relevant for current purposes, it’s interesting that at every horizon the coefficient is significantly below zero. What this tells us is that there is no actual time interval corresponding to the “long run” of the model– a period long enough for labor and the capital stock to be fully adjusted but short enough that technology is fixed. Over this hypothetical long run, the coefficient would be one. One way to think about the fact that the estimated coefficients are always smaller, is that any period long enough for labor to adjust, is already long enough to see noticeable autonomous changes in productivity. [2]

But what we’re interested in right now is not this normal pattern. We’re interested in how dramatically the post-2008 period has departed from it. The past eight years have seen close to the slowest employment growth of the postwar period and close to the slowest productivity growth. It is normal for employment and productivity to move together for a couple quarters or a year, but very unusual for this joint movement to be sustained over nearly a decade. In the postwar US, at least, periods of slow employment growth are much more often periods of rapid productivity growth, and conversely. Here’s a regression similar to the Okun one, but this time relating productivity growth to employment growth, and using only data through 2008.

prod on empWhile the significance lines can’t be taken literally given that these are overlapping periods, the figure makes clear that between 1947 and 2008, there were very few sustained periods in which both employment and productivity growth made large departures from trend in the same direction.

Put it another way: The past decade has seen exceptionally slow growth in employment — about 5 percent over the full period. If you looked at the US postwar data, you would predict with a fair degree of confidence that a period of such slow employment growth would see above-average productivity growth. But in fact, the past decade has also seen very low productivity growth. The relation between the two variables has been much closer to what we would predict by extrapolating their relationships over periods of a year. In that sense, the current slowdown resembles an extended recession more than it does previous periods of slower growth.

As I suggested in an earlier post, I think this is a bigger analytic problem than it might seem at first glance.

In the conventional story, productivity is supposed to be driven by technology, so a slowdown in productivity growth reflects a decline in innovation and so on. Employment is driven by demographics, so slower employment growth reflects aging and small families. Both of these developments are negative shifts in aggregate supply. So they should be inflationary — if the economy’s productive potential declines then the same growth in demand will instead lead to higher prices. To maintain stable prices in the face of these two negative supply shocks, a central bank would have to raise interest rates in order to reduce aggregate spending to the new, lower level of potential output. Is this what we have seen? No, of course not. We have seen declining inflation even as interest rates are at historically low levels. So even if you explain slower productivity growth by technology and explain slower employment growth by demographics, you still need to postulate some large additional negative shift in demand. This is DeLong and Summers’ “elementary signal identification point.”

Given that we are postulating a large, sustained fall in demand in any case, it would be more parsimonious if the demand shortfall also explained the slowdown in employment and productivity growth. I think there are good reasons to believe this is the case. Those will be the subject of the remaining posts in this series.

In the meantime, let’s pull together the historical evidence on output, employment and productivity growth in one last figure. Here, the horizontal axis is the ten-year percentage change in employment, while the vertical axis is the ten-year percentage change in productivity. The years are final year of the comparison. (In order to include the most recent data, we are comparing first quarters to first quarters.) The color of the text shows average inflation over the ten year period, with yellow highest and blue lowest. The diagonal line corresponds to the average real growth rate of GDP over the full period.

e-p scatter

What we’re looking at here is the percentage change in productivity, employment and prices over every ten-year period from 1947-1957 through 2006-2016. So for instance, growth between 1990 and 2000 is represented by the point labeled “2000.” During this decade, total employment rose by about 20 percent while productivity rose by a total of 15 percent, implying an annual real growth of 3.3 percent, very close to the long-run average.

One natural way to think about this is that yellow points below and to the right of the line suggest negative supply shocks: If the productive capacity of the economy declines for some reason, output growth will slow, and prices will rise as private actors — abetted by a slow-to-react central bank — attempt to increase spending at the usual rate. Similarly, blue points above the line suggest positive supply shocks. Yellow points above the line suggest positive demand shocks — an increase in spending can increase output growth above trend, at least for a while, but will pull up prices as well. And blue points below the line suggest negative demand shocks. This, again, is Delong and Summers’ “elementary signal identification point.”

We immediately see what an outlier the recent period is. Both employment and productivity growth over the past ten years have been drastically slower than over the preceding decade — about 5 percent each, down from about 20 percent. 2000-2010 and 2001-2011 were the only ten-year periods in postwar US history when total employment actually declined. The abruptness of the deceleration on both dimensions is a challenge for views that slower growth is the result of deep structural forces. And the combination of the slowdown in output growth with falling prices — especially given ultra-low interest rats — strongly suggests that we’ve seen a negative shift in desired spending (demand) rather than in the economy’s productive capacities (supply).

Another way of looking at this is as three different regimes. In the middle is what we might call “the main sequence” — here there is steady growth in demand, met by varying mixes of employment and productivity growth. On the upper right is what gets called a “high-pressure economy,” in which low unemployment and strong demand draw more people into employment and facilitates the reallocation of labor and other resources toward more productive activity, but put upward pressure on prices. On the lower left is stagnation, where weak demand discourages participation in the labor force and reduces productivity growth by holding back investment, new business formation and by leaving a larger number of those with jobs underemployed, and persistent slack leads to downward pressure on prices (though so far not outright deflation). In other words, macroeconomically speaking the past decade has been a sort of anti-1960s.


[1] There are actually two versions of Okun’s law, one relating the change in the unemployment rate to GDP growth and one relating the level of unemployment to the deviation of GDP from potential. The two forms will be equivalent if potential grows at a constant rate.

[2] The assumption that variables can be partitioned into “fast” and “slow” ones, so that we can calculate equilibrium values of the former with the latter treated as exogenous, is a very widespread feature of economic modeling, heterodox as much as mainstream. I think it needs to be looked at more critically. One alternative is dynamic models where we focus on the system’s evolution over time rather than equilibrium conditions. This is, I suppose, the kind of “theory” implied by VAR-type forecasting models, but it’s rare to see it developed explicitly. There are people who talk about a system dynamics approach, which seems promising, but I don’t know much about them.

Links for July 27, 2016

Labor dynamism and demand. My colleagues Mike Konczal and Marshall Steinbaum have an important new paper out on  the decline in new business starts and in labor mobility. They argue that the data don’t support a story where declining labor-market dynamism is the result of supply-side factors  like occupational licensing. It looks much  more like the result of chronically weak demand for labor, which for whatever reason is not picked up by the conventional unemployment rate.  This is obviously relevant to the potential output question I’m interested in — a slowdown in the rate at which workers move to new firms is a natural channel by which weak demand could reduce labor productivity. It’s also a very interesting story in its own right.

Konczal and Steinbaum:

The decline of entrepreneurship and “business dynamism” has become an accepted fact … Explanations for these trends … broadly fall on the supply side: that increasingly onerous occupational licensing impedes entry into certain protected professions and restricts licensed workers to staying where they are; that the high cost of housing thanks to restrictions on development hampers individuals from moving… But we find that the data reject these supply-side explanations: If there were increased restrictions on changing jobs or starting a business, we would expect those few workers and entrepreneurs who do manage to move to enjoy increased wage gains relative to periods with higher worker flows, and we would expect aggressive hiring by employers with vacancies. … Instead, we see the opposite…

We propose a different organizing principle: Declining business dynamism and labor mobility are features of a slackening labor market … workers lucky enough to have formal employment stay where they are rather than striking out as entrepreneurs …

Also in Roosevelt news, here’s a flattering piece about us in the New York Times Magazine.


John Kenneth who? Real World Economics Review polled its subscribers on the most important economics books of the past 100 years. Here’s the top ten. Personally I suspect Debt will have more staying power than Capital in the 21st Century, and I think Minsky’s book John Maynard Keynes is a better statement of his vision than Stabilizing an Unstable Economy, a lot of which is focused on banking-sector developments of the 1970s and 1980s that aren’t of much interest today. But overall it’s a pretty good list. The only one I haven’t read is The Affluent Society. I wonder if anyone under the age of 50 picked that one?


Deflating the elephant. Here is a nice catch from David Rosnick. Brank Milanovic has a well-known graph of changes in global income distribution over 1988-2008. What we see is that, while within most countries there has been increased polarization, at the global level the picture is more complicated. Yes, the top of the distribution has gone way up, and the very bottom has gone down. But the big fall has been in the upper-middle of the distribution — between the 80th and 99th percentiles — while most of the lower part has has risen, with the biggest gains coming around the 50th percentile. The decline near the high end is presumably working-class people in rich countries and most people in the former Soviet block —who were still near the top of the global distribution in 1988. A big part of the rise in the lower half is China. A natural question is, how much? — what would the distribution look like without China? Milanovic had suggested that the overall picture is still basically the same. But as Rosnick shows, this isn’t true — if you exclude China, the gains in the lower half are much smaller, and incomes over nearly half the distribution are lower in 2008 than 20 years before. It’s hard to see this as anything but a profoundly negative verdict on the Washington Consensus that has ruled the world over the past generation.


