Saturday, December 17, 2011

What We Talk About When We Don't Talk About Demand

There sure are a lot of ways to not say aggregate demand.

Here's the estimable Joseph Stiglitz, not saying aggregate demand in Vanity Fair:
The parallels between the story of the origin of the Great Depression and that of our Long Slump are strong. Back then we were moving from agriculture to manufacturing. Today we are moving from manufacturing to a service economy. The decline in manufacturing jobs has been dramatic—from about a third of the workforce 60 years ago to less than a tenth of it today. ... There are two reasons for the decline. One is greater productivity—the same dynamic that revolutionized agriculture and forced a majority of American farmers to look for work elsewhere. The other is globalization... (As Greenwald has pointed out, most of the job loss in the 1990s was related to productivity increases, not to globalization.) Whatever the specific cause, the inevitable result is precisely the same as it was 80 years ago: a decline in income and jobs. The millions of jobless former factory workers once employed in cities such as Youngstown and Birmingham and Gary and Detroit are the modern-day equivalent of the Depression’s doomed farmers.
This sounds reasonable, but is it? Nick Rowe doesn't think so. Let's leave aside globalization for another post -- as Stieglitz says, it's less important anyway. It's certainly true that manufacturing employment has fallen steeply, even while the US -- despite what you sometimes here -- continues to produce plenty of manufactured goods. But does it make sense to say that the rise in manufacturing productivity be responsible for mass unemployment in the country as a whole?

There's certainly an argument in principle for the existence of technological unemployment, caused by rapid productivity growth. Lance Taylor has a good discussion in chapter 5 of his superb new book Maynard's Revenge (and a more technical version in Reconstructing Macroeconomics.) The idea is that with the real wage fixed, an increase in labor productivity will have two effects. First, it reduces the amount of labor required to produce a given level of output, and second, it redistributes income from labor to capital. Insofar as the marginal propensity to consume out of profit income is lower than the marginal propensity to consume out of wage income, this redistribution tends to reduce consumption demand. But insofar as investment demand is driven by profitability, it tends to increase investment demand. There's no a priori reason to think that one of these effects is stronger than the other. If the former is stronger -- if demand is wage-led -- then yes, productivity increases will tend to lower demand. But if the latter is stronger -- if demand is profit-led -- then productivity increases will tend to raise demand, though perhaps not by enough to offset the reduced labor input required for a given level of output. For what it's worth, Taylor thinks the US economy has profit-led demand, but not necessarily enough so to avoid a Luddite outcome.

Taylor is a structuralist. (The label I think I'm going to start wearing myself.) You would be unlikely to find this story in the mainstream because technological unemployment is impossible if wages equal the marginal product of labor, and because it requires that output to be normally, and not just exceptionally, demand-constrained.

It's a good story but I have trouble seeing it having much to do with the current situation. Because, where's the productivity acceleration? Underlying hourly labor productivity growth just keeps bumping along at 2 percent and change a year. Over the whole postwar period, it averages 2.3 percent. Over the past twenty years, 2.2 percent. Over the past decade, 2.3 percent. Where's the technological revolution?

Just do the math. If underlying productivity rises at 2 percent a year, and demand constraints cause output to stay flat for four years [1], then we would expect employment to fall by 8 percent. In other words, lack of demand explains the whole fall in employment. [2] There's no need to bring in structural shifts or anything else happening on the supply side. A fall in demand, plus a stable rate of productivity increase, gets you exactly what we've seen.

It's important to understand why demand fell, but from a policy standpoint, no actually it isn't. As the saying goes, you don't refill a flat tire through the hole. The important point is that we don't need to know anything about the composition of output to understand why unemployment is so high, because the relationship between the level of output and employment is no different than it's always been.

But isn't it true that since the end of the recession we've seen a recovery in output but no recovery in employment? Yes, it is. So doesn't that suggest there's something different happening in the labor market this time? No, it doesn't. Here's why.

There's a well-established empirical relationship in macroeconomics called Okun's law, which says that, roughly, a one percentage point change in output relative to potential changes employment by one a third to a half a percentage point. There are two straightforward reasons for this: first, a significant fraction of employment is overhead labor, which firms need an equal amount of whether their current production levels are high or low. And second, if hiring and training employees is costly, firms will be reluctant to lay off workers in the face of declines in output that are believe to be temporary. For both these reasons (and directly contrary to the predictions of a "sticky wages" theory of recessions) employment invariably falls by less than output in recessions. Let's look at some pictures.

