The drop in potential GDP

Posted on by

Robert Murphy points me to Paul Krugman doing economic analysis.

The analysis concerns potential GDP, which is a concept that plays a big role in a paradigm that I have come to reject, that of aggregate supply and demand. But let’s roll with it, at least for a while. Potential GDP is what GDP would be at full employment, meaning that there is no shortfall in aggregate demand. Another term for potential GDP is long-run aggregate supply.

More below the fold.

Krugman starts with the fact that the Congressional Budget Office estimate of potential GDP in the U.S. today is 11 percent lower than it was ten years ago. This says that aggregate supply has underperformed in the past ten years.

Algebraically, potential GDP can be thought of as labor input times productivity. Even though we are at full employment, labor input is a bit lower than what was expected 10 years ago, because labor force participation is a bit lower than was expected back then. But both Krugman and I read the data as saying that the shortfall in labor force participation is small relative to the shortfall in potential GDP. That leaves unexpectedly low productivity growth as the main source of the shortfall.

Although Krugman does not use the term “productivity,” he clearly is looking for explanations of a productivity growth shortfall. He is most attracted to the idea that low aggregate demand has an adverse effect on productivity. But the connection between the two is a black box. I can think of possible stories–less investment holding down the capital stock, periods of unemployment holding down the human capital of workers–but it would call for larger quantitative effects than seem plausible.

Getting away from the AS-AD paradigm, I would raise two issues: re-allocation of resources; and possible mis-measurement.

The AS-AD paradigm says that all economic activity takes place within a single GDP factory, and it treats unemployment as if it were a temporary layoff from this GDP factory. Instead, what I once called the recalculation problem takes the view that layoffs are not temporary. They are permanent job losses at declining firms that need to be offset by job gains at expanding firms. When the job gains fail to offset the job losses, we call that a recession.

I think that the last ten years fit much better with the recalculation paradigm than with the GDP factory paradigm. Now that we seem close to full employment, how many workers are back at the same jobs that were lost in 2008-2010? I would venture a guess that it’s not very many.

The last ten years have seen a significant reallocation of labor. Not necessarily larger than the reallocations that take place during years without recessions, but certainly far larger than any layoff-recall process that took place.

This reallocation could have reduced productivity growth, for a number of reasons. Perhaps there was a net loss in human capital, as older workers with a lot of experience in, say, construction, aged out of the labor force and young workers with not much experience found jobs in, say, hospital administration. The former were producing measurable output, the latter not so much.

That leads me to the measurement problems. To describe the economy as a GDP factory, we have to add up health care services, streaming videos, and bushels of wheat as if they were all bushels of wheat. This means that statisticians have to come up with a bushels-of-wheat equivalent for health care services and streaming videos. Is the challenge in doing this plausibly large enough to be 10 percent of GDP or more? I think that it is. But others disagree. Chad Svyerson says,

the mismeasurement story, while plausible on its face, falls apart when examined. If productivity growth had actually been 1.5 percent greater than it has been measured since the mid-2000s, U.S. gross domestic product (GDP) would be conservatively $4 trillion higher than it is, or about $12,000 more per capita. So if you go with the mismeasurement story, that’s the sort of number you’re talking about and there are several reasons to believe you can’t account for it.

First, the productivity slowdown has happened all over world. When you look at the 30 Organization for Economic Co-operation and Development countries we have data for, there’s no relationship between the size of the measured slowdown and how important IT-related goods — which most people think are the primary source of mismeasurement — are to a country’s economy.

Second, people have tried to measure the value of IT-related goods. The largest estimate is about $900 billion in the United States. That doesn’t get you even a quarter of the way toward that $4 trillion.

Third, the value added of the IT-related sector has grown by about $750 billion, adjusting for inflation, since the mid-2000s. The mismeasure­ment hypothesis says that there are $4 trillion missing on top of that. So the question is: Do we think we’re only getting $1 out of every $6 of activity there? That’s a lot of mismeasurement.