By the way, you cannot interpret this — as I at first wrongly did — as meaning that 40 percent of the world’s people have lower incomes than in 1988. It’s less than that. Faster population growth in poor countries would tend to shift the distribution downward even if every individual’s income was rising.


Does nuclear math add up? Over at Crooked Timber, there’s been an interesting comments-thread debate between Will Boisvert (known around here for his vigorous defense of nuclear power) and various nuke antis and skeptics. I’m the farthest thing from an expert, I can’t claim to be any kind of arbiter. But personally my sympathies are with Will. One important thing he brings out, which I hadn’t thought about enough until now, is the difference between electricity and most other commodities. Part of the problem is the very large share of fixed costs — as the Crotty-Minsky-Perelman strain of Keynesians have emphasized, capitalism does badly with long lived capital assets. A more distinctive problem is the time dimension — electricity produced at one time is not a good substitute for electricity produced at a different time, even just an hour before or after. Electricity cannot be stored economically at a meaningful scale, nor — given that almost everything in modern civilization uses it — can its consumption be easily shifted in time.  This means that straightforward comparisons of cost per kilowatt — hard enough to produce, given the predominance of fixed  costs — can be misleading. Regardless of costs, intermittent sources — like wind or solar — have to be balanced by sources that can be turned on anytime — which in the absence of nuclear, means fossil fuels.

Do you believe, as I do, that climate change is the great challenge facing humanity in the next generation? Then this is a very strong argument for nuclear power. Whatever its downsides, they are not as bad as boiling the oceans. Still, it’s not a decisive argument. The big other questions are the costs of power storage and of more extensive transmission networks — since when the sun isn’t shining and the wind isn’t blowing in one place, they probably are somewhere else. (I agree with Will that using the price mechanism to force electricity usage to conform to supply from renewables is definitely the wrong answer.) The CT debate doesn’t answer those questions. But it’s still an example of how informative blog debate can be when there are people  both sides with real expertise who are prepared to engage seriously with each other.


On other blogs, other wonders. Here is a fascinating post by Laura Tanenbaum on the end of sex-segregated job ads and the false dichotomy between “elite” and “grassroots”  feminism.

This very interesting article by Jose Azar on the extent and economic significance of common ownership of corporate shares deserves a post of its own.

Here’s a nice little think piece from Bloomberg wondering what, if anything, is meant by “the natural rate” of interest. I’m glad to see some skepticism about this concept in the larger conversation. In my mind, the “natural rate” is one of the key patches covering over the disconnect between economic theory and the observable economy.

Bhenn Bhiorach has a funny post on the lengths people will go to to claim that low inflation is really high inflation.

Trump’s Tariffs: A Dissent

Last week, the Washington Post ran an article by Jim Tankersley on what would happen if Trump got his way and the US imposed steep tariffs on goods from Mexico and China. I ended up as the objectively pro-Trump voice in the piece. The core of it was an estimates from Mark Zandi at Moody’s that a 45% tariff on goods from China and a 35% tariff on goods from Mexico (I don’t know where these exact numbers came from) would have an effect on the US comparable to the Great Recession, with output and employment both falling by about 5 percent relative to the baseline. About half this 5 percent fall in GDP would be due to retaliatory tariffs from China and Mexcio, and about half would come from the US tariffs themselves. As I told the Post, I think this is nuts.

Let me explain why I think that, and what a more realistic estimate would look like. But first, I should say that Tankersley did exactly what one ought to do with this story — asked the right question and went to a respected expert to help him answer it. The problem is with what that expert said, not the reporting. I should also say that my criticisms were presented clearly and accurately in the piece. But of course, there’s only so much you can say in even a generous quote in a newspaper article. Hence this post.

I haven’t seen the Moody’s analysis (it’s proprietary). All I know is what’s in the article, and the general explanation that Tankersley gave me in the interview. But from what I can tell, Zandi and his team want to tell a story like this. When the US imposes a tariff, it boosts the price of imported goods but leads to no substitution away from them. Instead, higher import prices just mean lower real incomes in the US. Then, when China and Mexico retaliate, that does lead to substitution away from US goods, and the lost exports reduce US real incomes further. But only under the most extreme assumptions can you get Zandi’s numbers out of this story.

While this kind of forecasting might seem mysterious, it mostly comes down to picking values for a few parameters — that is, making numerical guesses about relationships between the variables of interest. In this case, we have to answer three questions. The first question is, how much of the tariff is paid by the purchasers of imported goods, as opposed to the producers? The second question is, how do purchasers respond to higher prices — by substituting to domestic goods, by substituting to imports from other countries, or by simply paying the higher prices? Substitution to domestic goods is expansionary (boosts demand here), substitution to imports from elsewhere is neutral, and paying the higher prices is contractionary, since it reduces the income available for domestic spending. And the third question is, how much does a given shift in demand ultimately move GDP? The answer to the first question gives us the passthrough parameter. The answer to the second question gives us two price elasticities — a bilateral elasticity for imports from that one country, and an overall elasticity for total imports. The answer to the third question gives us the multiplier. Combine these and you have the change in GDP resulting from the tariff. Of course if you think the initial tariffs will provoke retaliatory tariffs from the other countries, you have to do the same exercise for those, with perhaps different parameters.

Let’s walk through this. Suppose the US — or any country — increases taxes on imports: What can happen? The first question is, how is the price of the imported good set — by costs in the producing country, or by market conditions in the destination? If conditions in the destination country affect price — if the producer is unable or unwilling  to raise prices by the full amount of the tariff — then they will have to accept lower revenue per unit sold. This is referred to as pricing to market or incomplete passthrough, and empirical studies suggest it is quite important in import prices, especially in the US. Incomplete passthrough may result in changing profit margins for producers, or they may be able to adjust their own costs — wages especially — in turn. Where trade is a large fraction of GDP, some of the tax may eventually be translated into a lower price level in the exporting country.

Under floating exchange rates, the tariff may also lead a depreciation of the exporting country currency relative to the currency of the country imposing the tariff. This is especially likely where trade between the two countries is a large share of total trade for one or both of them. In this case, a tariff is more likely to cause a depreciation of the Mexican peso than of the Chinese renminbi, since the US accounts for a higher fraction of Mexico’s exports than of China’s, and the renminbi is actively managed by China’s central bank.

Taking all these effects together, passthrough for US imports is probably less than 0.5. In other words, the  majority of a tariff’s impact will probably be on dollar revenue for producers, rather than dollar costs for consumers. So a 10 percent tariff increases costs of imported goods by something less than 5 percent and reduces the revenue per unit of producers by something more than 5 percent.

The fraction of the tax that is not absorbed by lower exporter profit margins, lower wages in the export industry or a lower price level in the exporting country, or by exchange rate changes, will be reflected in higher prices in the importing country. The majority of trade goods for the US (as for most countries) are intermediate and capital goods, and even imported consumption goods are almost never purchased directly by the final consumer. So on the importing side, too, there will be firms making a choice between accepting lower profit margins, reducing wages and other domestic costs, or raising prices. Depending on exactly where we are measuring import prices, this might further reduce passthrough.

Let’s ignore this last complication and assume that a tax that is not absorbed on the exporting-country side is fully passed on to final price of imported goods. Purchasers of imported goods now respond to the higher price either by substituting to domestic goods, or substituting to imported goods from some third country not subject to the tax, or continuing to purchase the imports at the higher price. To the extent they substitute to domestic goods, that will boost demand here; to the extent they substitute to third-country goods, the tax will have no effect here.

These rates of substitution are described by the price elasticity of imports, computed as the ratio of the percentage change in the price, to the resulting percentage change in the quantity imported. So for instance if we thought that a 1 percent increase in the price of imported goods leads to a 2 percent fall in the quantity purchased, we would say the price elasticity is 2. There are two elasticities we have to think about — the bilateral elasticity and the overall elasticity. For example, we might think that the bilateral elasticity for US imports from China was 3 while the overall price elasticity for was 1. In that case, a 1 percent increase in the price of Chinese imports would lead to a 3 percent fall in US imports from China but only one-third of that would be through lower total US imports; the rest would be through higher imports from third countries.

To the extent the higher priced imported goods are purchased, this may result in a higher price of domestic goods for which the imports are an input or a substitute; to the extent this happens, the tax will raise domestic inflation but leave real income unchanged. For the US, import prices have a relatively small effect on overall inflation, so we’ll ignore this effect here. If we included it, we would end up with a smaller effect.

To the extent that the increase in import prices neither leads to any substitution away from the imported goods, nor to any price increase in domestic goods, it will reduce real incomes in the importing country, and leave incomes in the exporting country unchanged. Conversely, to the extent that the tariff is absorbed by lower wages or profit margins in the exporting country, or leads to substitution away from that country’s goods, it reduces incomes in the exporting country, but not in the importing country. And of course, to the extent that there is no substitution away from the taxed goods, government revenue will increase. Zandi does not appear to have explicitly modeled this last effect, but it is important in thinking about the results — a point I’ll return to.

