Thoughts on Summers on the Productivity Slowdown

I received this *.pdf and an invitation to read it from my colleague and sometimes co-author Alessandra Pelloni. I had a lot of thoughts which I post here as they can’t do any harm.

Two comments before the jump and 5 after the jump.

I think it will be hard to massage the data away from the conclusion that
when there are recessions, subsequent output growth is lower. The
question is, is the relationship best thought of as causal. After all, if
productivity slowed down, you might expect investment to slow down,
you might expect the stock market to go down. And so, if events were
happening that were reducing the underlying rate of growth, and people
came to realize that those events were happening, you might expect the
economy to go into recession.
And so the question where there’s certainly much more room for argument
is whether the relationships I described are causal. And here, there’s no
God’s truth. We don’t get to contrive recessions to do a controlled
experiment on the question. Blanchard and I do what I think is the sensible
way to approach the question, which is we look at recessions with
different causes. We look at recessions that are associated with tight
money to reduce inflation. That seems like a relatively pure case of a
demand side reduction. We look at recessions that follow the bursting of
bubbles. We look at other recessions not associated with those kinds of
events. We look at recessions associated with fiscal contractions. We look
at recessions associated with credit contractions.
Separately, I have looked at declines and output associated with fiscal
contractions and what has happened to potential GDP revisions in
response to those contractions and I would say that a summary of that
research is that economic logic is borne out the apparent hysteresis
estimate from a demand side recession is lower than the apparent
hysteresis estimate if you look

this is very interesting. I haven’t read the paper which he cites . I agree with Summers that more better be done on this topic.

I have some thoughts. One is housing again. Housing investment is just as much a part of aggregate demand as investment in plant and equipment. However, it has less to do with labor augmenting technological progress. I think conceptually it is more like durable goods purchases for an extremely extremely durable good. A housing bubble bursting is different from a high-tech bubble bursting.

Another is what about long term effects of periods of extremely high demand (also known as wars). Wars are not caused by good technological progress. This is government spending and X and works better for military spending than for total spending.

Another is what happens to estimates of the natural rate of unemployment. In theory it shouldn’t depend on productivity growth. The original story by Friedman referred to labor market institutions. The old Blanchard and Summers 1986 hysteresis paper was about employment not output. A productivity slowdown might cause a recession (if as argued firms don’t notice and keep on investing a lot, then suddenly cut investment when they realize). It should not cause elevated unemployment a decade later.

Suddenly loose monetary policy. After the stock market crash of 1987, the non independent Bank of England (aka Thatcher) panicked and loosened monetary policy. It turned out that UK unemployment didn’t have to be high after all.

I’d say the practical proposal here is to look at employment and unemployment not GDP.

Briefly, I have a kind of intermediate position on QE and negative interest
rates and stuff. [skip]

I am of a view that is in a sense intermediate. I believe that if the only
game in town is monetary policy, the risks of excessive unemployment
hardening in to structural non-employment and structural output losses
exceed any risks associated with financial instability. And so, when the
only game in town is monetary policy, I tend to be supportive of finding
ways for monetary policy to be expansure (sic).

But I take very seriously, unlike some of my friends who share that view,
the ideas that sufficiently low rates – A, are of questionable efficacy in
stimulating investment; B, the investment they stimulate is likely to be of
dubious quality. After all, think about the investments you would do with
a 1.5% tenure rate but you would not do with a 1% tenure rate, what is
their likely quality to be?

I agree that low interest rates don’t have that much effect on investment really and that they cause investment with a small effect on labor productivity. But really, the story is much simpler than the one Summers tells. Interest rates mainly affect residential investment. It isn’t easy to get a negative term on a real interest rate (achieved expectable whatever) using business investment or non residential fixed capital investment. This makes sense. One hint is that it is one of the many bees in Krugman’s bonnet. He is often right.

I don’t think that even larger houses would cause higher US labor productivity. I think US houses are plenty large enough already. In any case, the claim about the magnitude of the effect of interest rates on investment is testable. I think there is a lot of sloppy thinking (on the web at least) in which the obvious effects of monetary policy in the early 80s (involving changes of interest rates on the order of 1000 basis points) implies that a change of dozens of basis points is reason to get excited and a possible hypothetical change of say around 100 is the key issue.

So anyway, I sure agree with Summers on this one. Unconventional monetary policy is better than nothing but not a decent substitute for fiscal stimulus.

On the other hand, while I think his point about the low quality of investment which is made only if the required return is tiny is valid, actual required returns are much greater than the risk free interest rate. Market risk premiums are high. The one relevant to society is the sum of two terms. The first, subtraction of the expected value of losses due to default is really a correction to expected returns (it appears as part of the risk premium of a corporate bond but it is part of the expected return not a high expected return compensating risk). The second is the contribution to aggregate risk if, say, a bubble bursts and there is a recession. It is possible to estimate the second term (as in Mehra and Prescott). It is tiny compared to actual average returns on risky assets minus the risk free rate. It is very possible that socially desirable investment will only be made if the risk free real rate is negative. This is consistent with dynamic efficiency and everything. I agree with Summers’s conclusion, but he is using a bit of clever rhetoric discussing the interest rate not the many different interest rates.