These graphs show the quarter by quarter annualized change in output (vertical axis) and employment (horizontal axis) over recent US business cycles. The diagonal line is the regression line for the postwar period as a whole; as you would expect, it passes through zero employment growth around two percent output growth, corresponding to the long-run rate of labor productivity growth.

1960 recession

1969 recession

1980 and 1981 recessions
1990 recession

2001 recession
2007 recession

What you see is that in every case, there's the same clockwise motion. The initial phase of the recession (1960:2 to 1961:1, 1969:1 to 1970:4, etc.) is below the line, meaning growth has fallen more than employment. This is the period when firms are reducing output but not reducing employment proportionately. Then there's a vertical upward movement at the left, when growth is accelerating and employment is not; this is the period when, because of their excess staffing at the bottom of the recession, firms are able to increase output without much new hiring. Finally there's a movement toward the right as labor hoards are exhausted and overhead employment starts to increase, which brings the economy back to the long-term relationship between employment and output. [3] As the figures show, this cycle is found in every recession; it's the inevitable outcome when an economy experiences negative demand shocks and employment is costly to adjust. (It's a bit harder to see in the 1980-1981 graph because of the double-dip recession of 1980-1981; the first cycle is only halfway finished in 1981:2 when the second cycle begins.)

There's nothing exceptional, in these pictures, about the most recent recession. Indeed, the accumulated deviations to the right of the long-term trend (i.e., higher employment than one would expect based on output) are somewhat greater than the accumulated deviations to the left of it. Nothing exceptional, that is, except how big it is, and how far it lies to the lower-left. In terms of the labor market, in other words, the Great Recession was qualitatively no different from other postwar recessions; it was just much deeper.

I understand the intellectual temptation to look for a more interesting story. And of course there are obviously structural explanations for why demand fell so far in 2007, and why conventional remedies have been relatively ineffective in boosting it. (Tho I suspect those explanations have more to do with the absence of major technological change, than an excess of it.) But if you want to know the proximate reason why unemployment is so high today, there's a recession on still looks like a sufficiently good working hypothesis.

[1] Real GDP is currently less than 0.1 percent above its level at the end of 2007.

[2] Actually employment is down by only about 5 percent, suggesting that if anything we need a structural story for why it hasn't fallen more. But there's no real mystery here, productivity growth is not really independent of demand conditions and always decelerates in recessions.

[3] Changes in hours worked per employee are also part of the story, in both downturn and recovery.


  1. Neat. So when we (including me) thought that the US was violating Okun's Law this time, it was just bad GDP data. I'm now so glad I couldn't come up with a good explanation of why the US was violating Okun's Law this time!

    "For what it's worth, Taylor thinks the US economy has profit-led demand, but not necessarily enough so to avoid a Luddite outcome."

    But, if so, couldn't monetary and/or fiscal policy shift the AD curve right to avoid the Luddite outcome? Does Taylor say why not?

  2. Hi Nick.

    Sorry, this is not a very good post. (If you hadn't commented on it I would probably have taken it down.) It mixes up a bunch of different issues that are related (at least in my head) but should be kept conceptually distinct. (1) Should we expect the shift from manufacturing to service employment (or agriculture to manufacturing) to produce a chronic downward shift in AD, as Stiglitz suggests? (2) Has there been some long-term change in the trend of labor productivity growth such that GDP growth that produced acceptable employment outcomes in the past no longer will do so? (3) Has the short-term (Okun's law) relationship between output and employment in the US changed compared with earlier periods? And (4) Is the Okun's law relationship different in the US than in other countries? Personally, I think the answers are no, no, no and yes, but that's not at all clear in what I wrote. Let's see if we can do better.

    On the first question, I basically agree with your post (the one I linked). The one caveat would be that insofar as a sectoral shift involves a redistribution between wages and profits, and there's a different propensity to consume out of profits than out of wages which is not offset by the elasticity of investment with respect to profits, it can have effects on aggregate demand. It's not clear to me that this is what Stiglitz is arguing; if you think this sort of underconsumption story is an important structural cause for the current demand shortfall, the shift from manufacturing to services seems like a rather indirect way of talking about it. And as you suggest, even if one did think that demand was wage-led so that a higher profit share would ten to reduce demand, there is no reason that couldn't be offset by demand-boosting policies. The question Taylor is answering is just, under what conditions can an acceleration in labor productivity growth reduce employment.