Whether the increase in import prices increases domestic incomes (by leading to substitution to domestic goods) or reduces them, the initial effect will be compounded as the change in income leads to changes in other spending flows. If, let’s say, an increase in the price of Chinese consumer goods forces Americans to cut back purchases of American-made goods, then the workers and business owners in the affected industries will find themselves with less income, which will cause them to reduce their spending in turn. This is the familiar multiplier. The direct effect may be compounded or mitigated by financial effects — the multiplier will be larger if you think (as Zandi apparently does) that a fall in income will be accompanied by a fall in asset prices with a further negative effect on credit and consumption, and smaller if you think that a trade-induced change in income will be offset by a change in monetary (or fiscal) policy. In the case where central bank’s interest rate policy is always able to hold output at potential, the multiplier will be zero — shocks to demand will have no effect on output. This extreme case looked more reasonable a decade ago than it does today. In conditions where the Fed can’t or won’t offset demand impacts, estimates of the US multiplier range as high as 2.5; a respectable middle-of-the-road estimate would be 1.5.

Let’s try this with actual numbers.

Start with passthrough. The overwhelming consensus in the empirical literature is that less than half of even persistent changes in exchange rates are passed through to US import prices. This recent survey from the New York Fed, for instance, reports a passthrough of about 0.3:

following a 10 percent depreciation of the dollar, U.S. import prices increase about 1 percentage point in the contemporaneous quarter and an additional 2 percentage points over the next year, with little if any subsequent increases.

The factors that lead to incomplete passthrough of exchange rate movements — such as the size of the US market, and the importance exporters of maintaining market share — generally apply to a tariff as well, so it’s reasonable to think passthrough would be similar. So a 45% tariff on Chinese goods would probably raise prices to American purchasers by only about 15%, with the remainder absorbed by profits and/or wages at Chinese exporters.

Next we need to ask about the effect of that price on American purchases. There is a large literature estimating trade price elasticities; a sample is shown in the table below. As you can see, almost all the import price elasticities are between 0.2 and 1.0. (Price elasticities seem to be greater for US exports than for imports; they also seem to be higher for most other countries than for the US.) The median estimates is around 0.5 for overall US imports. Country-specific estimates are harder to find but I’ve seen values around 1.0 for US imports from both China and Mexico. Using those estimates, we would expect a 15% increase in the price of Chinese imports to lead to a 15% fall in imports from China, with about half of the substitution going to US goods and half going to imports from other countries. Similarly, a 10% increase in the price of goods from Mexico  (a 35% tariff times passthrough of 0.3) would lead to a 10% fall in imports from Mexico, with half of that being a switch to US goods and half to imports from elsewhere.

Screen Shot 2016-03-23 at 4.31.18 PM
Selected trade elasticity estimates for the US. The last column indicates if “price” was measured with an import price index (P), the exchange rate (E), or competitiveness, i.e. relative wages (C). The “P” estimates are most relevant for a tariff.

Finally, we ask how the combination of substitution away from imports from Mexico and China, and the rise in price of the remaining imports, would affect US output. US imports from China are about 2.7 percent of US GDP, and imports from Mexico are about 1.7 percent of GDP. So with the parameters above, substitution to US goods raises GDP by 7.5% x 2.7% (China) plus 5% x 1.7% (Mexico), or 0.29% of GDP. Meanwhile the higher prices of the remaining imports from China and Mexico reduce US incomes by 0.22 percent, for a net impact of a trivial one twentieth of one percent of GDP. Apply a standard multiplier of 1.5, and the tariffs boost GDP by 0.08 percent.

You could certainly get a larger number than this, for instance if you thought that passthrough of a tariff would be substantially greater than passthrough of exchange rate changes. And making US import demand just a bit less price-elastic is enough to  turn the small positive impact into a small negative one. But it would be very hard to get an impact of even one percent of GDP in either direction. And it would be almost impossible to get a negative impact of the kind that Zandi describes. If you assume both that the tariffs are fully passed through to final purchasers, and that US import demand is completely insensitive to price then with a multiplier of 1.5, you get a 2.7 percent reduction in US GDP. Since this is close to Zandi’s number, this may be what he did. But again, these are extreme assumptions, with no basis in the empirical literature. That doesn’t mean you can’t use them, but you need to justify them; just saying the magic word “proprietary” is not enough. (Imagine all the trouble Jerry Friedman could have saved himself with that trick!)

And the very low price elasticity you need for this result has some funny implications. For instance, it implies that when China intervenes to weaken their currency, they are just impoverishing themselves, since — if demand is really price-inelastic — they are now sending us the same amount of goods and getting fewer dollars for each one. I doubt Zandi would endorse this view, but it’s a logical corollary of the ultra-low elasticity he needs to get a big cost to the US from the initial tariff. Note also that the low-elasticity assumption means that the tariff creates no costs for China or Mexico: their exporters pass the increased tariff on completely to US consumers, and lose no sales as a result. It’s not clear why they would “retaliate” for this.

Let’s assume, though, that China and Mexico do impose tariffs on US goods. US exports to China and Mexico equal 0.7 and 1.3 percent of US GDP respectively. Passthrough is probably higher for US exports — let’s say 0.6 rather than 0.3. Price elasticity is also probably higher — we’ll say 1.5 for both bilateral elasticities and for overall export elasticity. (In the absence of exchange-rate changes, there’s no reason to think that a fall in exports to China and Mexico will lead to a rise in exports to third countries.) And again, we’ll use a multiplier of 1.5. This yields a fall in US GDP from the countertariffs of just a hair under 1 percent. Combine that with the small demand boost from the tariff, and we get an overall impact of -0.9 percent of GDP.

I admit, this is a somewhat larger hit than I expected before I worked through this exercise. But it’s still much smaller than Zandi’s number.

My preferred back-of-the-envelope for the combined impact of the tariffs and countertariffs would be a reduction in US GDP of 0.9 percent, but I’m not wedded to this exact number. I think reasonable parameters could get you an impact on US GDP anywhere from positive 1 percent to, at the worst, negative 2 percent or so. But it’s very hard to get Zandi’s negative 5 percent. You need an extremely high passthrough for both import and export prices, plus extremely price-inelastic US import demand and extremely price-elastic demand for US exports — all three parameters well outside the range in the empirical literature.  At one point a few years ago, I collected about 20 empirical estimates of US trade elasticities, and none of them had a price elasticity for US exports greater than 1.5. But even with 100% passthrough, and a generous multiplier of 2.0, you need an export price elasticity of 4 or so to get US GDP to fall by 5 points.

Still, while Zandi’s 5 percent hit to GDP seems beyond the realm of the plausible, one could perhaps defend a still-substantial 2 percent. Let’s think for a moment, though, about what this would mean.

First of all, it’s worth noting — as I didn’t, unfortunately, to the Post reporter — that tariff increases are, after all, tax increases. Whatever its effect on trade flows, a big increase in taxes will be contractionary. This is Keynes 101. Pick any activity accounting for 5 percent of GDP and slap a 40 percent tax on it, and it’s a safe bet that aggregate income will be lower as a result. The logic of the exercise would have been clearer if the tariff revenue were offset by a cut in some other tax, or increase in government spending. (Maybe this is what Trump means when he says Mexico will pay for the wall?) Then it would be clearer how much of the predicted impact comes from the tariff specifically, as opposed to the shift toward austerity that any such a big tax increase implies. The point is, even if you decide that a 2 percent fall in US GDP is the best estimate of the tariff’s impact, it wouldn’t follow that tariffs as such are a bad idea. It could be that a big tax increase is.

Second, let’s step back for a moment. While Mexico and China are two of our largest trade partners, they still account for less than a quarter of total US trade. Given passthrough of 0.3, the 45/35 percent tariff on Chinese/Mexican goods would raise overall US import prices by about 3 percent. Even with 100 percent passthrough, the tariffs would raise overall import prices by just 10 percent. The retaliatory tariffs would raise US export prices by about half this — 5 percent with full passthrough. (The difference is because these two countries account for a smaller share of US exports than of US imports). Now, let’s look at the movements of the dollar in recent years.

dollar exrate

Since 2014, the dollar has risen 15 percent. That’s a 15 percent increase in the price of US goods in all our export markets — three times the impact of the hypothetical Mexican and Chinese tariffs. But before that, from 2002 to 2008, the dollar fell by over 20 percent. That raised the price of US imports by twice as much as the hypothetical Trump tariff. And so on on back to the 1970s. If you believe Zandi’s numbers, then the rise in the dollar over the past two years should already have triggered a severe recession. Of course it has not. It would be foolish to deny that movements of the dollar have had some effect on US output and employment. But no one,  I think, would claim impacts on anything like this scale. Still, one thing is for sure: If you believe anything like Zandi’s numbers on the macro impacts of trade price changes, then it’s insane to allow exchange rates to be set by private speculators.