Those are strong arguments. But I still think that coming up with bushels-of-wheat equivalents for what most people do at work nowadays leaves a lot of room for measurement error.

22 thoughts on “The drop in potential GDP”

  1. Interesting post.

    There is the problem of productivity declining globally. That suggests a universal cause.

    That leads us to weak demand. Companies that face weak demand have less incentive to upgrade plant and equipment. Maybe labor is relatively cheap too.

    Also, you get labor productivity boost just from more volume, if fixed costs are a large part of factory costs. So I think we will start to see some upticks in productivity in H2 2018.

    Certainly, the 1990s saw good productivity gains, and lots of demand. That was not ancient history. In fact, it was a great decade.

    So what about China?

    “China’s Labour Productivity improved by 6.85 % YoY in Dec 2017, compared with a growth of 6.49 % in the previous year. China’s Labour Productivity Growth data is updated yearly, available from Dec 1953 to Dec 2017, averaging at 7.45 %.”

    Okay, China is China, and not the West, and an emerging economy and so on. Yet again we see big growth and productivity gains.

    https://www.ceicdata.com/en/indicator/china/labour-productivity-growth

    I advise the Fed let it rip. Shoot for full-tilt boogie boom times in Fat City. After a decade or so of that, maybe cool things off slowly, very slowly.

    When did “boom times” become dirty words?

  2. Productivity growth reportedly averaged 3% annually from 1947 to 1973, and half that from 1973 to 1995.

    The 1970s were the dawn of the era of government regulation. Since then, we are increasingly using part of our workday complying with regulations, rather than producing goods and services.

    My explanation is so obvious that it must be wrong. Someone please help me understand what I am missing. Thanks.

    • Yeah I think that regulation is likely a big part of the answer. That and there has been huge inflation in non-market sectors like education and health care. Real estate (is construction) costs are also increased through government restriction of real estate supply.

      I wonder if “older narrower deeper” applies as well. Many jobs are now “executive” jobs that involve deep thinking and specializion and experiences. Fewer and older workers are qualifies for that kind of work. Also it is much easier to manage fewer employees who are just as productive. Scaling by adding workers isn’t as obviously better in the days of the “mythical man month”

    • Regulation, especially the government levies on labor. Our government safety net is paid with labor wage. Just paying for safety net expenditures in the pst has resulted in some 600 billion in interest charges at the Swamp. about 3% of our labor just for interest charges from the past bailouts.

  3. To be cynical, perhaps the slowdown has something to do with the fact that more and more young people are spending more and more time in school, and not getting much of use out of it. In fact, perhaps some of it is anti-investment if it tells them that you are a failure if you take a blue-collar job and that if you can’t get a good job, it is the fault of an unjust American society.

  4. Real gross value added from housing is about a half trillion below where it would have been if long term trends had continued. Some potential is lost in addition to that because of the productivity gains from migration that have been lost. Some productivity growth in 2005 was coming from moving hundreds of thousands of folks out of the productive cities each year to make room for productive workers where population growth is limited by housing.

  5. The long run trend in the growth of real net private capital stock has slowed sharply.

    From 1959 to 1980 it average3d 3.3%.
    from 1980 to 2008 it averaged 2.5%.
    From 2OO8 to 2017 it averaged only 1.4%.

    If you lag the growth in capital stock one year and compare it to productivity growth you get an amazingly strong correlation, especially if you do one more step and use the growth of real net capital stock per private employee.

    This makes a strong argument that the slowdown in productivity it due largely to weak capital spending and in particular the rapid depreciation of IT equipment.

    In the old days if a worker shifted from the farm to manufacturing he moved from a low capital, weak productivity to a high capital and high productivity job. But in recent years as workers shifted from high productivity manufacturing jobs to low productivity service jobs the net gain in capital intensity and productivity fell sharply.

    I have real problems seeing how we get out from under this new trend.

      • Population growth in the US is about 30% lower now than it was during the 70s, why would we expect residential investment to stay the same % of GDP with different demographics?