    OK, on the second question, I was thinking more of stuff like the zero-marginal product worker story that Tyler Cowen was telling last year, or the books and articles about "Our Jobless Future" and "The End of Work" that you see after every recent recession. My point is just that if labor hoarding and overhead labor are important, then the recovery from a recession is always going to involve a period where output growth accelerates before employment growth does. And almost every recent recovery does seem to involve such a period. The mistake is to extrapolate from that to a permanent change in the behavior of labor productivity.

    On the last two, I'm sure you've seen Robert Gordon's stuff on the Okun coefficient as an index of the power of workers vs management (weaker workers are easier to hire and fire, "disposable"); he argues that with the decline of unions, weaker labor-market regulation, etc., the Okun coefficient has moved toward one. On the other hand, I recently saw Michael Reich present some new work (which doesn't seem to be online, unfortunately) disputing that, and claiming that there is no long run change in the US Okun coefficient. This is my sense too, altho I am very far from an expert in this area.

    Anyway, that's what the phase diagrams were trying to show. In a sort of ultra-Okun world, where firms had no ability to adjust employment, we'd see the points of a demand-induced business cycle lying along a vertical line, since output would fall and rise while employment grew steadily. In the opposite case -- a no-Okun world -- the points of a business cycle would lie along the long-term diagonal. (Or actually, since firms still can't adjust the capital stock in the short run, we'd see the recession points falling above the diagonal, as firms shifted toward more capital-intensive production processes as output fell.) My suggestion -- which would need to be tested more systematically -- is that the more recent business cycles don't look any closer to no-Okun as opposed to ultra-Okun than the earlier cycles do.

  3. I have some problems with the concept of "productivity".

    [note: I'm just a dilettante in economics so if I say something really stupid please explain to me anyway]

    1) For what I can understand, "productivity" is supposed to measure the material efficiency of workers: for example, if a shoemaker is able to make 10 pairs of shoes in 8 hours, but later because of a tech improvement he becames able to make 15 pairs of shoes in the same 8 hours, his "productivity" increased by 50%.
    Let's call this concept "material productivity". This is what an "intuitive" notion of productivity is.

    2) But how can you compare the "material productivity" of a shoemaker (measured in shoes per hour) whit that of a plumber or of a banker?
    We can't, so we measure the "added value" instead of the material products. For example if the shoemaker in 10 hours makes 20 shoes, using 5$ of materials each, and sells each shoe at 10$, his productivity will be (10$-5$)x20/10h = 10$/h.
    Let's call this "economic productivity". This is what is usually meant by "productivity".

    3) This means that the average productivity of a nation is, basically, total GDP/ total hours worked.
    However, "economic productivity" is really different from "material productivity", since market forces that affect prices also affect economic, but not material productivity.
    For example, say that the shoemaker in 2) has to sell his shoes for 9$ each because of, e.g., Chinese competition. His economic productivity falls to (9$-5$)x20/10h = 8$/h, even if his material productivity didn't change.

    4) This also means that monetary/fiscal effects affect economic (but not necessariously material) productivity. For example, national GDP is limited by aggregate demand. When employment falls, the number of hours worked falls, but the GDP falls less because some people who lost their job still receive unemployment benefits, hence keeping AD up.
    Since "economic productivity" is GDP/total hours worked, this produces an increase in "economic productivity", without necessariously affect "material productivity".

    This seems very stupid to me, so I guess I'm wrong, but I don't understand were I'm wrong.

  4. @Josh,

    I’m not too satisfied with these graphs, either. They are almost as confusing as those other graphs I complained about.

    The problem is that you keep wanting to graph the temporal dynamics of recessions without including time as an explicit variable on a coordinate axis. That approach is reliably so opaque that it requires you to write lengthy expository captions to persuade the reader that there’s a cyclic regularity to the resulting enigmatic clouds of data points.

    At the very least, if you’re graphing such a gnarly cyclic path through a 2-D space (especially one so cluttered with data), you should highlight the path with color, as you did previously, but also mark the path’s start and endpoint, with separate symbols, and insert arrow-heads denoting the direction of the cycle.