So if Zandi is wrong about the macro impact of tariffs, does that mean Trump is right? No. First of all, while I don’t think there’s any way to defend Zandi’s claim of a very large negative impact on GDP of a tariff (or of a more respectable, but economically equivalent, depreciation of the dollar), it’s almost as hard to defend a large positive impact. Despite all the shouting, the relative price of Chinese goods is just not a very big factor for aggregate demand in the US. If the goal is stronger demand and higher wages here, there are various things we can do. A more favorable trade balance with China (or Mexico, or anywhere else) is nowhere near the top of that list. Second, the costs of the tariff would be substantial for the rest of the world. It’s important not to lose sight of the fact that China, over the past generation, has seen perhaps the largest rise in living standards in human history. We can debate how critical exports to the US were in this process, but certainly the benefits to China of exports to the US were vastly greater than whatever costs they created here.

But the fact that an idea is wrong, doesn’t mean that we can ignore evidence and logic in refuting it. Trumpism is bad enough on the merits. There’s no need to exaggerate its costs.


UPDATE: My spreadsheet is here, if you want to play with alternative parameter values.


Unemployment and Productivity Growth

I write here frequently about “the money view” — the idea that we need to see economic relationships as a system of money flows and money commitments, that is not reducible to the “real” production and exchange of goods and services. Seeing the money-game as a self-contained system is the first step; the next step is to ask how this system interacts with the concrete activities of production.

One way to look at this interface is through the concept of potential output, and its relationship to current expenditure, or demand. In the textbook view, there is no connection between the long-run evolution of potential output with demand. This is a natural view if you think that economic quantities have an independent material existence. First we have scarce resources, then the choice about which end to devote them to. Knut Wicksell suggests somewhere an evocative metaphor for this view of economic growth: It’s as if we had a cellar full off wine in barrels, which will improve with age. The problem of economic growth is then equivalent to choosing the optimal tradeoff between having better wine, and drinking it sooner than later. But whatever choice we make, all the wine is already there. Ramsey and Solow growth models, with their “golden rule” growth rate, are descriptions of this kind of problem. Aggregate demand doesn’t come into it.

From our point of view, on the other hand, production is a creative, social activity. Economic growth is not a matter of allowing an exiting material process to continue operating through time, but of learning how to work together in new ways. The fundamental problem is coordination, not allocation.  From this point of view, the technical conditions of production are endogenous to the organization of production, and the money payments that structure it. So it’s natural to think that aggregate expenditure could be an important factor determining the pace at which productive activity can be reorganized.

Now, whether demand actually does matter in the longer run is hotly debated point in heterodox economics. You can find very smart Post Keynesians like Steve Fazzari arguing that it does, and equally smart Marxists like Dumenil and Levy arguing that it does not. (Amitava Dutt has a good summary; Mark Setterfield has a good recent discussion of the formal issues of incorporating demand into Kaldorian growth models.) But within our framework, at least it is possible to ask the question.

Which brings me to this recent article in the Real World Economic Review. I don’t recommend the piece — it is not written in a way to inspire confidence. But it does make an interesting claim, that over the long run there is an inverse relationship between unemployment and labor productivity growth in the US, with average labor productivity growth equal to 8 minus the unemployment rate. This is consistent with the idea that demand conditions influence productivity growth, most obviously because pressures to economize on labor will be greater when labor is scarce.

A strong empirical regularity like this would be interesting, if it was real. But is it?

Here is one obvious test (a bit more sensible to me than the approach in the RWER article). The figure below shows the average US unemployment rate and real growth rate of hourly labor productivity for rolling ten-year windows.

It’s not exactly “the rule of 8” — the slope of the regression line is just a big greater than -0.5, rather than -1. But it is still a striking relationship. Ten-year periods with high growth of productivity invariably also have low unemployment rates; periods of high average unemployment are invariably also periods of slow productivity growth.

Of course these are overlapping periods, so this tells us much less than it would if they were independent observations. But the association of above-average productivity growth with below-average unemployment is indeed a historical fact, at least for the postwar US. (As it turns out, this relationship is not present in most other advanced countries — see below.) So what could it mean?

1. It might mean nothing. We really only have four periods here — two high-productivity-growth, low-unemployment periods, one in the 1950s-1960s and one in the 1990s; and two low-productivity-growth, high-unemployment periods, one in the 1970s-1980s and one in the past decade or so. It’s quite possible these two phenomena have separate causes that just happened to shake out this way. It’s also possible that a common factor is responsible for both — a new technology-induced investment boom is the obvious candidate.

2. It might be that high productivity growth leads to lower unemployment. The story here I guess would be the Fed responding to a positive supply shock. I don’t find this very plausible.

3. It might be that low unemployment, or strong demand in general, fosters faster productivity growth. This is the most interesting for our purposes. I can think of several versions of this story. First is the increasing-returns story that originally motivated Verdoorn’s law. High demand allows firms to produce further out on declining cost curves. Second, low unemployment could encourage firms to adopt more labor-saving production techniques. Third, low unemployment might associated with more rapid movement of labor from lower-productivity to higher-productivity activities. (In other words, the relationship might be due to lower visible unemployment being associated with lower disguised unemployment.) Or fourth, low unemployment might be associated with a relaxing of the constraints that normally limit productivity-boosting investment — demand itself, and also financing. In any of these stories, the figure above shows a causal relationship running from the x-axis to the y-axis.

One scatterplot of course hardly proves anything. I’m really just posing the question. Still, this one figure is enough to establish one thing: A positive relationship between unemployment and labor productivity has not been the dominant influence on either variable in the postwar US. In particular, this is strong evidence against the idea the idea of technological unemployment, beloved by everyone from Jeremy Rifkin to Lawrence Summers. (At least as far as this period is concerned — the future could be different.) To tell a story in which paid labor is progressively displaced by machines, you must have a positive relationship between labor productivity and unemployment. But historically, high unemployment has been associated with slower growth in labor productivity, not faster. So we can say with confidence that whatever has driven changes in unemployment over the past 75 years, it has not been changes in the pace at which human labor is replaced by technology.

The negative relationship between unemployment and productivity growth, whatever it means, turns out to be almost unique to the US. Of the dozen or so other countries I looked at, the only one with a similar pattern is Japan, and even there the relationship is weaker. I honestly don’t know what to make of this. But if you’re interested, the other scatterplots are below the fold.

Note: Labor productivity is based on real GDP per hour, from the BLS International Labor Comparisons project; unemployment is the harmonized unemployment rate for all persons from the OECD Main Economic Indicators database. I used these because they are (supposed to be) defined consistently across countries and were available on FRED. Because the international data covers shorter periods than the US data does, I used 8-year windows instead of 10-year windows.

Work, Unemployment and Aggregate Demand

(I originally posted this as a series of comments on a 2012 post at Steve Randy Waldman’s Interfluidity. In that post, Steve suggested that we should think of redistribution under capitalism as “the poor collectively sell[ing] insurance against riot and revolution, which the rich are happy to pay for with modest quantities of efficiently produced goods.”)

In Theories of Surplus Value, Marx writes:

Assume that the productivity of industry is so advanced that whereas earlier two-thirds of the population were directly engaged in material production, now it is only one-third. Previously 2/3 produced means of subsistence for 3/3; now 1/3 produce for 3/3. Previously 1/3 was net revenue (as distinct from the revenue of the labourers), now 2/3. Leaving [class] contradictions out of account, the nation would now use 1/3 of its time for direct production, where previously it needed 2/3. Equally distributed, all would have 2/3 more time for unproductive labour and leisure. But in capitalist production everything seems and in fact is contradictory… Those two-thirds of the population consist partly of the owners of profit and rent, partly of unproductive labourers (who also, owing to competition, are badly paid). The latter help the former to consume the revenue and give them in return an equivalent in services—or impose their services on them, like the political unproductive labourers. It can be supposed that—with the exception of the horde of flunkeys, the soldiers, sailors, police, lower officials and so on, mistresses, grooms, clowns and jugglers—these unproductive labourers will on the whole have a higher level of culture than the unproductive workers had previously, and in particular that ill-paid artists, musicians, lawyers, physicians, scholars, schoolmasters, inventors, etc., will also have increased in number.

A large and growing share of employment, in other words, is unnecessary from a technical standpoint. It exists because useless jobs are more conducive to social stability than either mass poverty or a social wage. The payments the majority of the population receives for not rioting or rebelling look better when they are dressed up as payment for our work as mistresses, grooms, jugglers — or as yoga instructors or economics professors. This way, people are still dependent on a boss. In a differently organized world, we could dispense with most of these jobs and take the benefits of increased productivity in some combination of shorter hours for productive workers and a shift toward more intrinsically fulfilling (craft-like) forms of productive work.

By starting from here we can think more sensibly about employment and unemployment. From a macroeconomic standpoint, all we need is that expenditure on unproductive labor changes in some rough proportion with income.

From my point of view, the essential facts about employment are (1) As long as the most socially accepted form of claim on the social product is wages for work, work will be found for people, along the lines Marx suggests. (This is not true in poor societies, where a large portion of the poor engage in subsistence labor, of either the traditional or garbage-picking variety.) And (2) In the short run, employment will rise and fall as the rich feel a smaller or a greater need for the insurance-value of financial wealth.

As soon as you being to think about employment in terms of an input of labor to a production process, you’ve taken a wrong turn. We should not try to give supply-based explanations of unemployment, i.e. to show how the allocation of some stock of productive resources by some decision makers could generate unemployment. Unemployment is strictly a phenomenon of aggregate demand.