        • There might be something to this. But much of the fall in population growth is post-crisis and this is causally dense. A lack of homes and construction jobs has something to do with that.

          Also, if my work changes the national set of presumptions from “we misallocated trillions of dollars in over investment in residential construction in the 2000s” to “sure we have been reducing residential investment over time, but there are good reasons for that.” that would be fine. That’s an improvement.

          The baseline I am arguing against has been unequivocal and extreme. I’d love to see the goal posts move as long as you’re not just doing a motte and bailey move.

  6. My “Handle-Baumol Jobs” story about technology and labor productivity is consistent with it being a global phenomenon for all developed countries. US-specific explanations like local Aggregate Demand, Housing Construction (Kevin Erdmann call your office), Regulations, or even Mismeasurement to some degree (since other countries have different consumption baskets which would probably reveal more variation) can’t account for widespread and seemingly synchronized stagnation.

    Instead, it’s much more reasonable to conclude that the past guesses about the future path of productivity growth were simply wrong because based on bad assumptions and models which are proving to have been false. As one of the clearer false assumptions, natality and population growth is lower (and much lower in some countries like South Korea) than experts anticipated even a decade ago. But that only tells us about the aggregate, not about per-capita productivity rates, at least not directly.

    I’ll try to briefly recap the story. I’ll be focusing on a “real output per man-hour” conception of labor productivity for ease of illustration, though it’s possible to extend the analysis to nominal / welfare frameworks.

    Think about the classic Baumol cost-disease jobs don’t increases in labor productivity. We have the classic example of the number of musicians needed to play a Beethoven string quartet, and the time they need to play it. Obviously that applies to live performances in general, and certain kind of educational lectures are like live performances. Nurses don’t change a bandage any faster than they did fifty years ago. Haircuts take about the same time. Manhours per restaurant meal seem about the same. An hour massage takes an hour. If the vast majority of commercial air travel takes place in planes that are no larger or faster than planes 50 years, then a pilot takes the same number of manhours to produce passenger-miles as his remote predecessors (and one can reason similarly for the transportation sector in general once they hit important physical constraints in size and speed so long as the platforms require ‘drivers’).

    Now, at this point we should ask what these tasks – and note they are mostly service jobs – have in common. What is it about them that makes it such that they don’t experience labor productivity growth?

    In general, the reason is because “average humans can’t go faster than average humans.” If all the low-hanging fruit of process and technique improvements has been picked, then any part of the output which depends on human action or when man-hour use is a fundamental part of the consumption-experience can’t get much faster when it can’t be economically augmented by technological capital either by complementing labor or, what usually involves a more radical transformation of the way the output is created and experienced, substituting for it.

    For example, YouTube could deliver the audio of that quartet for millions or billions of listening experiences for near zero marginal cost and maybe the equivalent of a single minute of human labor input (which gives rise to winner-take-all dynamics), but we don’t usually count that as the “same” kind of experience or labor-input process at all. Same goes for MOOCs.

    So we see that some jobs and tasks have this issue as an inherent factor of the particular kind of human action and output they involve. If there is to be growth in labor productivity throughout the economy, it will have to take place in other jobs. If there are lots of those other kinds of jobs with plenty of room for improvement (RFI) – as one saw historically during the expansion of the manufacturing workforce – then on average across the whole economy there will be fast productivity growth.

    But if most jobs are Baumol-type jobs, then the few RFI jobs have a huge burden to carry if there is to be high productivity growth.

    And what is happening is that most of the labor force in developed countries has shifted – and continues to shift – into Baumol-type service jobs.

    Why? Because in economic conditions like ours, when certain sectors become more productive, they don’t increase output more than they increase productivity (because additional demand would be insufficient to pay for the necessary amount of scarce inputs), and so shed redundant workers. That excess labor is necessarily displaced to fields with lower productivity growth.