    It would be better if you could revert to the good old visual vernacular—something no graph-maker should take lightly--and just put time prodeeding to the right on the x axis, perhaps for a normalized generic period of from 16 quarters prior to 16 quarters after a recession (or something). What you might then do is make the y axis some evocative composite function of employment and output. The ratio of delta output to delta employment is tricky because of division by zero problems. Maybe the y axis could be the ratio of indices of output and employment; the time-dependent effect of the recession might then manifest as a transient perturbation around a steadier trend line (or constant).

  5. Bravo, Random Lurker! What you just said about 'productivity' I have been wanting to say for a while.

  6. RL-

    Not stupid at all. The problem of how to aggregate all the heterogeneous products of economic activity into one number is a profound one, that most most economists breeze over.

    That said, your number 3 isn't really a problem. A case like that will reduce nominal GDP *and* the price level in equal proportions, so real output per hour won't change. In general, labor productivity as conventionally measured will not be affected by a change in relative prices, so there shouldn't be any divergence between measured productivity and what you're calling "material productivity."

    4 doesn't really work as you give it, either. Insofar as unemployment benefits are maintaining demand, they are being spent on currently produced output. And it takes as much labor to give a haircut (say) to an unemployed person as an employed one. So the fraction of income that is coming from UI benefits as opposed to wage income will have no effect on the relationship between GDP and hours worked.

    But even though I'm not really convinced by either of your cases, I agree with you on the broader point that there are problems with measuring productivity over time.

    For instance, go back to your shoemaker. Suppose he has an assistant. In a recession, he can't sell as many shoes as before, so he doesn't need the assistant. But since hiring & training a good assistant takes time, he doesn't want to fire the one he's got. So the national accounts show the same amount of labor producing fewer shoes, an apparent fall in productivity that has nothing to do with any regression in shoemaking technology. This is the Okun's law relationship. (Nick Rowe has suggested that we don't see this in the current recession, but I don't agree. Okun's law doesn't require an absolute fall in measured labor productivity during recessions, just a deceleration relative to trend.)

    Or again, suppose he does fire the assistant, and let's suppose, reasonably, that she can't find another job making shoes. But if she's not eligible for UI benefits, or reaches the time limit, or feels morally constrained to have a job, she'll find work doing something else, presumably something that pays less. Maybe -- Joan Robinson's example -- she'll end up selling pencils on a street corner. This kind of "disguised unemployment," as Robinson called it, will again show up as an apparent fall in labor productivity, since the value added in pencil-vending is less than in shoemaking. This will naively look like some lack of human capital -- how unproductive do you have to be to have nothing better all day than hawk pencils -- giving rise to well-meaning arguments by, unfortunately, people like Joe Stiglitz, that what's needed is more education & job training. When really it's just a symptom of inadequate demand.

    And then all this assume we can measure output. In the shoemaker case, no problem: shoes per hour. But over the longer term, as tastes and technology change qualitatively, it becomes sort of meaningless to compare quantities of output. Or as Keynes said,

    "To say that net output today is greater, but the price level lower, that ten years ago or one year ago, is a proposition of a similar character to the statement that Queen Victoria was a better queen but not a happier woman than Queen Elizabeth - a proposition not without meaning and not without interest, but unsuitable as material for the differential calculus."

    People usually treat as a throwaway line, but I think he meant it.

  7. @JW

    Thanks for the clear answer.

    About my point 3, I see you're right. I was thinking to the Italian case. Here in Italy we have a problem of falling productivity since some years; I think that if we speak of a non-closed market my point still makes some sense (Italy has a rather low-tech kind of business for a developed country AFAIK).

    About my point 4: Ouch, you're right!

  8. Do you really think the recessions of 2001 and 2008, subsequent recoveries were not qualitatively different from usual pattern of post-WWII recessions?

    You don't think the pace of recovery in employment seems a bit prolonged? The rise in wages in line with rising productivity, a bit non-existent?

    The most prominent parallel I see seems to prove the case for different: the general pattern is for housing and autos to lead the way into recovery and into recession, as those sectors respond to interest rates -- or more precisely, yield curve inversions. The recovery from 2001, as anemic as could have been imagined, (apparently) required blowing a huge housing bubble.

    Viewed over a 60 year horizon, it looks to me like the U.S. economy is experiencing something akin to rising insulin resistance. The Fed insulin of a positive-sloping yield curve, just doesn't process the sugar of an expanding suburban economy (housing & autos) like it used to. This last injection, given our apparent transition to Type 2 diabetes, seems to have put the economy into a near-coma.