Unemployment in advanced countries is not characterized by exogenous factor supplies and Leontief-type production functions, where some factors are exhausted leaving an excess supply of their complements.  (The implicit model that lies behind various robots-will-take-all-the-jobs stories.) Unemployment in capitalist economies involves laid-off workers and idle factories; it involves unemployed construction workers and rising homelessness; it involves idle farmworkers and apples rotting on the trees. Unemployment never develops because we need fewer people to make the stuff, but because less stuff is being made. (Again, things are different in poor countries, and in the early stages of industrialization historically.) Unemployment cannot be explained without talking about aggregate demand any more than financial crises can be explained without talking about money and credit. It exists only to the extent that income and expenditure are determined simultaneously.

Unemployment rises when planned money expenditure falls for a given expected money income. Unemployment falls when planned money expenditure rises for a given expected money income. Conditions of production have no (direct) effect one way or the other.

Recognizing that unemployment is an aggregate expenditure phenomenon, not a labor-market phenomenon, helps avoid many errors. For example:

It is natural to think of unemployed people as people not engaged in productive work. This is wrong. The two things have nothing to do with each other. Unemployed people are those whose usual or primary claim on the social product takes the form of a wage, but who are not currently receiving a wage. There are lots of people who do not receive wages but are not unemployed because they have other claims on the social product — children, retirees, students, caregivers, the institutionalized, etc. Almost all of tehse people are capable of productive work, and many are actively engaged in it — caregiving and other forms of household production are essential to society’s continued existence. At the same time, there many people who do receive wages but who are not engaged in productive work; one way to define these is as people whose employment forms part of consumption out of profits or rents.

While there is no relationship between people’s capability for and/or engagement in useful work, on the one hand, and employment, on the other, there is a close link between aggregate expenditure and employment, simply because a very large fraction of expenditure takes the form directly or indirectly of wages, and aggregate wages adjust mainly on the extensive rather than the intensive margin. So when we see people unemployed, we should never ask, why does the production of society’s desired outputs no longer require their labor input? That is a nonsense question that will lead nowhere but confusion. Instead we should ask, why has there been a fall in planned expenditure?


Going beyond the 2012 conversation, two further thoughts:

1. The tendency to talk about unemployment in terms of why some peoples’ labor is no longer needed for production, is symptomatic of a larger confusion. This is the confusion of imagining money claims and payments as a more or less transparent representation of physical and social realities, as opposed to a distinct system that rests on but is substantially independent from underlying social and biological existence. Baseball requires human beings who can throw, hit and run; but the rules of baseball are not simply shorthand for people’s general activity of throwing, hitting and running. Needless to say, economics education assiduously cultivates the mixing-up of the money game with the substrate upon which it is played.

2. It’s natural to think of productive and unproductive labor as two distinct kinds of employment, or at least as opposite poles on a well-defined continuum. Marx usually writes this way. But I don’t think this is right, or at least it becomes less valid as the division of labor becomes more extensive and as productive activity becomes more directly social and involves more coordination of activities widely separated in space and time, and more dependent on the accumulation of scientific and technical knowledge.

Today our collective productive and creative activity requires the compliance of a very large number of people, both active and passive. This post will never be read by anyone if I don’t keep on typing. It will also never be read if the various tasks aren’t performed that are required to operate the servers where this blog is hosted, my internet connection and yours, the various nodes between our computers, the utilities that supply electricity to all the above, and so on. It would not be read if someone hadn’t assembled the computers, and transported and sold them to us; and if someone hadn’t developed the required technologies, step by step as far back as you want to go. It would not be read, or at least not by anyone except me and a few friends, if various people hadn’t linked to this blog over the years, and shared it on social media; and more broadly, if the development of blogs hadn’t gotten people into the habit of reading posts like this. Also, the post won’t be read if someone breaks into my house before before I finish writing it, and steals my laptop or smashes it with a hammer.

All of these steps are necessary to the production of a blog post. Some of them we recognize as “labor” entitled to wages, like whoever is watching the dials at Ravenswood. Some we definitely don’t, like the all-important not-stealing and not-smashing steps. And the status of some, like linking and sharing,  is being renegotiated. Again, a factory only runs if the workers choose to show up rather than stay home in bed; we reserve a share of the factory’s output to reward them for making that choice. It also only runs if passersby choose not to throw bricks through the windows; we don’t reserve any share of the output for them. But if we were going to write down the physical requirements for production to take place, the two choices would enter equivalently.

In a context where a large part of the conditions of production appear as tangible goods with physically rival uses; where the knowledge required for production was not itself produced for the market; where patterns of consumption are stable; where the division of labor is limited; where most cooperation takes the form of arms-length exchanges of goods rather than active coordination of productive activity; where production does not involve large commitments of fixed capital that are vulnerable to disruption; then the idea that there are distinct identifiable factors of production might not be too big a distortion of reality. In that context, splitting claims on the social product into shares attributable to each “factor” is not too disruptive; if anything, it can be a great catalyst for the development of productive capacities. But as the development of capitalism transforms and extends the division of labor, it becomes more and more difficult to separate out which activities that are contributing to a particular production process. So terms like productivity or productive labor lose touch with social reality.

You can find this argument in chapters 13-14 and 32 of Capital Volume 1. The brief discussion in chapter 32 is especially interesting, since Marx makes it clear that it is precisely this process that will bring capitalism to an end — not a fall in the rate of profit, which is never mentioned, nor a violent overthrow, which is explicitly rejected. But that thought will have to wait for another time.

Debt and Demand

One interesting issue in the ongoing secular stagnation debate is the relationship between debt and aggregate demand. In particular, there’s been a revival of the claim that there is something like a one to one relationship between changes in the ratio of debt to income, and final demand for goods and services.

I would like to reframe this claim a bit, drawing on my recent work with Arjun Jayadev. [1] In a nutshell: Changes in debt-income ratios reflect a number of macroeconomic variables, and until you have a specific story about which of those variables is driving the debt-income ratio, you can’t say what relationship to expect between that ratio and demand. We show in our paper that the entire post-1980 rise in household debt ratios can be explained, in an accounting sense, by higher real interest rates. Conversely, if the interest rates faced by households are lower in the future, debt-income ratios will decline without any fall in demand for real goods and services.

You might not know it from the current discussion, but there is an existing literature on these questions. The relationship between leverage — especially household debt — and aggregate demand was explored in a number of papers around the time of the last US credit crisis, in the late 1980s. Perhaps I’ll write a proper review of this material at some point; a short list would include Benjamin Friedman (1984 and 1986), Caskey and Fazzari (1991), Alfred Eichner (1991) and Tom Palley (1994 and 1997). It’s unfortunate that these earlier papers don’t get referred to in today’s discussion of debt and demand, by either mainstream or heterodox writers. [2]

For most of these writers, the important point was that the effect of debt on demand is two-faced: new borrowing can finance additional expenditure on real goods and services, but on the other hand debt service payments (in the presence of credit constraints) subtract from the funds available for current expenditure. Eichner, for instance, uses the equation E = F + delta-D – DS, or aggregate expenditure equals cashflow plus debt growth minus debt service payments.

More generally, to think systematically about the relationship between debt and household expenditure, we need to start from a consistent set of accounts. The first principle of financial accounting is that, for any economic unit, total sources of funds must equal total uses of funds. There are many ways of organizing accounts, at the level of the individual household or firm, at the level of the sector, or at the level of the nation, but this equality must always hold. You can slice up sources and uses of funds however you like, but total money coming in must equal total money going out.

The standard financial accounts for the United States are the Flow of Funds, maintained by the Federal Reserve. A number of alternative accounting frameworks are reflected in the social accounting matrixes developed by the late Wynne Godley and Lance Taylor and their students and collaborators.

Here’s one natural way of organizing sources and uses of funds for the household sector:

compensation of employees
capital income
transfer receipts
net borrowing 
consumption (including consumer durables)
residential investment
tax payments
interest payments
net acquisition of financial assets

The items before the equal sign are sources of funds; the items after are uses. [3] The first two uses of funds are included in GDP measured as income, while the latter two are not. Similarly, the first two uses of funds are included in GDP measured as expenditure, while the latter three are not.

When we look at the whole balance sheet, it is clear that borrowing cannot change in isolation. An increase in one source of funds must be accompanied by some mix of increase in some use(s) of funds, and decrease in other sources of funds. So if we want to talk about the relationship between borrowing and GDP, we need a story about what other items on the balance sheet are changing along with it. One possible story is that changes in borrowing are normally matched by changes in consumption, or in residential investment. This is the implicit story behind the suggestion that lower household borrowing will reduce final demand dollar for dollar. But there is no reason in principle why that has to be the main margin that household borrowing adjusts on, and as we’ll see, historically it often has not been.

So far we have been talking about the absolute levels of borrowing and other flows. But in general, we are not interested in the absolute level of borrowing, but on the ratio of debt to income. It’s common to speak about changes in borrowing and changes in debt-income ratios as if they were synonyms. [4]  But they are not. The debt-income ratio has a denominator as well as a numerator. The denominator is nominal income, so the evolution of the ratio depends  not only on household borrowing, but on real income growth and inflation. Faster growth of nominal income — whether due to real income growth or inflation — reduces the debt-income ratio, just as much as lower borrowing does.

In short: For changes in the debt-income ratio to be reflected one for one in aggregate demand, two things must be true. First, changes in the ratio must be due mainly to variation in the numerator, rather than the denominator. And second, changes in the numerator must be due mainly to variation in consumption and residential investment, rather than variation in other balance sheet items. How true are these things with respect to the rise in debt-income ratios over the past 30 years?

To frame the question in a tractable way, we need to simplify the balance sheet, combining some items to focus on the ones we care about. In our paper, Arjun and I were interested in debt ratios, not aggregate demand, so we grouped together all the non-credit flows into a single variable, which we called the household primary deficit. We defined this as all uses of funds except interest payments, minus all sources of funds except borrowing.

Here, I do things slightly differently. I divide changes in debt into those due to nominal income growth, those due to expenditures that contribute to aggregate demand (consumption and residential investment), and those due to non-demand expenditure (interest payments and net acquisition of financial assets.) For 1985 and later years, I also include the change in debt-income ratios attributable to default. (We were unable to find good data on household level defaults for earlier years, but there is good reason to think that household defaults did not occur at a macroeconomically significant level between the Depression and the Great Recession.) This lets us answer the question directly: historically, how closely have changes in household debt-income ratios been linked to changes in aggregate demand?

Figure 1 shows the trajectory of household debt for the US since 1929, along with federal debt and non financial business debt. (All are given as fractions of GDP.) As we can see, there have been three distinct episodes of rising household debt ratios since World War II: one in the decade or so immediately following the war, one in the mid-1980s, and one in the first half of the 2000s.

Figure 1: US debt-GDP ratios, 1929-2011

Figure 2 shows the annual change in the debt ratio, along with the decomposition described above. All variables are expressed as deviations from the 1950-2010 average. The heavy black line is the change in the debt-income ratio. The solid red line is final-demand expenditure, i.e. non-interest consumption plus residential investment. The dashed and dotted blue lines show the contributions of nominal income growth and non-demand expenditure, respectively. And the purple line with diamonds shows the contribution of defaults. (Defaults are measured relative to the 1985-2010 average.)

Figure 2: Decomposition of changes in the household debt-income ratio, 1949-2011

It’s clear from this figure that there is an important element of truth to the Keen-Krugman view that there is a tight link between the debt-incoem ratio and demand. There is evidently a close relationship between household demand and changes in the debt ratio, especially with respect to short-term variation. But that view is also missing something important. In some periods, there are substantial divergences between final demand from household and changes in the debt ratio. In particular, the increase in the household debt ratio in the 1980s (by about 20 points of GDP) took place during a period when consumption and residential investment by households were near their lowest levels since World War II. The increase in household debt after 1980 has often been described as some kind of “consumption binge”; this is the opposite of the truth.

The ambiguous relationship between household debt and aggregate demand can be seen in Table 1, which compares the periods of rising household debt with the intervening periods of stable or falling debt. The numbers are annual averages; to facilitate comparisons between periods, the averages for sub periods are again expressed as deviations from the 1950-2010 mean. (Or from the 1985-2010 mean, in the case of defaults.) The numbers are the contributions to the change i the debt-income ratio, so a positive value for nominal income growth indicates lower inflation and/or growth than the postwar average.

Table 1: Decomposition of changes in the household debt-income ratio, selected periods

Change in debt-income ratio Contribution of nominal income growth Aggregate-demand expenditure Non-demand   expenditure Defaults
1950-2010 mean 1.5 -4.9 89.1 17.7 -0.9
Difference from mean:
1949-1963 1.3 2.3 2.9 -4.3 N/A
1964-1983 -1.6 -1.4 -1.8 1.1 N/A
1984-1989 1.4 -0.3 -2.1 3.8 0.4
1990-1998 -0.5 0.3 -0.8 0.3 0.2
1999-2006 3.2 -1.2 3.1 1.7 0.1
2007-2010 -3.5 1.7 -1.4 -2.0 -1.3

What we see here is that while the first and third episodes of rising debt are indeed associated with higher than average household expenditure on real goods and services, the 1980s episode is not. The rise in debt in the 1980s is explained by a rise in non-demand expenditures. Specifically, it is entirely due to the rise in interest payments, which doubled from 3-4 percent of household income in the 1950s and 1960s to over 8 percent in the late 1980s. (Interest payments continued around this level up to the Great Recession, falling somewhat only in the past few years. The reason “non-demand expenditures” is lower after 1990 is because the household sector sharply reduced net acquisition of financial assets.) Also, note that while the housing booms of 1949-1963 and 1999-2006 saw almost identical levels of household expenditure on real goods and services, the household debt ratio rose nearly twice as fast in the more recent episode. The reason, again, is because of much higher interest payments in the 2000s compared with the immediate postwar period. Finally, as I’ve pointed out on this blog before, the deleveraging since 2008 would have been impossible without elevated household defaults, which approached 4 percent of outstanding household debt in 2009-2010 — partly offset by the sharp fall in household income in 2009, which raised the debt-income ratio.

Figure 3, from our paper, offers another way of looking at this. The heavy black line is the actual trajectory of the household debt-income ratio. The other lines show counterfactual scenarios in which non-interest household expenditures are at their historical levels, but growth, inflation and/or interest rates are held constant at their 1946-1980 average levels.

Figure 3: Counterfactual scenarios for the evolution of household-debt income ratios, 1946-2010

All these counterfactual scenarios show a spike in the 2000s: People really did borrow to pay for new houses! But the counterfactual scenarios also show lower overall trends of household debt, indicating that slower income growth, lower inflation and higher interest rates all contributed to the rise of household debt post-1980, independent of changes in borrowing behavior. Most interestingly, the red line shows that new borrowing after 1980 was lower than new borrowing in the 1950s, 60s and 70s; if households had engaged in the exact same spending on consumption, residential investment and financial assets as they actually did, but inflation, growth and interest rates had remained at their pre-1980 levels, the household debt-income ratio would have trended gradually downward.

To the extent that rising debt-income ratios after 1980 were the result of higher interest rates and disinflation, they were not contributing to aggregate demand. And if lower interest rates and and, perhaps, higher inflation and/or higher default rates bring down debt ratios in the future, deleveraging will not be a headwind for demand. 

It is customary to see rising debt as the result of private choices to finance higher expenditures by issuing new credit-market liabilities. But historically, it is equally correct to see rising debt as the result of political choices that increase the real value of existing liabilities.

[1] I’m pleased to report that a version of this paper has been accepted for publication by American Economic Journal: Macroeconomics. This has caused some adjustment in my view of the permeability of the “mainstream-heterodox” divide.

[2] This neglect of the earlier literature is especially puzzling since several of the protagonists of the 1990-era discussion are active in the sequel today. Steve Fazzari, for instance, in his several superb recent papers (with Barry Cynamon) on household debt, does not refer to his own 1991 paper, tho it is dealing with substantially the same questions. 

[3] Only a few minor items are left out. This grouping of sources and uses of funds essentially follows Lance Taylor’s social accounting matrices, as presented in Reconstructing Macroeconomics and elsewhere. Neither the NIPAs nor the Flow of Funds present household accounts in exactly this way. The Flow of Funds groups all three sources of household income together, treats consumer durables as a separate category of household investment, and treats interest payments as consumption. The NIPAs treat residential investment and mortgage interest payments as their own sector, separate from the household sector, and omits borrowing and net acquisition of financial assets. The NIPAs also include a number of noncash items, of which the most important is the imputed flow of housing services from the owner-occupied housing sector to the household sector and the corresponding imputed rental payments from the household sector to the owner-occupied residential sector.

[4] For example, a recent paper on the causes of “The Rise in U.S. Household Indebtedness” begins with the sentence, “During the past several decades in the United States, signi ficant changes have occurred in household saving and borrowing behavior,” with no sign of realizing that this is a different question than the one posed by the title.

Secular Stagnation, Progress in Economics

It’s the topic of the moment. Our starting point is this Paul Krugman post, occasioned by this talk by Lawrence Summers.

There are two ways to understand “secular stagnation.” One is that the growth rate of income and output will be slower in the future. The other is that there will be a systematic tendency for aggregate demand to fall short of the economy’s potential output. It’s the second claim that we are interested in.

For Krugman, the decisive fact about secular stagnation is that it implies a need for persistently negative interest rates. That achieved, there is no implication that growth rates or employment need to be lower in the future than in the past. He  is imagining a situation where current levels of employment and growth rates are maintained, but with permanently lower interest rates.

We could also imagine a situation where full employment was maintained by permanently higher public spending, rather than lower interest rates. Or we could imagine a situation where nothing closed the gap and output fell consistently short of potential. What matters is that aggregate expenditure by the private sector tends to fall short of the economy’s potential output, by a growing margin. For reasons I will explain down the road, I think this is a better way of stating the position than a negative “natural rate” of interest.