    And now one has to consider the labor force reallocation process in response to the radical changes brought about by automation, globalization, and extremely cheap transportation and communication, to include internet services with extreme economies of scale and which can scale up to serve the entire human population at practically zero marginal cost and with a tiny amount of human labor.

    In general, the cost of labor in developed countries is very high (not just because of compensation, but also taxes, regulations, etc.) To the extent sectors can arbitrage and replace that costly labor with cheaper alternatives they will. If a good or service can be done abroad, it will be outsourced. If the worker can be replaced with a machine economically, he will be. If a service can be provided digitally at low cost to everyone, many people will take the free option, and if you add up all the employees of the entire software sector together, it’s still a tiny portion of the overall labor force, and, at any rate, a lot of people can’t “learn to code”. Even the production of goods which has to be done locally like agriculture or some manufacturing requires a decreasing portion of the labor force because increasingly augmented with robots, and productivity increases that affect a tiny portion of the labor force get diulted out of the average signal when combined with the full workforce overall.

    So the question is, with all this replacement, arbitrage, outsourcing, and substitution going on, if laborers are by necessity going to reallocate themselves and find some other tast to perform, then what jobs are left for most people to do?

    The jobs that are left are necessarily the tasks which are inherently very hard to shift to cheaper or alternative sources – that by nature resist outsourcing and automation.

    The only kind of job that can resist outsourcing, and which has the potential to employ a large portion of the overall labor force, is one that depends on economies of agglomeration, that is, dealing with other humans nearby. And jobs that resist automation are tasks only humans can perform, and humans can only perform human tasks at human speeds.

    One sees right away that most of these jobs are necessary Baumol cost disease jobs for which there is very little opportunity to improve labor productivity at all.

    So, because so many people – and increasing numbers of us all the time – are working in Baumol jobs where the ratio of output to manhour inputs doesn’t change much from year to year, overall labor productivity stagnates compared to what happened in the past, and what productivity growth we do observe is due to the contribution of a few sectors which only employ a small and decreasing fraction of the labor force.

    • Handle is mostly nailing it, with a story consistent with Arnold’s reallocation story. The human services, like limo drivers & truck drivers, don’t see much productivity improvements. Until fully automated.

      Kevin Erdman is also right about housing — ending house construction is a big reduction in capital formation. He doesn’t quite draw a difference between an existing Palo Alto house being sold after an increase in value from $1 mil to $1.5 million, and the building of 2 new houses in Fresno at $250k each. I’d say the building of new houses is “real” capital investment, while the increasing value is some kind of non-productive scarcity investment. I don’t think there was really $1 million of capital investment, tho, only the $500k from the two new houses. Too much “housing investment” is that scarcity investment, not actual new capital.

      Also, Kevin was MUCH better in having longer time frames in his graphs. Krugman is silly to extend from the 2006-2008 crash 10 years later, but only go back to 2007, Fatas 2004, 2007, (FRED unemployment) 1990, (FRED labor force) 2000 (labor force should be used instead of unemployment going forward).

      “we’ve returned to sort-of full employment at a much lower level of real GDP than informed people projected we’d reach before the crisis struck.” – Krugman

      I’m sure regulations have been partly responsible, and that Trump’s deregulating is helping — none know how much. But obviously, effort to comply with regulations specifically is “non-productive” as compared to effort to produce the service or good. If truck drivers also have to fill out more forms, it reduces their time driving trucks to get the same of truckloads.

      I’m pretty sure Roger Sweeny also has a good point about more young folk spending time “non-productively” in go-nowhere schooling plus videos (plus opioids?).

      Finally, my own point – reduced illegals. After the house construction crash starting 2006, illegal aliens unable to find construction work left the US, maybe 5-6 million. Since they were illegal, their work was falsely attributed to others, the legal workers. But it was some 1-2% of the manual value added, with the built houses visible and counted, but the illegal persons not so much.

      Measuring and accounting for illegals is a significant part of the measurement problem.