Summers’s practical practical proposal is to invest more in infrastructure now that interest rates are very low and demand is slack (and would be slacker at normal real interest rates). I totally agree with Summers (and many others including the above mentioned Alessandra Pelloni) and we all know it isn’t going to happen.

I want to talk to you about what is to me, the big puzzle in
all of this, having to do with the fairly compelling evidence of productivity
slowdown and what seems to me to be the equally compelling evidence of
substantial dis-employment effects in the economy.


And yet, if technical change is a major source of dis-employment, it is
hard to see how it could be a major source of dis-employment without also
being a major source of productivity improvement. In part, if the
technology is replacing people that means that productivity should be
expected to go up at least if you measure simple labor productivity. And if
more of that is happening than used to be happening, then you would
expect productivity to be rising more rapidly than it used to be rising.

Summers asked how can there be reduced employment/population and slow productivity growth ? Capital and skill biased technological progress can explain the first, but where is the resulting productivity ? Here Summers doesn’t mention the supply of workers with different levels of formal education. In the past (up to 1973 when the draft was eliminated) there was rapid increase in US educational enrollment. This increase slowed dramatically. This naturally explains both problems & solves both puzzles. A lower rate of human capital accumulation should cause lower productivity growth. A gap between demand for human capital and the low human capital of lots of people can explain why their real wages have stagnated (or fallen) and why fewer of them are employed.

Now for one thing Summers knows about this (he talks to other Larry = Larry Katz). He left it out to make his story more interesting.


I am struck that there is likely what may well be an increase in
unmeasured quality improvement. To take the first example that comes to
mind and I’ll do an experiment with this group. I’ve done this experiment
with other groups – which would you rather have for you and your family,
1980 healthcare at 1980 prices or 2015 healthcare at 2015 prices? How
many people would prefer 1980 healthcare at 1980 prices? How many
people would prefer 2015 healthcare at 2015 prices?
There are a fair number of abstainers but your answer was pretty clear.
What does that mean? That means that healthcare inflation was negative
from 1980. That is very differe

2) miss measurement. I am sure this is an important issue. I very much like the health care spot survey (but his audience is relatively rich). Here the point is that the official price indices are not affected by the introduction of new goods. It is assumed that a new good is a perfect substitute for an old good and people are indifferent between buying it at its first ever price and buying old goods at their current price. This is silly. It matters a lot. One could also say that if a good or service doesn’t exist at all its price is infinite (you can’t buy it for any amount of money). This means that the cost of, say, a hip replacement has declined 100% of the old price = infinity % of the current price. People just couldn’t buy the current average consumption bundle in 1973.

1% a year compounds to a huge amount.

If you think there is mismeasured productivity growth – so if you
think the people who say, you know, there is a line of thought that likes to
say: “Well, GDP growth’s been slow but that doesn’t really mean much
because the labor market is tight and so maybe the GDP growth has been
mismeasured.” If you want to go down that line of argument, you have to
say: “Well the inflation rate is 1 percentage point less than we thought we
are.” And that those raise a question which is if the inflation rate is 1%
less than we are measuring it, why would we be thinking about tightening
monetary policy?

3) so inflation is 1% a year lower — this matters a lot if you care about the 2% inflation target. So ? 2% is silly in any case. There is nothing really special about the inflation rate (the shoe leather costs of reduced real balances at 4% are trivial as they are at 2% while the theoretically ideal rate in an absurd model with full employment and price flexibility is -r %) . Still a good debating point in a good cause.

On miss measurement in general

Summers needs not just that inflation is over estimated but that the difference between measured and true inflation has increased. He doesn’t argue this in his talk. He didn’t ask, for another 30 year interval, if people would prefer 1970 health care at 1970 prices or 1940 healthcare at 1940 prices. I’d say that they’d go for the one which includes penicillin — oh and vaccinations for all sorts of things which were terrible threats in 1940. He is absolutely right that the value of improved health care is huge compared to the measured value (measured at cost). But it is just silly to think that the rate of progress has increased in the USA (look at life expectancy — oh hey he mentioned that).

I see he comes back to this in response to John Fernald’s question. I get his explanation that the miss measurement problem should have increased as the share of sectors where measurement is hard (that is services) has increased. This argument makes a lot of sense. It can be addressed by looking at data disaggregated just at tiny bit. The fact is that measured manufacturing productivity has the same general pattern as overall business productivity (I am using FRED which for some reason has manufacturing only after 1987). Summers (that is a research assistant) can make up constant measurement (of growth rates) error assumptions for each sector and calculated how the shift in GDP composition affects the aggregate measurement error. I am sure he is right about the sign of growth measurement error and the sign of the change in the error. I think part of the puzzle remains.