Tuesday, November 03, 2015

Big TFP data mystery! (Probably solved!)

NOTE: Mystery probably resolved! See update below. Here was the original post, for posterity:

While recently complaining about the overselling of static-efficiency policies, I asserted that rich countries have all grown at about the same long-term rate, despite decade-long divergences. I was talking, of course, about Total Factor Productivity, which at long horizons should be determined by technology.

I had been under the impression that over the last three decades or so, the rich countries had all experienced similar rates of TFP growth. My source for that was the OECD's time-series on multifactor productivity (another name for TFP). Here is a chart of those OECD productivity numbers since 1985:

As you can see, most rich countries grew their TFP at the same average rate, consistent with the idea that TFP mostly measures technology in the long term, and that technology spreads rather easily between rich countries. A few countries, like Korea, Ireland, and Finland, did much better over this period, and a few countries, like Italy, Spain, and Portugal, lagged behind. But most rich countries were clustered along the same basic line. The U.S., UK, France, and Germany (highlighted on the graph) all stayed very close to each other.

But I now see that FRED has its own TFP numbers for various countries, taken from the Penn World Tables. And here's what happens when I plot the TFP numbers for the U.S., UK, France, and Germany over the same time period (1985-2011):


The U.S. and UK lines match up as before, but the Germany and France lines are wildly, totally different! In fact, according to the Penn World Tables, Germany's TFP actually steadily declined from the mid-80s to 2011! 

What on Earth is going on here?? Obviously the two measurement methodologies are very different. So I tried to track down the source of the discrepancy, and I found some interesting stuff. 

First of all, it turns out that the Penn World Tables, currently assembled by a team of economists from UC Davis and the University of Groningen, have undergone substantial revisions to their methodology in recent years. They switched to a new growth accounting method developed by Francesco Caselli in the early 2000s (which I plan to study in detail when I get the chance). As Antonio, and Marek JarociƄski pointed out in 2010, these revisions were enough to substantially change the results of all cross-country growth regressions. Simon Johnson, William Larson, Chris Papageorgiou, Arvind Subramanian criticized the new Penn methodology, and suggested possible changes. 

Meanwhile, the OECD methodology for calculating TFP has some questions surrounding it as well. To get TFP you need measures of labor and capital inputs. The OECD uses a pretty textbook method for doing this - simply stick in the raw estimates for the dollar values of labor and capital. But when they tried using another database called EU-KLEMS that tries to adjust for "quality" of inputs, they found totally different numbers.

I am not experienced enough in growth accounting to wade into these disputes in a substantive manner; it would take me at least a month of serious study to be able to say with any confidence which of these methodologies I believe most. The real takeaway here, though, is that TFP measurements are HIGHLY suspect, and will continue to be so for the foreseeable future.

That is bad news for most of modern macroeconomics, both on the growth theory and on the business cycle theory side of things. If differing methodologies for measuring labor and capital inputs diverge by this much, it means that any series you use probably has tons of stuff in it that it shouldn't have. That means that changes in the series at business-cycle frequencies - the good old TFP shocks of RBC models, which are also part of "kitchen sink" DSGE models like Smets-Wouters - are also unreliable. Basically, all those "shocks" are as likely as not to just be noise. That's probably true whether you compare across countries or look only at one country.

So this is a very pessimistic finding, and a huge challenge for the growth accounting field. Hopefully, a meeting of the brightest minds will get to the bottom of the problem and arrive at a consensus solution. If not, it means that any model that relies on measures of aggregate TFP, or factor inputs in general, is unreliable until the accounting problems are worked out.


Robert Inklaar of the University of Groningen contacted me and explained what was wrong! The most recent version of the Penn World Tables, version 8, did not take into account changes in averaged hours worked in some countries. Also, it used a Barro-Lee data source that apparently had some questionable data on trends in education. Inklaar says that the next version of the PWT, version 9, will fix the problems, and until it comes out, to use OECD data.

Well, I am mostly relieved. It's not really a methodology disagreement (except for the Barro-Lee education data). All of macro does not have to be scrapped, just yet. :-)

Thanks to Robert Inklaar for helping me out!

...But the growth economists I talked to about this mystery all expressed deep skepticism about these TFP data sets in general...