      Plus, as they leave, the knock on effects of their leaving, low cost living areas, low cost restaurants, low cost services, all reduced demand. Their leaving would be similar to an ebola outbreak killing 5 million US workers — except
      these would have been counted workers where the legal employed numbers are better measured.

    • Handle,

      Thank you for an excellent comment!

      Your comment explains a lot and quite clearly I might add. (Also, I’m starting to follow your blog.)

    • It’s not fair to dismiss Regulations as an explanation because the productivity slowdown has occurred in “all developed countries”. Regulation has been increasing in all developed countries over the relevant timeframe, so could still be a cause.

      It’s also wrong to say it still takes a nurse the same amount of time to replace a bandage. Although I’m not familiar with hospital procedures, I think hospital regulation and reporting to various governmental agencies has increased substantially since the 1960s. Although it takes the same amount of time to change the bandage, the number of people working in administrative jobs to support the bandage change has vastly increased.

      Even a one hour massage now is more likely to have attendant duties like tax reporting, OSHA and Labor and Industries compliance and reporting, Employment Security Department compliance and reporting, more time spent understanding, reporting to, and paying income taxes, PPACA research, compliance, and reporting, understanding potential environmental restrictions and taxes (how do I dispose of unused massage oil?).

      The point is that although a one hour massage still only takes one hour, the administrative overhead required to deliver that massage has reduced the number of one hour massages that can be delivered during a 40 hour period.

      There is a lot of merit to the Baumol-Handle model. It does not dismiss the possibility that regulation also plays a large part in the productivity slowdown.

    • Anyone who has aging parents or old friends knows that nursing homes and hospitals are staffed largely by people who do “tasks only humans can perform, and humans can only perform human tasks at human speeds.” They have both been growth industries recently, and demographics suggests they will continue to be.

      The implications for aggregate productivity growth are scary.

    • This goes back to the measurement problem, but it strikes me that there are a lot of services that could improve their productivity not by reducing inputs, but instead by increasing the quality of their outputs. One example, from the ACA, is the government’s efforts to reduce unnecessary deaths from hospitalizations caused by central line infections. I forget the stats, but the reimbursement mechanism worked at reducing preventable deaths by tens of thousands of people a year, without really changing how much money was being spent on health care. Hospitals just hadn’t adopted cheap to implement best practices yet, and the government gave them a sufficient incentive to do so. I could see the same sort of quality effects being possible in education as well, and healthcare and education make up a decent size of the US economy. An example from education, my daughter is in kindergarten, and she already knows how to write all of the letters, and isn’t too far away from being able to read. I didn’t really learn to read until 2nd grade. I think part of this is that in 30 years time there have been improvements in curriculum and pedagogy (the whole “Letterlander” thing where each letter has a name and a story is brilliant, IMO). So I think that there is more room for productivity improvement than Handle is accounting for. However, those productivity improvements would have to happen in highly regulated industries.

  7. I like Handle’s commentary, but that still does not explain why we see a decrease in productivity even within the manufacturing sector.

    Erdmann is irrefutable.

        • It doesn’t matter at a national level. If your new units are causing vacancies in old units in other cities then the productivity rate of the old investments will be in decline. Net productivity of all housing units is going to be heavily driven by the vacancy rate, and you can’t take the value of new housing for the country at its cost or sale price alone.

    • Some day when he’s near DC, I’m going to have to invite Kevin Erdmann over for dinner to hash out our perspectives on what’s really going on with housing.

      I think my general approach is pretty reasonable. When one is puzzling over some economic phenomenon, first look around at what’s going on in other comparable countries and check if the phenomenon is also going on in a lot of those other places. If it is, then the important, core explanation most likely isn’t a special, local matter.

      That’s not to say that local factors can’t contribute, but that contribution would better be described as ‘exacerbation’ than the ‘explanation’. Because then one would have to go find similar local factors in all those other countries, and the odds that one is dealing with a coincidence of simultaneous local effects all pushing the same way are very low.

