Counterfactual Failure Alert!

5d0320d4-f194-468f-927a-0ea2798f35e3We’ve been focusing on productivity in recent issues and had some questions concerning possible measurement of high-tech productivity from some readers in the last weeks. Wouldn’t that be nice, but, alas, two new studies dim such hopes.

The two papers make many of the same points. All agree that there is a measurement problem, but that it was likely larger during the high growth period of 1995 to 2004 than it is today. In “Does the United States have a productivity slowdown or a measurement problem?” authors David M. Byrne, John G Fernald, and Marshall B. Reinsdorf suggest some adjustments to IT-related sectors, which make recent growth in investment and GDP look a bit better, and adjustments to durable goods manufacturing, which make labor productivity look even worse.

In his “challenge to mismeasurement explanations,” University of Chicago professor Chad Syverson argues that the highest estimate of the surplus generated by internet-linked technologies is only one-third of the “purportedly mismeasured GDP,” even using his “very expansive assumptions about the value of leisure time.” Equally bad, to account for even a modest piece of the $2.7 trillion hole left by the productivity slow-down, output and productivity in information and communications technologies (ICT) would have to be multiples of what is currently reported. For example, combined employment in computer/electronic manufacturing, information, and computer systems design rose 3.6%, from 5.58 million to 5.78 million, between 2004 and 2015. If those workers actually did produce the output lost to the productivity slowdown, real value added per worker would have risen an “astounding” 363% over those years, far outpacing the 83% growth in durable goods manufacturing between 1995 and 2004.

Syverson throws cold water on the argument that Gross Domestic Income being slightly higher than Gross Domestic Product since 2004 suggests that people are being paid to make things that are sold at heavy discounts: that disparity is based on “unusually high” capital income [read profits], not labor income. He points out that if the annualized 1.5% drop in labor productivity were to persist over the next 25 years it would compound to a 45% decrease in per capita income.

Byrne et al. argue that since domestic production of IT hardware has decreased, the effect of mismeasurement was likely larger before the as-measured slowdown after 2004. They also make the point that consumer benefits from Google searches and fun on Facebook are non-market (akin to playing soccer with one’s children), and anyway are too small to compensate for the loss in “overall wellbeing” driven by slower market-sector productivity growth. They see real improvement in matching products and services with consumers, but nest these in a continuum of changes that once included the shift in responsibility for doing the laundry from a domestic washer-person, to a family-member, and finally to a washing machine.

They point the finger at the decline in economic dynamism including the dearth of start-ups and slower reallocation of resources following productivity shocks, and note the possibility that the fast-growth period between 1995 and 2004 might be the anomaly. And they admit they don’t know what gains will come from cloud computing, and that more IT-driven increases may be in the offing.

The productivity showdown is widespread, and Syverson matches OECD labor productivity data with ICT intensity, which he measure as access to broadband, for close to 30 countries. He regresses labor productivity with broadband penetration for a -0.001 co-efficient, a point estimate that implies that one standard deviation (SD) in broadband penetration comes with one-fiftieth of a SD in the magnitude of the slowdown. This coincides with work showing that differences in the slowdown in total factor productivity in U.S. states are not correlated with each state’s ICT intensity.

Syverson writes that the “intuitive and plausible empirical case for the mismeasurement hypothesis faces a higher bar in the data,” and that the slowdown is real. In closing he notes that whether the slowdown will end “anytime soon is an open question.” We’re not seeing it yet, so that’s a question we have to address.