I think this conversation is a step forward for mainstream macroeconomic thought. There are further steps still to take. In this post I describe what, for me, are the positive elements of this new conversation. In subsequent posts, I will talk about the right way of analyzing these questions more systematically — in terms of a Harrod-type growth model — and  about the wrong way — in terms of the natural rate of interest.

The positive content of “secular stagnation”

1. Output is determined by demand.

The determination of total output by total expenditure is such a familiar part of the macroeconomics curriculum that we forget how subversive it is. It denies the logic of scarcity that is the basis of economic analysis and economic morality. Since Mandeville’s Fable of the Bees, it’s been recognized that if aggregate expenditure determines aggregate income, then, as Krugman says, “vice is virtue and virtue is vice.”

A great deal of the history of macroeconomics over the past 75 years can be thought of as various efforts to expunge, exorcize or neutralize the idea of demand-determined income, or at least to safely quarantine it form the rest of economic theory. One of the most successful quarantine strategies was to recast demand constraints on aggregate output as excess demand for money, or equivalently as the wrong interest rate. What distinguished real economies from the competitive equilibrium of Jevons or Walras was the lack of a reliable aggregate demand “thermostat”. But if a central bank or other authority set that one price or that one quantity correctly, economic questions could again be reduced to allocation of scarce means to alternative ends, via markets. Both Hayek and Friedman explicitly defined the “natural rate” of interest, which monetary policy should maintain, as the rate that would exist in a Walrasian barter economy. In postwar and modern New Keynesian mainstream economics, the natural rate is defined as the market interest rate that produces full employment and stable prices, without (I think) explicit reference to the intertemporal exchange rate that is called the interest rate in models of barter economies. But he equivalence is still there implicitly, and is the source of a great deal of confusion.

I will return to the question of what connection there is, if any, between the interest rates we observe in the world around us, and what a paper like Samuelson 1958 refers to as the “interest rate.” The important thing for present purposes is:

Mainstream economic theory deals with the problems raised when expenditure determines output, by assuming that the monetary authority sets an interest rate such that expenditure just equals potential output. If such a policy is followed successfully, the economy behaves as if it were productive capacity that determined output. Then, specifically Keynesian problems can be ignored by everyone except the monetary-policy technicians. One of the positive things about the secular stagnation conversation, from my point of view, is that it lets Keynes back out of this box.

That said, he is only partway out. Even if it’s acknowledged that setting the right interest rate does not solve the problem of aggregate demand as easily as previously believed, the problem is still framed in terms of the interest rate.

2. Demand normally falls short of potential

Another strategy to limit the subversive impact of Keynes has been to consign him to the sublunary domain of the short run, with the eternal world of long run growth still classical. (It’s a notable — and to me irritating — feature of macroeconomics textbooks that the sections on growth seem to get longer over time, and to move to the front of the book.) But if deviations from full employment are persistent, we can’t assume they cancel out and ignore them when evaluating an economy’s long-run trajectory.

One of the most interesting parts of the Summers talk came when he said, “It is a central pillar of both classical models and Keynesian models, that it is all about fluctuations, fluctuations around a given mean.” (He means New Keynesian models here, not what I would consider the authentic Keynes.) “So what you need to do is have less volatility.” He introduces the idea of secular stagnation explicitly as an alternative to this view that demand matters only for the short run. (And he forthrightly acknowledges that Stanley Fischer, his MIT professor who he is there to praise, taught that demand is strictly a short-run phenomenon.) The real content of secular stagnation, for Summers, is not slower growth itself, but the possibility that the same factors that can cause aggregate expenditure to fall short of the economy’s potential output can matter in the long run as well as in the short run.

Now for Summers and Krugman, there still exists a fundamentals-determined potential growth rate, and historically the level of activity did fluctuate around it in the past. Only in this new era of secular stagnation, do we have to consider the dynamics of an economy where aggregate demand plays a role in long-term growth. From my point of view, it’s less clear that anything has changed in the behavior of the economy. “Secular stagnation” is only acknowledging what has always been true. The notion of potential output was never well defined. Labor supply and technology, the supposed fundamentals, are strongly influenced by the level of capacity utilization. As I’ve discussed before, once you allow for Verdoorn’s Law and hysteresis, it makes no sense to talk about the economy’s “potential growth rate,” even in principle. I hope the conversation may be moving in that direction. Once you’ve acknowledged that the classical allocation-of-scarce-means-to-alternative-ends model of growth doesn’t apply in present circumstances, it’s easier to take the next step and abandon it entirely.

3. Bubbles are functional

One widely-noted claim in the Summers talk is that asset bubbles have been a necessary concomitant of full employment in the US since the 1980s. Before the real estate bubble there was the tech bubble, and before that there was the commercial real estate bubble we remember as the S&L crisis. Without them, the problem of secular stagnation might have posed itself much earlier.

This claim can be understood in several different, but not mutually exclusive, senses. It may be (1) interest rates sufficiently low to produce full employment, are also low enough to provoke a bubble. It may be (2) asset bubbles are an important channel by which monetary policy affects real activity. Or it may be (3) bubbles are a substitute for the required negative interest rates. I am not sure which of these claims Summers intends. All three are plausible, but it is still important to distinguish them. In particular, the first two imply that if interest rates could fall enough to restore full employment, we would have even more bubbles — in the first case, as an unintended side effect of the low rates, in the second, as the channel through which they would work. The third claim implies that if interest rates could fall enough to restore full employment, it would be possible to do more to restrain bubbles.

An important subcase of (1) comes when there is a minimum return that owners of money capital can accept. As Keynes said (in a passage I’m fond of quoting),

The most stable, and the least easily shifted, element in our contemporary economy has been hitherto, and may prove to be in future, the minimum rate of interest acceptable to the generality of wealth-owners.[2] If a tolerable level of employment requires a rate of interest much below the average rates which ruled in the nineteenth century, it is most doubtful whether it can be achieved merely by manipulating the quantity of money.  Cf. the nineteenth-century saying, quoted by Bagehot, that “John Bull can stand many things, but he cannot stand 2 per cent.”

If this is true, then asking owners of money wealth to accept rates of 2 percent, or perhaps much less, will face political resistance. More important for our purposes, it will create an inclination to believe the sales pitch for any asset that offers an acceptable return.

Randy Wray says that Summers is carrying water here for his own reputation and his masters in Finance. The case for bubbles as necessary for full employment justifies his past support for financial deregulation, and helps make the case against any new regulation in the future. That may be true. But I still think he is onto something important. There’s a long-standing criticism of market-based finance that it puts an excessive premium on liquidity and discourages investment in long-lived assets. A systematic overestimate of the returns from fixed assets might be needed to offset the systematic overestimate of the costs of illiquidity.

Another reason I like this part of Summers’ talk is that it moves us toward recognizing the fundamental symmetry between between monetary policy conventionally defined, lender of last resort operations and bank regulations. These are different ways of making the balance sheets of the financial sector more or less liquid. The recent shift from talking about monetary policy setting the money stock to talking about setting interest interest rates was in a certain sense a step toward realism, since there is nothing in modern economies that corresponds to a quantity of money. But it was also a step toward greater abstraction, since it leaves it unclear what is the relationship between the central bank and the banking system that allows the central bank to set the terms of private credit transactions. Self-interested as it may be, Summers call for regulatory forbearance here is an intellectual step forward. It moves us toward thinking of what central banks do neither in terms of money, nor in terms of interest rates, but in terms of liquidity.

Finally, note that in Ben Bernanke’s analysis of how monetary policy affects output, asset prices are an important channel. That is an argument for version (2) of the bubbles claim.

4. High interest rates are not coming back

For Summers and Krugman, the problem is still defined in terms of a negative “natural rate” of interest. (To my mind, this is the biggest flaw in their analysis.) So much of the practical discussion comes down to how you convince or compel wealth owners to hold assets with negative yields. One solution is to move to permanently higher inflation rates. (Krugman, to his credit, recognizes that this option will only be available when or if something else raises aggregate demand enough to push against supply constraints.) I am somewhat skeptical that capitalist enterprises in their current form can function well with significantly higher inflation. The entire complex of budget and invoicing practices assumes that over some short period — a month, a quarter, even a year — prices can be treated as constant. Maybe this is an easy problem to solve, but maybe not. Anyway, it would be an interesting experiment to find out!

More directly relevant is the acknowledgement that interest rates below growth rates may be a permanent feature of the economic environment for the foreseeable future. This has important implications for debt dynamics (both public and private), as we’ve discussed extensively on this blog. I give Krugman credit for saying that with i < g, it is impossible for debt to spiral out of control; a deficit of any level, maintained forever, will only ever cause the debt-GDP ratio to converge to some finite level. (I also give him credit for acknowledging that this is a change in his views.) This has the important practical effect of knocking another leg out from the case for austerity. It’s been a source of great frustration for me to see so many liberal, “Keynesian” economists follow every argument for stimulus with a pious invocation of the need for long-term deficit reduction. If people no longer feel compelled to bow before that shrine, that is progress.