  1. It's very simple. TFP is a residual, not a factor.

    1. Krzys wins the thread. And does the fossil-fuel consumption boom of 1945-1978 count as a boom in TFP or capital? I've never gotten a clear answer to this, but then again, I haven't asked any professional.

    2. You dorks. Obviously it's a residual. The problem is how to measure labor and capital.

    3. Labor's easy enough to measure (hours, employment, maybe IQ). I've never understood by what "capital" is supposed to mean in the Solow model. Did the USSR have more "capital" per worker than the U.S. in 1985?

    4. Well, pithom, you could actually try looking these things up...that's why God created Google!

    5. If the residual jumps around, then the inputs are not well defined, and probably shouldn't be. In other words, aggregation does not work for growth accounting. Give it up.

    6. That may be the only conclusion. But let's not jump to that conclusion...

    7. No, it's done. It is not physics. You cannot get another data set to resolve the noise with a better theory. You can only overfit. You, guys, are done.

    8. Of course it's not physics, but what does that mean? Nothing except physics is physics. Microeconomics has a ton of provable, impressive successes that you probably don't even know about, because you probably don't care enough to look into the matter. I agree that macro is screwed, but I think you should be a little better-informed before you agree with me! ;-)

    9. But even if aggregation "does not work", we can still draw useful conclusions from TFP movements. Suppose two countries have similar levels of capital and labor, but one country has higher GDP and therefore higher TFP. Maybe this is some aggregate productivity shock or maybe this is institutional. Maybe one country is better at knowing in which industries to build capital or maybe one has better labor search markets and has more efficiently sorted workers. Whatever the underlying reason for the TFP difference, it is still useful to know that one country somehow seems to be better at using the capital and labor it has than is the other country.

    10. It's like with God, Multiverse and the true nature of macroeconomy. You will never know. The only thing, you can know about macro is a couple of simple heuristics. That's it.

    11. By the way, I have nothing against micro. I used to do mechanism design long time ago and it was cool, and more, importantly meaningful. The data is there to resolve things.

  2. I looked at some studies once, but they had contradictory methods of measuring capital.

  3. Also, this is Google. Blogger is a part of Google, which is a part of Alphabet.

  4. Your are not the only one puzzling about it. http://www.wsj.com/articles/alan-blinder-the-unsettling-mystery-of-productivity-1416873038



    One explanation that might make for interesting research is to see if a prolong period of declining real wages decreases the incentive to invest in business and train workers to make them more productive. To quote the voxeu article:

    "Lower real wages seem to have helped reduce lay-offs because they have made it cheaper for firms to hang on to their workers even in the face of falling demand. But low wages also made it relatively cheaper for firms to take on more workers rather than invest in new capital or technologies."

    See also falling labor share and slower productivity growth. http://angrybearblog.com/?s=productivity

    1. I thought low real wages was what led to the Chinese taking "our jobs" and rapidly growing their economy. And that Naked Keynesianism link is ridiculous, for multiple reasons. The USSR also grew fast after the war, as did Portugal, South Africa, Sweden, Denmark, and Mexico. The post-war boom was mostly about energy consumption.

      Manufacturing productivity is especially worrying. It's been virtually flat since 2011. Never has this happened before.

    2. Thanks, rickstersherpa!!

  5. You could always be a dick about it and get him a cheaper, lower quality beer from a low TFP country...

    1. Anonymous12:48 PM

      I get it. Unfortunately, as a group economists often advocate policies they can't back up. They deserve the scorn. If we want to make this personal I could certainly start digging up dirt, but that isn't in my character. Sticking up for those that can't is worth all of the thoughtless insults you could hurl my way.

    2. If you are so down on economists, you might want to consider finding another blog to patronize.

      You also commented in the wrong comment thread.

      Unless of course you are Russ Robertson, which seems unlikely.

    3. Anonymous8:34 PM

      John, I started this thread.

      You have no idea wheat what you're talking about.

    4. Anonymous8:48 PM

      I started this discussion. If you don't know that I did it, you're fucked up or misinformed.

      I know personally the people on the opposite side, and I'm going to defect because I don't like you people.

      Give me a reason to like you…

      Your'e fuckin history…

      I'm joining the opposition.