    The Nick Rowe "gizmo economy" interpretation of Stiglitz reminds me of the way macroeconomists have traditionally dismissed Fisher's debt-deflation theory: "debt-deflation represented no more than a redistribution from one group (debtors) to another (creditors). Absent implausibly large differences in marginal spending propensities among the groups, it was suggested, pure redistributions should have no significant macroeconomic effects."[Ben Bernanke]
    Arrogant, and, tragically wrong.

    I think Stiglitz is groping, rather than grasping, at this point, but I think we can be kinder and humbler.

    The post-WWII period has been over for a decade. We do need some massive spending on public goods, and some rationale for motivating and structuring that spending.

  9. Bruce, you're absolutely right about the importance of being kind and humble, especially with people like Stiglitz. My next post will revisit his argument in, hopefully, a more constructive spirit.

    On the substance - you're right, recent recessions are different somehow, yes, but why? The point I'm most confident on -- which isn't really an argument with Stiglitz -- is that the issue is slower growth of demand in the recovery, as opposed to a changed relationship between output and employment.

    So why does demand recover more slowly? I don't know. But let's think about it. We can talk about two families of explanations. One says interest-sensitivity has declined, the other says there's some secular downward pressure on demand.

    So on the declining interest-sensitivity side you've got the zero-lower bound, basically just bad luck. Or the variant of that argument Steve Fazzari makes in his new Great Recession anthology (which looks *really* good), that the Great Moderation was the more or less unintended result of the space the high Volcker rates gave for big rate cuts afterward.

    Or you've got the version my friend Hasan Comert developed in his dissertation, that financial deregulation/innovation has basically delinked the Fed policy rate from the economically important rates. It's not that economic activity is less sensitive to interest rates, it's that the Fed can't move the rates that matter. I am moderately sympathetic to this argument.

    Then you've got the Richard Koo balance-sheet argument. Units are trying to optimize their debt levels while satisficing with respect to some ceiling on leverage; when the latter binds the former doesn't. So high levels of debt -> low interest rate sensitivity. Probably something to this one too.

    You could add a pure lower-interest-elasticity argument based, maybe, on changes in investment finance, but that would have trouble with the fact that business investment has never been noticeably sensitive to interest rates, monetary policy has always mainly worked through the housing market. Which is a good explanation for why it isn't working now, but not so good for the last two jobless recoveries.

    OK, then we've got the secular stagnation arguments. Most obvious is the income distribution/underconsumption argument. This might be what Stiglitz's story turns out to be;; it isn't really clear in the VF piece. Lance Taylor doesn't buy it, he thinks a redistribution to profits should boost investment demand by more than it reduces consumption demand. I think one can be agnostic about this question without falling into the know-nothingism of that Bernanke quote. (Great quote, btw.)

    There's Keynes' version, human demands are finite and eventually get satiated. Don't think that fits the current situation very well.

    On the other side, we could talk about factors that reduce business investment by reducing expected profitability. One version I think people should talk about more is the Ricardian (in the sense of bringing in back landlords as a distinct group in the functional distribution of income) story that growth today is typically choked off by a runup in the price of oil and other raw materials. (This is sort of an updating of the cyclical profit squeeze story that is, IMO, one of Marxism's more useful contributions to macro.)

    Then there's a secular downward shift in investment demand. This is the story I'm inclined towards. Some combination of greater power of rentiers within the firm, an absence of epochal innovations, and a shift in the source of profits from accumulation of physical capital to appropriation of rents (via brands, IP, etc.) means that businesses tend to invest less at a given level of profitability and interest rates.

    But from a policy standpoint, as I said in the post, it almost doesn't matter. If you believe any demand-side story, then the answer is just, more demand.

  10. Will,

    You may be right about the graphs. I don't think the specifically cyclical character of the relationship between output and employment in recessions would be visible in a standard time series graph; but I'm also pretty sure that my solution is not the best one.

  11. I'm puzzled by the claim that an increased profit share would increase investment demand. If one is already making a fine profit, then further investment would diminish one's rate-of-profit by boosting productivity and output without any corresponding increase in consumption demand. That would occur, of course, under highly competitive market conditions, but does anyone really think that competitive rather than oligopolistic conditions really prevail for major investing firms?

