Symposium on measures of economic growth

One of the most important issues in recent U.S. macroeconomic performance is the productivity slowdown. Since 2004, U.S. labor productivity has slowed dramatically. From 2004 Q1 to 2017 Q1, real output per hour of the nonfarm business sector growth has averaged 1.3 percent per year. In contrast, during the preceding 9 years (from 1995 Q1 to 2004 Q1) labor productivity had grown an average of 3.0 percent per year. If the earlier period’s productivity growth rate could have been maintained after 2004, U.S. GDP would now be at least $3 trillion larger,

Among the various explanations that economists have floated, one is the idea that maybe the slowdown is an illusion—an artifact of mismeasurement in the statistics for real GDP and and price indexes that are used as deflators.

The latest issue of the Journal of Economic Perspectives includes a symposium addressing the question, “Are measures of economic growth biased?” It features articles by Martin Feldstein of Harvard University, by Chad Syverson of the University of Chicago, and by Erica Groshen, Brian Moyer, Ana Aizcorbe, Ralph Bradley, and David Friedman, all formerly or currently affiliated with either the Bureau of Labor Statistics or the Bureau of Economic Analysis.

Feldstein’s article, “Underestimating the Real Growth of GDP, Personal Income, and Productivity,” presents the case that the growth of real GDP and productivity are systematically underestimated. To those who are familiar with the Boskin Commission’s 1996 report on the consumer price index, the sources of bias are familiar. For most components of GDP, growth in real GDP is derived by deflation—that is, dividing the relative growth in nominal spending by the relative growth in a price index, or “deflator.” Inadequate adjustment for quality and for new products cause the CPI or producer price index to be upward biased (that is, to overstate the growth in prices). This bias, in turn, causes GDP and productivity growth to be downward biased, understating overall growth in the nation’s output.

If you haven’t read about these biases, Feldstein’s article is a pretty good summary of the literature, though I’d also recommend reading the statistical agency article by Groshen, et al. to provide a more accurate explanation of how the agencies attempt to adjust for quality in practice. I agree with much of what Feldstein says. For example, I’ve long thought that standard statistical methods tend to overestimate price change and underestimate quantity change for new goods or for goods and services that are undergoing substantial quality improvement. How large are these biases? For personal consumption expenditures, a point estimate of about a half percentage point per year, with a confidence range running from zero to one percent per year would be consistent, for example, with David Lebow and Jeremy Rudd‘s estimates in a 2003 Journal of Economic Literature article, after accounting for the fact that PCE is not subject to the CPI’s upper level sustitution and weighting biases. Groshen, et al., derive estimates of similar magnitude.

I’ll mention three areas where I disagree with Feldstein’s analysis.

• Feldstein ignores the possibility that in some cases errors in quality adjustment may go the other direction and underestimate price change. This can represent a strategy by producers or by retailers to use the introduction of new varieties of a good as an opportunity to raise their prices. Robert Gordon’s 2009 article on apparel prices and papers on rental housing by Crone, Nakamura, and Voith and by Gordon and vanGoethem document examples of price indexes with downward biases. Another example of a type of bias that runs counter to the traditional quality/new goods bias is the offshoring bias, in which GDP and productivity is overstated by failing to properly reflect the reduction in price when producer switches its source of supply from a domestic producer to an imported source. This bias was discussed by Susan Houseman, Christopher Kurz, Paul Lengermann, and Benjamin Mandel in the 2011 Journal of Economic Perspectives.
• While Feldstein presents evidence for downward bias in GDP growth, to explain the post-2004 productivity slowdown he really would need to point to a marked change or deterioration in quality adjustment methods around the time of the slowdown. He points to the the growth of services, which tend to be harder to measure than services, but that has been a gradual, long-term change and wouldn’t explain the abrupt change in the productivity trend.
• The final section of Feldstein’s article is on using imperfect data, and it doesn’t really fit in with the rest of the article. For example, he begins with a discussion of assessing business cycle conditions (that is, are we in a recession or a boom). But his reservations about using GDP as a measure of the business cycle have little or nothing to do with the quality and new-goods biases that are the subject of the rest of the paper. We’ve long known that when all data series are subject to measurement error—as they surely are—analysts may be able to draw better inferences by combining information from multiple data series. Indeed, the Council of Economic Advisers describes using that approach, and that general approach is also widely used by analysts at the Fed and elsewhere. But that conclusion is driven more by volatility rather than bias. Because quality and new goods biases are small in the short run compared with typical macroeconomic shocks, I doubt that price index bias has any substantial adverse effect on monetary policy. Even if price indexes are biased, the BLS uses consistent methods for measuring prices, which suggests that any upward bias is stable and relatively consistent. This consistency is confirmed by external big-data price indexes such as PriceStats. I’ve always thought that one of the reasons the Fed set its inflation target at 2 percent rather than zero is because the Board is aware that inflation measures may be biased upward.

Syverson’s article, “Challenges to Misemeasurement Explanations for the US Productivity Slowdown,” directly addresses the hypothesis that the post-2004 productivity slowdown can be explained by mismeasurement. He finds, however, that there is considerable evidence that mismeasurement cannot account for most or all of the productivity slowdown. First, the slowdown in measured productivity is not just a United States phenomenon, but has occurred at about the same time in more than two dozen other advanced economies. These countries include a variety of statistical systems and varying degrees of production and use of information and communications technology (ICT). Second, the attempts to measure the consumers’ surplus associated with Internet services that are missing from GDP have come up with estimates that are much too small to explain the productivity slowdown. Third, the slowdown is too widespread across industries to be accounted for by mismeasurement of ICT or digital technologies. Fourth, the income-side measure (called “gross domestic income” in the U.S. accounts) provides a check on nominal GDP, and the comparisons of these two measures suggest that mismeasurement of nominal GDP is much smaller than the productivity shortfall. Syverson’s conclusion that mismeasurement doesn’t explain the productivity slowdown was supported by another recent article by David Byrne, John Fernald, and Marshall Reinsdorf in Brookings Papers on Economic Activity.

The article by Groshen and her statistical agency coauthors, “How Government Statistics Adjust for Potential Biases from Quality Change and New Goods in an Age of Digital Technologies: A View from the Trenches,” provides a detailed overview of the methods that BLS currently uses for quality adjustment. They also compare the official methods to some alternative methods that have been proposed by researchers but which are not currently used for official indexes. In addition, they update Lebow and Rudd’s 2003 estimates of quality and new goods bias with the results of more recent estimates of medical care and of computer and information equipment and  services. They conclude that the best estimate of quality and new goods bias in consumption and investment is about 0.4 percentage point per year, which remains close to Lebow and Rudd’s 2003 estimate. The article concludes with a discussion of research and other initiatives that are underway at the statistical agencies to improve the measurement of prices and output.

Anyone who cares about long-term economic questions such as whether productivity and living standards can continue to grow, or whether we the age of rapid improvement in living standards is now past, should be interested in these papers and the issues that they discuss.