On a more abstract level, the possibility of sub-g or sub-zero interest rates helps break down the quarantining of Keynes discussed above. Mainstream economists engage in a kind of doublethink about the interest rate. In the context of short-run stabilization, it is set by the central bank. But in other contexts, it is set by time preferences and technological tradeoff between current and future goods. I don’t think there was ever any coherent way to reconcile these positions. As I will explain in a following post, the term “interest rate” in these two contexts is being used to refer to two distinct and basically unrelated prices. (This was the upshot of the Sraffa-Hayek debate.) But as long as the interest rate observed in the world (call it the “finance” interest rate) behaved similarly enough to the interest rate in the models (the “time-substitution” interest rate), it was possible to ignore this contradiction without too much embarrassment.

There is no plausible way that the “time substitution” interest rate can be negative. So the secular stagnation conversation is helping reestablish the point — made by Keynes in chapter 17 of the General Theory, but largely forgotten — that the interest rates we observe in the world are something different. And in particular, it is no longer defensible to treat the interest rate as somehow exogenous to discussions about aggregate demand and fiscal policy. When I was debating fiscal policy with John Quiggin, he made the case for treating debt sustainability as a binding constraint by noting that there are long periods historically when interest rates were higher than growth rates. It never occurred to him that it makes no sense to talk about the level of interest rates as an objective fact, independent of the demand conditions that make expansionary fiscal policy desirable. I don’t mean to pick on John — at the time it wasn’t clear to me either.

Finally, on the topic of low interest forever, I liked Krugman’s scorn for the rights of interest-recipients:

How dare anyone suggest that virtuous individuals, people who are prudent and save for the future, face expropriation? How can you suggest steadily eroding their savings either through inflation or through negative interest rates? It’s tyranny!
But in a liquidity trap saving may be a personal virtue, but it’s a social vice. And in an economy facing secular stagnation, this isn’t just a temporary state of affairs, it’s the norm. Assuring people that they can get a positive rate of return on safe assets means promising them something the market doesn’t want to deliver – it’s like farm price supports, except for rentiers.

It’s a nice line, only slightly spoiled by the part about “what the market wants to deliver.” The idea that it is immoral to deprive the owners of money wealth of their accustomed returns is widespread and deeply rooted. I think it lies behind many seemingly positive economic claims. If this conversation develops, I expect we will see more open assertions of the moral entitlement of the rentiers.

The Cash and I

Martin Wolf in the FT the other day:

The third challenge is over the longer-term sources of demand. I look at this issue in terms of the sectoral financial balances – the balances between income and spending – in the household, business, external and government sectors. The question, then, is where expansion will come from. In the first quarter of this year the principal offset to fiscal contraction was the declining household surplus. 

What is needed, as well, is a big swing towards surplus in the US current account or a jump in corporate investment, relative to retained profits. Neither seems imminent, though the second seems more likely than the first. The worry is that the only way to balance the economy will be via big new bubbles. If so, this is not the fault of the Fed. It is the fault of structural features of the domestic and global economies…

This is a good point, which should be made more. If we compare aggregate expenditure today to expenditure just before the recession, it is clear that the lower level of demand today is all about lower  consumption. But maybe that’s not the best comparison, because during the housing boom period, consumption was historically high. If we take a somewhat longer view, what’s unusually low today is not household consumption, but business investment. Weak demand is about I, not C.

This is especially clear when we compare investment by businesses to what they are receiving in the form of profits, or, better cashflow from operations — after-tax profits plus depreciation. [1] Here is the relationship over the past 40 years:

Corporate Investment and Cashflow from Operations as Fraction of Total Assets, 1970-2012

The graph shows annualized corporate investment and cashflow, normalized by total assets. Each dot is data from one quarter; to keep the thing legible, I’ve only labeled the fourth quarter of each year. As you can see, there used to be a  clear relationship between corporate profitability and corporate investment. For every additional $150, more or less, that a corporation took in from operations, it would increase capital expenditures by $100. This relationship held consistently through the 1960s (not shown), the 1970s, the 1980s and 1990s.

But now look at the past ten years, the period after 2001Q4.  Corporate investment rates are substantially lower throughout this period than at any earlier time (averaging around 3.5% of total assets, compared with 5% of assets for 1960-2001). And the relationship between aggregate profit and investment rates has simply disappeared.

Some people might say that the problem is in the financial system, that even profitable businesses can’t borrow because of a breakdown in intermediation, a shortage of liquidity, an unwillingness of risk-averse investors to hold their debt, etc. I don’t buy this story for a number of reasons, some of which I’ve laid out in recent posts here. But it at least has some certain prima facie plausibility for the period following the great financial crisis. Not for for the whole decade-plus since 2001. Saying that investment is low today because businesses can’t find anyone to buy their bonds is merely wrong. Saying that’s why investment was low in 2005 is absurd.

(And remember, these are aggregates, so they mainly reflect the largest corporations, the ones that should have the least problems borrowing.)

So what’s a better story?

I am going to save my full answer for another post. But regular readers will not be surprised that I think the key is a shift in the relationship between corporations and shareholders. I think there’s a sense in which the binding constraint on investment has changed from the terms on which management can get funds into the corporation, from profits or borrowing, to the terms are on which they can keep them from going out, to investors. But the specific story doesn’t matter so much here. You can certainly imagine other explanations. Like, “the China price” — even additional capacity that would be profitable today won’t be added if it there’s a danger of lower-cost imports entering that market.

The point of this post is just that corporate investment is historically low, both in absolute terms and relative to profitability. And because this has been true for a decade, it is hard to attribute this weakness to credit constraints, or believe that it will be responsive to monetary policy. (This is even more true when you recall that the link between corporate borrowing and investment has also essentially disappeared.) By contrast, household consumption remains high. I have the highest respect for Steve Fazzari, and agree that high income inequality is a key metric of the fucked-upness of our economy. But I don’t think it makes sense to think of the current situation in terms of a story where high inequality reduces demand by holding down consumption.

Consumption is red, on the right scale; investment is blue, on the left. Both as shares of GDP.

As I say, I’ll come back in a future post to my on preferred explanation for why a comparably profitable firm, facing comparable credit conditions, will invest less today than 20 or 30 years ago.

In the meantime, one other thing. That first graph is a nice tool for showing how a Marxist thinks about business cycles.

If you look at the graph carefully, you’ll see the points follow counterclockwise loops. It’s natural to see this as cycles. Like this:

Start from the bottom of a cycle, at a point like 1992. A rise in profits from whatever source leads to higher investment, mainly as a source of funds and but also because it raises expectations of future profitability. That’s the lower right segment of the cycle. High investment eventually runs into supply constraints, typically in the form of a rising wage share.[1] At that point profits begin to fall. Investment, however, continues high for a while, as the credit system allows firms to bridge a growing financing gap. That’s the upper right segment of the cycle. Eventually, though, if profits don’t recover, investment will follow them downward. This turning point often involves a financial crisis and/or abrupt fall in asset values, like the collapse of tech stocks in 2000. This is the upper left segment of the cycle.  Finally, in the  lower left, both profits and investment are low. But after some time the conditions for profitability are restored, and we move toward the right and begin a new cycle. This last step is less reliable than the others. It’s quite possible for the economy to come to rest at the lower left and wobble there for a while without any sustained change in either profits or investment. We see this in 2002-2003 and in 1988-1991.

(I think the investment boom of the late 70s and the persistent slump of the early 1990s are two of the more neglected episodes in recent economic history. The period around 1990, in particular, seems to have all the features that are supposed to be distinctive to the current macroeconomic conjuncture. At the time, people even called it a balance-sheet recession!)

For now, though, we’re not interested in the general properties of cycles. We’re interested in how flat and low the most recent two are, compared with earlier ones. That is the structural feature that Martin Wolf is pointing to. And it’s not a new feature of the post financial crisis period, it’s been the case for a dozen years at least, only temporarily obscured by the housing bubble.

UPDATE: In comments, Seth Ackerman asks if maybe using total assets to normalize investment and profits is distorting the picture. It’s a good question, but the answer is no. Here’s the same thing, with trend GDP in the denominator instead:

As you can see, the picture is basically the same. Investment in the 200s is still visibly depressed compared with earlier decades, and the relationship between profits and investment is much weaker. Of course, it’s always possible that current high profits will lead an investment boom in the next few years…

[1] Cash from operations is better than profits for at least two reasons. First, from the point of view of aggregate demand, we are interested in gross not net investment. A dollar of investment stimulates demand just as much whether it’s replacing old equipment or adding new. So our measure of income should also be gross of depreciation. Second, there are major practical and conceptual issues with measuring depreciation. Changes in accounting standards may result in very different official depreciation numbers in economically identical situations. By combining depreciation and profits, we avoid the problem of the fuzzy and shifting line between them, making it more likely that we are comparing equivalent quantities.

[2] A rising wage share need not, and often does not, take the form of rising real wages. In recent cycles especially, it’s more likely to combine flat real wages with a rising relative cost of wage goods.