    If I understand Keynes, he was according a key role to investment demand in overall demand levels, based on a still mainly industrial economy. As investment demand declined, workers in the capital goods sector would be unemployed, decreasing consumption demand, leading to increased unemployment in the consumption goods sector, further lowering consumption and thus investment demand, etc. This was put down somewhat vaguely to the declining marginal efficiency of investment and the "animal spirits" of investors, (though it hard not to detect the Marxian account of over-accumulation of capital, with declining opportunities for profitable investment and a falling rate-of-profit there). However, technological development has meant that the share of industry in output and employment has been steadily diminishing.

    It occurred to me in the mid-1990's that those famous contradictions of capitalism were re-emerging in a high-tech form. IT and telecom were increasing the flexibility and reducing the costs of capital goods, while increasing their substitutability for (and control over) labor. And the synergy of the two technological complexes was essential both to globalized platforming of production supply chains and the intensive financialization of global and domestic economies. Hence a reduction in wages and wage-based consumption demand and an increase in the profit share, which can only be maintained for further real productive investment strategies of financialized rent extractions. Which hollows out the contribution of investment to aggregate demand rather than incentivizing further real investment.

  12. Er, "can only be maintained by substituting"...

  13. How is Taylor's Reconstructing Macro? Is it a good intro to the Post Keynesians?

  14. I'm puzzled by the claim that an increased profit share would increase investment demand. If one is already making a fine profit, then further investment would diminish one's rate-of-profit by boosting productivity and output without any corresponding increase in consumption demand. That would occur, of course, under highly competitive market conditions, but does anyone really think that competitive rather than oligopolistic conditions really prevail for major investing firms?

    Have to disagree with you here, John. There's nothing specifically neoclassical about the idea that there's a positive relationship between profits and investment. This will be true whatever the degree of monopoly. I think Marx and most (almost all?) Marxist economists would say the same thing.

    I think you're getting at something interesting & important in the rest of your comment, tho, which I'll hopefully have something too say about shortly.

    How is Taylor's Reconstructing Macro? Is it a good intro to the Post Keynesians?

    Probably the best I've read. (Altho that may be a matter of taste -- he's not at all into the Davidson/Shackle epistemology stuff, which some people would say is central to the PK project.)

  15. I should add, it depends on your background. You'll need some economics background or it will be heavy weather; Maynard's Revenge is a lot more accessible.

  16. My point wasn't that higher profits can't and don't increase investment demand, just that there is no such simple generalization, without specifying conditions where that would be the case and where not, at a "meso" level. Increased productivity and thus lower unit output costs at higher mark-ups would tend to run against the relative elasticity of demand for that output, and possible syndromes such as Baumol's cost disease, where gains in productivity reaped by less productive sectors. If there is a new technological complex that effectively solicits increased investment for its expansion, (such as, e.g. railroads, automobiles, computers), then an investment boom might well be forthcoming. Or if there are highly competitive conditions, then increased investment will be attracted and thus reduce initially higher profits. But where oligopolistic conditions generally prevail, since for a number of reasons oligopolies already maintain excess productive capacity and since they will generally invest to maintain their market-dominant quasi-rents, not to erode them, if there is a general tendency to an increased profit share and a decreased wage share in output, what would instigate an investment-led boom that could only lead on to a foreseeable crisis of over-production and under-consumption?

    What we've hadin the U.S., according to its official theorists, is a financial asset-based economy, in which asset prices were regarded as essential to maintaining READ amidst ever-increasing inequality of income and wealth, which effects the level of demand otherwise. Hence we have a crisis rooted in a huge debt overhang resulting from an economy running on hot air, er, financialized rent extractions from extant production capacity rather than real investment in increased productivity, (which, again, is not simply a macro-economic tend that can be simply assumed). The figures in fact show declining corporate investment/increased corporate savings in the last decade. And I think there is at least some good reason for thinking that recorded productivity gains reflect the effects of off-shoring low productivity labor, while the MNCs reap huge profits/rents from what amounts to FX arbitrage strategies, rather than from actual increases in productivity from technically improved real investments.

    (That said, the Stiglitz article was strange. According to my received opinion view, the 1920's were an agricultural depression accompanied by a huge boom in industrial investment and productivity, some much of the postulated shift from agricultural to industrial labor had already occurred before the Depression, rendering it hard to believe that the failure to write-down ag. debt rather than industrial over-investment, financial speculation and the failure to raise wage levels in accordance with productivity gains, hence rising inequality, were not the key factors. I would have to see Stiglitz' evidence for ag. debt being a key quantitative factor).