Tuesday, September 20, 2011

Is growth accounting really useless?

I came across this Mankiw’s comment on Oliner, Sichel and Stiroh (2007):

To be honest, in my own life as a practical macroeconomist, I do not spend a lot of time thinking about growth accounting. In fact, I can estimate with a fair degree of precision that I spend fifteen minutes a year on the activity. Those are the fifteen minutes that I teach growth accounting to undergraduate students in my macroeconomics course.

While reading this paper I found myself reflecting on my almost complete lack of attention to the growth accounting literature, to which this paper very ably contributes.

One reason is that this literature seems mired in a host of issues that quickly make a reader’s eyes glaze over. Some of these issues are technical, such as distinctions between gross output and value added and the index number theory that bridges that gap. Others involve data availability, such as the potentially important role of unmeasured intangible capital. Out of necessity, many of these issues get resolved by imposing assumptions on the production process which, although not outlandish, are neither compelling nor verifiable. This paper, for example, at times makes an assumption about the complementarity between information technology and intangible capital that seems to be just pulled out of a hat. But I think there is a more fundamental reason why the growth accounting literature fails to have a larger impact. Even if one grits one’s teeth to make it through all the technical issues, and even if one has enough credulity to buy into all the necessary assumptions, the exercise does not deliver what we really want. Ultimately, God put macroeconomists on earth for two reasons: forecasting and policy analysis. We want to know how the world is likely to look in the future, and we want to know how alternative policies would change the future course of history. Unfortunately, growth accounting contributes relatively little to either forecasting or policy analysis. Instead it is a deeply data-intensive exercise that often gets so deeply enmeshed in its own internal logic that it never returns to the big questions of macroeconomics.

Long ago, some economist—I believe it was Moses Abramovitz—called multifactor productivity “a measure of our ignorance.” That is, we account for changes in capital, labor, labor quality, and the many other determinants of output we can measure, and the changes in output left unexplained are called “multifactor productivity.” But that is really just giving a fancy name to something about which we are pretty clueless. When reading this paper I started playing a game where every time I read the authors say something about “multifactor productivity,” I imagined putting some version of “a measure of our ignorance” in its place.

Let me give an example. At one point the authors write, “MFP growth strengthened in the rest of nonfarm business, adding roughly 3/4 percent-age point to annual labor productivity growth during 2000–06 from its 1995–2000 average.” I rewrote the sentence as follows: “our ignorance strengthened in the rest of nonfarm business, adding roughly 3 ⁄4 percentage point to annual labor productivity growth during 2000–06 from its 1995–2000 average.” Framed in this alternative way, the statement carries an almost comical hollowness. It also makes it clear why statements about multifactor productivity are of limited use for either forecasting or policy analysis. Measured ignorance is probably better than unmeasured ignorance, but it would be a mistake to confuse it for real knowledge.


No comments: