Tuesday, April 05, 2011

About that TFP stagnation...

Via Tyler Cowen, I see that David Beckworth has posted a graph of U.S. total factor productivity (TFP) from 1947 to 2010, using data from John Fernald at the SF Fed (who BTW is a co-author of my advisor, Miles Kimball). Total factor productivity, recall, is the part of production that can't be explained in terms of inputs of capital, labor, and other known factors of production; TFP is often called "technology", though it probably includes a few other things.

Anyway, here is Beckworth's graph:

Assuming that TFP = technology, this graph definitely seems to support Tyler Cowen's hypothesis of a "Great Stagnation" in technological progress around 1973.

However, just for fun, I decided to update the graph. John Fernald, ever the careful empiricist, breaks his TFP series down into two sectors: TFP in the production of durable goods (cars, buildings, TVs, machinery) and TFP in the production of nondurable goods (clothing, food, services).

Here's what it looks like when we graph both of those on the same graph:

Wow! If you look only at the durables sector, there was no Great Stagnation at all - in fact, quite the reverse, since durables TFP has been growing more strongly post-1993 than it ever did in the post-WW2 boom! Consider this: In the 26 years from '47 to '73, durables TFP nearly doubled, but in the 15 years from '94-'09, durables TFP more than doubled.

Yes, from this graph, it definitely looks like something big did happen to technological progress. But it looks like it happened not in 1973, as Cowen claims, but a decade earlier. In the 15 years to 1963, the two sectors progressed pretty much in tandem. But sometime in the early- to mid-60s, they diverged wildly, with nondurables TFP rising anemically through the late 70s and then basically flatlining until now. Durables TFP looks to have suffered its own very minor slowdown in the mid-70s (which is probably the reason why overall TFP looks like it took a turn around that time), but then exploded with unprecedented vigor after '93.

Like David Beckworth, I am also more convinced of a Great Stagnation than I was before I looked at Fernald's data. But I think Cowen's hypothesis needs a bit of updating. It is only the nondurables sector that has stagnated, and it happened in the early 60s. Why did it happen? My first guess was agriculture, but nope, it's not that. Did the years after WW2 simply see an unprecedented one-off boom in nondurables production, with a return to normalcy in the 60s?

Anyway, figuring out what really happened to technology in the 60s will take far more data-gathering and careful analysis than I can put into this blog post. A detailed historical breakdown of TFP by industry would probably be the place to start. But in the meantime, I think we should look at the "Great Stagnation" as a more subtle phenomenon than simply the exhaustion of the "low-hanging fruit" of nature. Our technologies for producing durable goods are improving faster than ever.

Update: Thanks to commenter Andy for pointing out that services are also included in nondurables. 

Update 2: Slow service-sector TFP growth  appears to have been a big factor in the 70s and 80s, but not since the early 90s. Also, commenter Mark reports that utilities and mining (nondurables) have experienced drops in TFP since the 90s, while agriculture, trade, and transportation have seen strong productivity gains in that time frame. Hmm...

Update 3: Tyler Cowen responds, saying: "note the importance of sectoral shifts into lower-growing sectors". First of all, such a shift may be happening, but it is not apparent in Fernald's data set; the GDP share of the durables sector, in which TFP growth continues apace, has held steady at 20%. Also, note that a shift into sectors with slower TFP growth does not necessarily indicate that technological innovation is slowing down...imagine an economy in which we get infinitely better at producing right shoes, but no better at producing left shoes, and you'll see what I'm talking about.


  1. I was wondering why this breakdown doesn't include services, but after looking at Fernald's data, I take it they are included in what you call "nondurable goods" (which I am assuming corresponds to his category of "non-equipment output").

    1. This is vital. Finance has got a much larger share of GDP lately.

  2. Assuming my previous comment is correct, let me suggest that the important distinction here may not be between durables and nondurables but between goods and services. I suspect that services are driving the "nondurables" category over the past few decades. In fact, this may be the whole explanation for the great stagnation: services productivity was always stagnating, but before 1963 services were not so important.

    Indeed services have become much more important compared to nondurable goods specifically because their productivity has stagnated, which has resulted in their becoming a larger and larger fraction of expenditures. In 1954, households spent roughly the same amount on services as on nondurable goods. In 2010, they spent 3 times as much on services. The ratio of nondurable goods to durable goods expenditures has actually fallen over the same period (and even since 1973), which suggests that nondurable goods productivity has kept up with durable goods productivity.

