Shares of Real GDP Don't Give the Real Story of Manufacturing
(Bloomberg View) -- Here's a striking chart, from a 2014 Journal of Economic Perspectives article by Brookings Institution economists Martin Neil Baily and Barry P. Bosworth on the state of manufacturing in the U.S.:
What this seems to show is that while manufacturing has shed jobs, its importance to the U.S. economy is undiminished. The contrast is remarkable, and the chart makes it quite clear.
As I discovered when working on a column last week about how manufacturing is holding up in the U.S., though, the chart is also quite wrong. The reasons it's wrong are pretty wonky, and I'm still not sure I fully understand them. But given that the above chart -- along with a similar one produced by a couple of economists at the Federal Reserve Bank of St. Louis earlier this year -- keeps getting brought up in economic debates as evidence that manufacturing is holding up better in the U.S. than is widely thought, it seems important to explore why it's problematic.
The problem has to do with the concept "share of real GDP," which, when you start unpacking it, turns out not to mean what it sounds like it should. Real gross domestic product is GDP adjusted for inflation, which allows us to better measure and compare growth over time. If you're measuring and comparing different industries' shares of GDP over time, it's not immediately obvious why you would need to do such an inflation adjustment. The very act of dividing manufacturing's contribution to GDP in 1960 by overall GDP in 1960 does the inflation adjustment for you, given that both amounts are in 1960 dollars. And, in fact, the government agency responsible for producing the National Income and Product Accounts that include GDP, the Bureau of Economic Analysis, provides only nominal GDP shares in its GDP-by-Industry statistics. Here's manufacturing's share of nominal GDP since 1947, as reported by the BEA (a chart I already used in my Wednesday column):
What is not reflected in this chart is changing relative prices. Because the prices of manufactured goods have gone up more slowly than those of other goods and services (think health care), Baily and Bosworth reasoned that manufacturing's shrinking share of nominal GDP understates its continuing impact on the economy. They have a point, but their solution of relying on share-of-real-GDP numbers delivers some very strange results.
First, let us consider what comparing manufacturing's share of real GDP in, say, 1960 with what it is today is even supposed to mean. Ideally, it would provide the answer to this question: Suppose all prices had remained at 1960 levels; what proportion of the total value of this year’s output would have been accounted for by manufacturing output? That sounds like an interesting thought experiment, but it's not all that revealing a measure of manufacturing's current economic importance or clout. It also reflects how real GDP was calculated for many decades.
The problem with this "fixed weight" method of calculating real GDP was that the further you got from the base year, the less meaningful the numbers were. As economist Karl Whelan put it in 2000:
Why should we care about how the value of output would have grown had all prices remained at their year-b level? What’s so special about year b?
For a long time, the BEA combated this problem by "rebasing" real GDP numbers every five years, but this had the side effect of making it almost impossible to compare real GDP over long periods of time. So in 1996, the agency switched to the chain index formula devised in 1920 by the brilliant if not always prudent Yale economist Irving Fisher (he was the guy who declared in 1929 that stock prices had reached a "permanently high plateau"). Chain indexing updates the prices used to calculate real output every year, rendering the choice of base year much less important and allowing for meaningful real GDP comparisons over periods of many decades.
This had a side effect, too, though. For reasons that I'm not going to get into here but that you can read all about in Whelan's 2000 paper, "A Guide to the Use of Chain Aggregated NIPA Data," the different components of real GDP can no longer be added together. That is, they can be added together but, except in the base year, they don't add up to real GDP. As time goes by, the sum of the different parts of real GDP gets bigger and bigger relative to the real GDP total. So calculating manufacturing's share of real GDP over periods of decades generates percentages that are effectively nonsense.
I asked Whelan, who was at the Federal Reserve in 2000 and is now a professor at University College Dublin, to take a look at the above chart and the one generated by the St. Louis Fed. "Yes, these papers report mistaken calculations," he concluded. I then emailed the authors of said mistaken calculations. This was part of the response I received from Baily, who was chief of the Council of Economic Advisers for the final year and a half of the Bill Clinton administration:
The share of nominal value added created in manufacturing has declined steadily. Because the relative price of manufactured goods has declined, the real value added share has not declined in the same way. That was the basic story we wanted to tell, and holds up. Saying the real share is constant may be an overstatement.
Close to the base year, the comparison of real value added is fine. As you get away from the base year, the calculation goes off, as you point out. I am not sure there is a good way to remedy that problem.
The BEA's remedy to the problem is to put up warnings against doing share-of-real-GDP calculations all over its website. It even presents its industry data in a way that makes it close to impossible for a non-expert user to calculate shares of real GDP further back than 1997. But economists know how to get around such barriers, and are conditioned to see real (that is, inflation-adjusted) data as superior to nominal data. Whelan wrote his paper in 2000 in part because "mistaken calculations based on real NIPA data have become common in both academic publications and in the work of business economists."
Baily and Bosworth hadn't read Whelan's paper, but they knew the measure was problematic. They just figured it was still superior to share of nominal GDP. By this point, I'm certain it isn't, although I had yet to fully convince Baily and Bosworth (we've been emailing a lot) when I turned this in to my editor.
What does all this mean for our understanding of the state of U.S. manufacturing? Adjusted for inflation, U.S. manufacturing production really has grown a lot since the early 1970s, even as many industries have struggled and the sector as a whole has shed millions of jobs. That real growth has slowed since 2000, though, and as I've written before (and Baily and Bosworth discuss in their 2014 article), a lot of the manufacturing output growth since the mid-1980s, when the BEA started factoring quality improvements in computers into its measures of real output, has been driven by the rising quality of computers and electronic products, not by increases in actual, you know, output. In the meantime, U.S. economic activity has steadily shifted away from goods and toward services. In short, manufacturing isn't going away, but its share of the economy is in no meaningful sense the same now as it was in the 1960s.
This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
Justin Fox is a Bloomberg View columnist. He was the editorial director of Harvard Business Review and wrote for Time, Fortune and American Banker. He is the author of “The Myth of the Rational Market.”
The Journal of Economic Perspectives is the excellent American Economic Association publication that "attempts to fill a gap between the general interest press and most other academic economics journals," and articles in it sometimes reach a wide readership.
This a paraphrase of a passage from the 2000 Whelan paper discussed elsewhere in the column.
Including an economist at the St. Louis Fed, but I haven't heard back from him.
A sample: "Comparisons of two or more different chained-dollar series must be made with caution, because the prices used as weights in the chained-dollar calculations usually differ from the prices in the reference period, and the resulting chained-dollar values for detailed GDP components usually do not sum to the chained-dollar estimate of GDP or to any intermediate aggregate."
I am conditioned to do this, too, and I cannot guarantee that I have not at some point published a share-of-real-GDP chart.
The Federal Reserve publishes a monthly index of real industrial production in manufacturing and many of its subsectors, which is available back to 1972 The BEA publishes quarterly and annual value-added and gross output by industry data which it makes available in real terms back to It also provides nominal value-added and gross output data back to 1947, along with the chain quantity indexes needed to convert these into real terms. All of these different measures of real manufacturing output seem to have followed pretty similar trajectories.
To contact the editor responsible for this story: Brooke Sample at bsample1@bloomberg.net.
For more columns from Bloomberg View, visit http://www.bloomberg.com/view.
©2018 Bloomberg L.P.