    5. If you are David Pechacek below, see below.

      Otherwise I can't be expected to sort one Anonymous from another.

      I also have absolutely no idea what you are talking about in your claims to have started the discussion, since you don't appear to be Noah Smith, the author of this blog.

      Insofar as I can tell you are making a completely irrelevant and hysterical response to an extremely bad joke I made above.

  6. I've looked at the PWT TFP numbers before and decided they made little sense...

  7. Cool post. I've wondered about the underlying numbers on TFP, and what actually goes into them. But I'm not sure what it says about the field if you already fear political motivations for the differences between the two measures prior to digging into the details.

  8. Anonymous10:44 AM

    This is my second try at posting a comment. On the first try I accidentally hit the 'Sign out' button below instead of the 'Publish' button, and there was no confirmation screen to decline. Yuck!

    Anyway, my point is that the US uses deliberately deceptive methods of measuring labor inputs due to foreign labor, and I would suspect the UK follows suit. The US doesn't even measure much of the foreign labor inputs to US production as labor, rather they measure it as capital inputs. While this wouldn't change the TFP calculation, it would change the perception that TFP measures technology gains.

    Note that this accounting problem doesn't affect all trade relations, but it does affect the relations with countries of low wages compared to the US. The effect here depends upon the source of commodities and the destination of finished goods.

    France and Germany are not nearly as dependent upon foreign labor as the US and the UK are, so this should be the first place to look for more specific insight.

  9. I'll admit that the only other time I've heard about a debate on Total Factor Productivity calculation has been in relation to the book "Time on the Cross", which makes the (extremely difficult for me to accept) assertion that slave labor was more economically efficient than free labor in mid-19th century America.


    One has to ask "Then where were the great slave run factories, and why was slave raised wheat and corn not driving Northern produce out of the market with its cheapness?"

    1. Anonymous12:59 PM

      "Then where were the great slave run factories, and why was slave raised wheat and corn not driving Northern produce out of the market with its cheapness?"

      I doubt than an accurate measure of TFP could really be made with regard to slavery. I don't doubt some of the assertions of the book as described on WikiP, but slavery is morally wrong and the slaves were brought against their will. Everything else is moot.

      Is it correct? It might possibly be, and why would we have a predisposed expectation that because slavery is morally wrong that it should be unproductive or unprofitable? I wish economics were inherently fair this way but they aren't.

      It is a common mistake in my opinion, to look for market-based reasons why basic human dignity, morality and rights are protected by market forces. This is the essence of free-market fundamentalism, and it is a scary movement that invades both sides of our economic debate.

    2. You're confusing "profitable" with "maximally efficient". More generally, you're confusing "made slaveholders richer" with "made the country as a whole richer".

      Nobody's pretending that slavery was not profitable for the slaveowners. The general consensus is that set back economic development in the South by decades or centuries.

      What the authors of the book I cited were saying was that slave labor was more efficient than free labor in the US, and that's utter balderdash, because had that actually been the case, market forces would have made it impossible to be anything but a slave or slaveowner.

      The basic microeconomic theory behind this is that there are two variables in determining labor productivity, the cost of the wage, or in the case of slaves, cost of food, shelter, etc., and the amount of labor the worker produced, which is partly a function of morale. (Conservatives like to pretend the latter factor doesn't exist.)

      As Adam Smith put it:

      "From the experience of all ages and nations, I believe, that the work done by free men comes cheaper in the end than the work performed by slaves. Whatever work he does, beyond what is sufficient to purchase his own maintenance, can be squeezed out of him by violence only, and not by any interest of his own."

      You get what you pay for, in other words.

      Slavery was only adopted because the cost of hiring workers in the functionally empty New World (once the Native Americans had succumbed to disease) was absolutely exorbitant.

      You are correct insofar as it goes that the moral questions are paramount over economic arguments, but I disagree that this is a reason not to make economic arguments.

      I also resent the unfounded accusation of market fundamentalism.

    3. Anonymous2:19 PM

      Resent as you will.

      I can flip your argument around and say that if Northern productivity was superior to Southern productivity that Southern plantations should have been economically unviable. I don't think you considered this fully.