  3. From the price indexes, it's clear that I was wrong about nondurable goods productivity keeping up with durable goods productivity. It kept up prior to 1963, just as your chart would suggest, and then it diverged. But nondurable goods productivity has clearly risen much faster than services productivity, both before and after 1963. I'm still inclined to think that the relative share of goods and services in overall output is the main driver of the changing growth rate of TFP.

  4. First of all, here is the complete historical record via Alexander Field:

    Average annual TFP

    Note that the "Golden Age" (1948-73) is behind three other periods in the rate of TFP growth.

    With respect to the divergence in "nondurable" TFP growth after 1963 the devil is in the details.

    If one looks at the historical record from the "Golden Age" what you see is the following for the fastest growing sectors. the following are average annual rates of change.

    Public utilities-4.8-1.2
    Real estate-3.3-1.3

    (Manufacturing includes both durable and nondurable manufacturing.) Note that there was a dramatic slowdown in TFP growth in all of the rapidly growing nonmanufacturing sectors with the exception of Trade and Transportation after 1966. (It's useful to remember that Alexander Field credits much of the rapid TFP growth during this period to Trade and Transportation, and specifically the Interstate Highway System.) Thus the divergence between durable manufacturing and the rest of the economy in the mid 1960s was due to the slowdown in the rapid growth in these sectors. "Nondurable" TFP growth came to a halt when Transportation and Trade joined these sectors in stagnating after 1973.

    With respect to the resurgence in TFP growth in 1995-2005, again the details matter.

    No subsector of nondurable manufacturing experienced above average TFP growth during 1995-2005. Within the nonmanufacturing sector there is a great variation in performance. Mining, Utilities and Construction have all seen declines in TFP. On the other hand there has been above average growth in Agriculture, Trade, Transportation and Information.

    In the durable manufacturing sector there are ten subsectors. Electrical Equipment, Appliances and Components has actually seen a decline in TFP. Only two subsectors of durable manufacturing have seen above average TFP growth: Computers and Electronic Equipment, and Miscellaneous. In fact Computers and Electronic Equipment has seen a six fold increase in TFP between 1993 and 2007.

    It would be tempting to say most TFP growth during 1995-2005 was the result of durable manufacturing TFP but upon examining the details we see this is not at all the case. A better generalization is to say that TFP growth during this period is related to the ICT revolution.

    Virtually all of the above average performance in durable manufacturing is due to the manufacture of computers and electronic equipment. And most of the "nondurable" sectors that had above performance in 1995-2005 are so due to the ICT revolution. In particular, Bart van Ark has shown that computers have led to a big increase in TFP in "distribution" which is reflected in the Trade and Transportation sectors. In fact the improvement in "distribution" TFP during this period dwarfs the contribution of computers and electronic equipment manufacturing to the surge in TFP growth in 1995-2005.

    One last note. It is rare (exceptional) in the historical record to find an age when improvements in TFP were dominated by manufacturing. The TFP growth during the Gilded Age came about in Trade, Transportation and Communications (railroads and the telegraph). The TFP growth during the Great Depression came mainly in Transportation (the U.S. Route System and the five fold increase in the share by the interstate trucking industry). The only time manufacturing dominated TFP growth was during the 1920s (the final burst in the electrification of factories). The most recent period really does not add to this exception.

  5. In about 1997, Alan Greenspan became convinced that the economy was in the midst of a productivity boom and therefore no tightening in monetary policy was required, despite the unfolding boom. It wasn't at all obvious then; even if it is now. To prove his hypothesis, he sent the Board's staff to look for evidence on productivity by industry. He got his proof by (sort of) contradiction: the Board's Industrial Output section produced data that showed that productivity growth in banking (a service), had been about zero over the previous 20 years. This was obviously false, a manifestation of ,first, the fact that TFP is measured as a residual item; and, second, that little is done to correct for changes in the quality and nature of the final product. Banking isn't just checking and savings accounts anymore. Hedonic measures are used in some industries where the data are good (autos), but not everywhere. The moral of the story is be wary of the data, particularly in services where measurement is really difficult.

  6. Yep. TFP = technology is a dodgy assumption. Estimates for TFP in the Canadian mining and oil and gas sector show technical regress: TFP in 1961 was 3 times higher than it is now. That says more about the problems of interpreting TFP than about technical change in the resource sector.

  7. Diego Espinosa2:54 PM

    As David Collum noted, it seems we got better at producing things (IT) that wear out quicker. Adjusting GDP for depreciation would reduce TFP. The adjustment only makes a difference when there are large changes in (actual, not accounting) aggregate depreciation rates.