      The long-term average decline in manufacturing “real output per hour” growth rates (which are very noisy and volatile series that are best to assess with caution) seems to be happening in most large developed countries with important, and sophisticated manufacturing sectors, i.e., U.S., Canada, Japan, Germany, South Korea, etc. Unfortunately the BLS discontinued some of these series in 2012, which was particularly bad timing, but it’s possible to get some of the data elsewhere.

      Canada in particular makes for a good example of this. Output per hour tracks that in the US very closely. Yet the experiences of the Canadian housing and real estate finance sectors were entirely different from what occurred in America. It would be bizarre to claim that the vagaries of the American housing sector somehow caused Canada’s manufacturing productivity growth rate to change in the same way. Yes, Canada’s economy is obviously closely tied to America’s and they catch our colds when we get sick, but it’s much more likely that some common factor is driving what is going on in both countries, especially when one starts throwing in yet more countries more distant from North America.

      Now, there are a lot of guesses as to what may be going on. One is that we’ve already picked all the low hanging fruit in a lot of these industries. Some of these places are so automated already that the joke is that the factory only needs a man and a dog, the dog to guard the factory, and the man to feed the dog.

      Related to that guess is that even manufacturing often runs into ‘Moravec’s Paradox’ tasks that are very easy for human workers but terribly difficult to automate, and will likely require additional and challenging major advances toward Strong AI and dextrous manipulation devices.

      Consider the challenges Elon Musk and Tesla has faced with automating their production line (which are, in reality, common industry-wide, since all the other automakers face similar incentives to maximize automation). It turned out that there are a lot of tiny things that vary and can change and which humans just naturally adjust to or automatically figure out, but which are hard to get machines to recognize and treat appropriately. An example I came across is that if a nut falls off a feeder onto the factory floor, the robot can’t do anything and stops the whole line, but a human just picks it up and carries on and doesn’t really think anything of it, and certainly doesn’t report it as some kind of potential process issue. It’s easy to underestimate just how much of that kind of thing goes on in any complex process with lots of human inputs at various stages. Once one has automated the easy stuff, one starts to run into the hard stuff like that.

      Indeed, many industries are now using their line labor force as ‘machine helpers’ (not exactly Cowen’s freestyle chess man-machine teams, but analogous) when the particular task is hard to automate. Even Amazon with all its robots and software geniuses still hires huge numbers of gift-wrappers and “pickers and packers” because human hands and minds are still just much better at that kind of thing, and often the computers just tell the humans what to do, e.g., “Grab the size-3 box,” or, “Paper jam, remove tray two.”

      Another explanation for some products is that they are so capital intensive already and non-labor inputs already represent such a large fraction of the cost that there’s just no important incentive to try and economize on labor and research ways to switch out tools to increase labor productivity, even if it’s technologically feasible.

      Related to this is an interesting problem with “bottlenecks” in multi-step processes where one step could technically be sped up considerably even though it is already only using a small amount of labor, but subsequent steps can’t absorb output at that pace, perhaps limited by human speeds, and so when one optimizes across various capital-labor alternatives for that Room-For-Improvement step, one still ends up stuck with a lower capital-to-labor ratio than what is technically possible and stagnant productivity.

      • Handle, I don’t disagree with what you’re saying. Clearly, in education and health care, and other sectors, there are issues. All of these things can be in effect. I wouldn’t claim housing directly is responsible for all of it, or even most of it.

        But, one way housing has an indirect role that is in effect in many countries is that it limits access. What has happened is that current tech and culture demand a new wave of urbanization which postmodern political culture is unwilling to allow. Consider the amount of venture capital that goes to places like San Francisco because it has a monopoly on certain innovative labor networks. Most of the cost of those innovations goes into human capital these days. And it costs a lot more for human capital to locate there. So, even though housing is just a background factor there, it is the primary reason why there is a sort of tax on certain kinds of frontier innovations. In addition to the higher costs, there is a lot of deadweight loss because we have capped the size of innovative centers.

Comments are closed.