      The book (which will definitely not read) appears to mention some pretty big differences between Northern and Southern conditions that could explain some of the differences in productivity of slave vs. paid labor.

      I think the most important thing to take away is that slave labor can still be as productive as other forms of production and they can compete just fine.

      Adam Smith was largely a philosopher, and your quote only suggests he was largely a free market fundamentalist in that he believed the free markets must provide that slave labor is less productive than paid labor. I strongly disagree.

      The implication, which many economists wish to avoid facing, is that markets don't provide any guarantee of fairness or morality regardless of competition.

      The key is to understand the symmetry in flipping your first argument about Northern vs. Southern agriculture.

    4. Yeah, sure, whatever.

      Believe what you want irrespective of academic consensus.

      I do not believe that markets provide any guarantee of morality or fairness.

      Thanks for torturing my statements into an apparent affirmation of that.

      Thank you also for derailing an attempt to start a discussion on historic miscalculations of TFP to self-indulgently preach a sermon on morality, which I note, you are not willing to attach your name to.

      Keep on boldly speaking truth to power there, micro-Chomsky.

    5. I was going to respond with contempt, but I thought otherwise.

      What is 'academic consensus' and why does it matter to reality?

      I would like to have a more lengthy discussion on this, and unfortunately, there is no appropriate forum for this.

      I have no idea what that last statement is about. I know who Chomsky is, and I agree with some of his assessments. Not all, but some. You've obviously read him, so I ask, to continue the conversation, what is your take on his viewpoints?

    6. And John, I was the one who started this conversation in the first place.

      I started the conversation on the blogs about TFP.

      Krugman responded and even got Solow involved. They're responding to what I said, and while I find it somewhat validating, I also find it alienating.

      I will never be part of this group. I can only say what I think. Why did I get involved in the first place? I was concerned. Perhaps I had something to add that few others understood. I'm not sure anymore.

      You're impenetrable as a group. Why did I try?

    7. Plantations were the factories of the day. Most plantations used upwards of 100 slaves working in a division of labor system where each group of 10 or so slaves performed a different task. The jobs were arranged in such a way that speed and precision mattered so that one group was always pushing the next group to get their job so the second group could do theirs. It resembled an assembly line.

      This system only leant itself to a few crops. It isn't that slave labor wasn't tried with corn and wheat. It's that it wasn't any better than using free hired help. The best crop for slave labor was sugar cane. Cotton wad second. Indigo, rice, and tobacco were marginally more profitable with slave labor.

    8. Yeah, so I lost my temper.

      It happens.

      I apologize for the tone of the above, but not necessarily the content.

      I'm not entirely sure how many of the posts up and down the comment chain are cases where Anonymous equates to David Pechacek, but I'll try and explain my exasperation with you in this comment because, frankly, I don't want to engage with you any more.

      To the extent that there's an agreement that I know of in the economics field on the microecon of slavery as a labor source, it's reflected here in this very old Paul Krugman article from before the Times.


      The relevant passage is: "Indeed, my old teacher Evsey Domar wrote a classic paper arguing that serfdom and slavery often were the response of ruling classes to labor shortage. Thus Russia imposed strict restrictions on the mobility of peasants from the late 16th century onward, precisely because the availability of rich new farmland on the expanding frontier would otherwise have increased their bargaining power against landlords. (The frontier expansion itself was probably the result of the growing effectiveness of firearms, which turned the military balance against the steppe nomads; so gunpowder, which contributed to the rise of the bourgeoisie in the West, created serfdom in the East). And, of course, slavery - which, along with serfdom, had died out in the West during the labor-abundant high Middle Ages - was reintroduced in the land-abundant conditions of the New World."

      As Ben Wheeler notes above, the slavery system only lent itself to a few cash crops with a high dollar value per acre. Once labor got more plentiful in America though, that system is likely to have become less and less profitable compared to hired hands, and died slowly over the course of urbanization as serfdom did in the high middle ages. The reason there were no plantations in the North was the rapidly increasing population density driving down the cost of labor there. (And the climate was unsuited to Cotton, Rice, Sugar and Tobacco.)

      You clearly don't accept that view, but I can sense no valid economic reasoning in your arguments against it above. You just invoke symmetry without taking into account the local population densities involved.

      So then come the moral questions.

      I don't mean to sound callous, but I find the application of absolute deontology to moral questions to be a waste of time. I don't accept your dictum that economic reasoning can ever be a justification for moral norms because I'm a utilitarian. So there is basically no point in the two of us arguing on that matter.

      It's the apparent position on your side that any appeal to market forces is a slippery slope downward to a market fundamentalism that I find pigheadedly stupid. It strikes me an as anti-market fundamentalism, just as dumb as the freshwater prejudice that markets always get things right.

      The thing regarding Chomsky is a simple expression of frustration at the apparent grandstanding on your part. Suppose that you are correct and my attitude towards the bad economics of slavery is morally wrong. How does you confronting me with that fact on a semi-obscure economics blog help anyone who is economically disadvantaged?

      Who, to put it bluntly, the fuck cares what either of us think?

    9. The plantation system was highly resistant to capital investment. It was profitable, but far from optimal. For example, after a vacuum based sugar extraction system was developed, it was never adopted. Slave labor was incrementally cheap enough so that there was no need for expensive equipment upgrades like this. (This is a common problem with investment. Burger joints could be much more efficient and heavily automated, but labor is too cheap to make this worthwhile.)

      In the north, where labor was relatively expensive, there was all sorts of automation. McCormick's harvester, pulled by teams of horses using swingleback hitches, was called Lincoln's ace-in-the-hole during the Civil War as it dramatically reduced the need for harvest manpower. Expensive labor has always been the main driver of automation.

    10. Slavery or serfdom or other kinds of forced labour are indeed a common response to situations where labour is scarce relative to land or other opportunities. But this points to the third factor: the structure of society (hierarchy, roles, ability to collect rents and taxes..). The Russian problem for instance, was not just that landlords could be left without labour; it was that the Russian state could not function without the military and administrative services from the landlords and, without the state, Russia was basically fodder for the neighbours (the Crimean Tatars, for instance, lived from raiding Russia for people, who they then sold to the Ottomans). So the Russian state coerced the lower classes into serfdom and the upper classes into compulsory state service.

      This is an area where economists could learn much more from sociologists, historians and anthropologists, as the core doctrines of economy do not have much to say about social structure (a market is basically socially flat).

  10. Anonymous2:51 PM

    In your update, I suspect that Groningen U is just trying to cross its ts and dot its is. (How do you write this phrase properly?). If they're going to be getting a lot more attention they're going to need a new level of internal scrutiny that they don't have in place yet. I don't doubt their numbers in their general implications. I suspect they will be corrected somewhat closer to OEPC, but will they totally correct the problem?

    They could be mistaken in trying to address some things I've said in terms of productivity alone, and most likely productivity calculations are not going to show the problem we're trying to get at here. My main point is that productivity measures don't necessarily measure technology. They can measure other things, such as shifts in accounting methods due to corporate inversion.

    1. Anonymous4:16 PM

      Oh, and to finish the last thought, productivity measures pretty accurately the exploitation of low wage labor.

      You have a robot that produces at X rate, and I have a robot that produces at 5X rate. Duh…

      The question for the ethically informed was always about whether our economy could absorb and help foreign workers without harming domestic workers. The clear answer is no. And the political instability that resulted might well end this union. Give it some time.

      These workers are just robots in this calculation. Do I agree? No. Is this how things are done? Yes.

    2. Anonymous4:19 PM

      Somehow "measures" was omitted from my last comment. Oh well, nobody's listening anyway.

  11. May I suggest http://www.iariw.org/papers/2013/Weipaper.pdf

  12. I think the subject here is Wisconsin policy in this district.

    Where are these people? Sean, want a debate?

    Fellows, want a debate?

  13. This comment has been removed by the author.

  14. How many times do I have to tell you that TFP is just the geometrically share-weighted factor prices? I know you've seen my paper on twitter: https://ideas.repec.org/p/new/wpaper/1513.html

    1. Achim6:12 AM

      While I don't like Mike's selfcongratulatory tone (on the other hand, other people here are not much better in that), he has a point. It's a while since I read Anwar Shaikh's paper, "Laws of Production and Laws of Algebra: The Humbug Production Function." I found it pretty disturbing. I read similar papers since then, and they dramatically reduced the attractiveness of macro or at least growth theory. If somebody writes a blog article with the title "Lazy econ critiques", this suggests that the econ aren't "lazy". My impression in this respect is different (even though I would not call myself "econ critique"; I am just a micro guy).

  15. If the think fate of macro hangs on the accuracy of measuring TFP "shocks," then yes, of course, you have been miseducated and you need to start over from scratch. There are no TFP shocks driving business cycles, unless maybe that once if you counted the oil embargo as a TFP shock, in which case you probably don't need to measure it in terms of TFP.

    I'm not sure if it's one of the reasons for that particular dataset to be nonsense, but generally the main reason the Penn World Tables are garbage is the PPP.

    1. Are there any established methodological problems with PPP?

      I became somewhat suspicious of the CIA Factbook and World Bank measures because they apparently don't include locally produced services that are public goods in many countries outside the USA. (Like Sweden and the UK.)

  16. Noah,
    I'm not so sure why you are so keen on Cochrane. My view is that good guys are:
    1. honest
    2. want rising median welfare
    Bad guys are
    1. dishonest
    2. uninterested in distribution of benefits.

    I'm not sure how Cochrane comes out a good guy on that basis.

  17. Duh, of course the OECD TFP numbers converge. They have set 2010 as 100 for all countries. By design all countries must then converge (and have not hsd much time to diverge). It's an issue of relative vs absolute effects .....

  18. The residuals (TFP)are just the composite of the effect of variables for which you don't account. Why don't you account for them? Because economists can't be bothered to systematically investigate the unknowns.

    Example of TFP

    A guy already working in a company and already being paid for doing X comes up with an idea (Y) that will improve labor productivity at no measurable additional cost of either labor or capital, and with no additional input labor effort, difficulty, stress, moral, etc.. but simply by utilizing an already highly developed labor skill set that nobody prior had thought of using in this application.

    Two simple examples of Y:
    1. Utilizing intricate sewing skills in an industrial high volume fine wire winding application. -- a 200% productivity improvement occurred. Real life example.... no exaggeration.
    2. Rearranging points in a process to weed out stuff that would ultimately fail quality criteria at points further along in the process at higher value added loss. Net greater productivity per unit produced and shipped by 100%. Another real life example, no exaggeration. This can be called "paying more attention to detail".
    3. Similar to 2) above: Implementing ongoing statistical trend data in real time processes to identify systematic drifts and spurious outliers followed by rigorous identification of root cause(s) and tooling adjustments to eliminate them. No additional engineering effort -- just more productive use of same. Imperceptible increase in labor with 1000% productivity improvement over 2 years. Real life example... no exaggeration. And BTW, Japan used this in a far more systematic and rigorous form which did add measurable labor and capital following WWII and was the predominant reason for Japan's post-war production machine and exports. The common name for it was "control charting".

    These productivity improvements spread rapidly intra-company and then even more rapidly thereafter inter-company, then inter-industry and ultimately inter-globally.

    These end up in the residuals (TFP). They only become apparent economically in creeping form over time as the effects spread from one unit of company to the entire company, then to more company's in an industry, then to more industries, etc.

    These types of improvements can be simply classified as "human efficiency" or " human factor yield improvement", where there is no perceptible or measurable increase in human effort applied, no or virtually no measureable capital expense. I simply refer to them as "attention to detail". A early example of this kind of efficiency gain which was widely proliferated for a long time was known as "time & motion" implementations... Smith's "pin" production example is probably the earliest reference to it in macro economics.

  19. What b.s....this is becoming my favorite blog. Like Scott Sumner's blog and Tyler Cowen's blog, this idiot says things that are provocative, that make you want to reach out and comment even when you wish to just lurch!

    "the real takeaway here, though, is that TFP measurements are HIGHLY suspect, and will continue to be so for the foreseeable future" - yes, and I knew that as a non-economist. TFP is measured as a 'residual' meaning it's not directly measured, hence prone to error due to everything and the kitchen sink included in it, as a residual. It's a fudge factor, akin to 'expectations' in macro models, or "natural interest rates", that also can't be measured. Actually even physics has this, with occult variables in quantum entanglement, but the difference is physicists can actually test their models, instead of just debating them online, and achieve results. Peace!