Tuesday, September 21, 2010

Multi-Decadal Slowdown in US Growth

I know this is not a novel observation, but I wanted to put up a reasonably current graph documenting the fact that the US economy has been growing more and more slowly over time.  The above shows the BEA's "change over prior period" in real GDP (blue), a linear fit to that (black) and the compound average growth rate over the prior ten years (red).  The data go through 2010 Q2.  Any way you look at it, the US has been experiencing gradually slowing growth for many decades.

Given deleveraging and peaking oil, I doubt the long-term trend is about to improve.


bmerson said...


I'm assuming that this is the overall growth change, right? It would be interesting to see whether the trend for the yearly change in per capita growth rate is more or less pronounced.


Gary said...

Two thoughts...
1) I look at the growth curves and under the noise see a drop in growth around 1% starting in the late 70's and early 80's, coincident with the first oil shocks - US peak oil - and Reaganomics. Perhaps we are seeing another leg down now.

2) With U.S. growth generally dropping, is it all that surprising that most profits are now coming from abroad where growth rates are higher? (your next post) Especially today when a large fraction of profits are "financial", it's difficult for the capitalist system to generate income without decent general economic growth.

Benno said...

It also makes a very arresting graph if one plots real GDP deltas (5,10,15 &20 year) since the war.

The economy in 2009 was only 15% larger than it was in 1999!

Through the much maligned 70s, the economy was still twice the size it was 2 decades earlier.

Michael Cain said...

A simple linear regression is suspect in this case because the process is clearly heteroskedastic -- the fancy statistical term for "the variance is not consistent over time". The so-called Great Moderation from 1985 is quite clear. If you do the regression against the 10-year moving average, the curve is significantly more level (the slope is reduced by almost a third).

Also, I would note that the data go back to 1947. There are certainly good reasons for choosing to start in 1950 (eg, eliminating post-war effects), but it also turns out to be the early starting point that makes the decline look worst. Starting in 1950 makes a significant difference compared to starting either a few years before or a few years after. If you want to make it look even worse, start around 1962.

Stuart Staniford said...


You're right on heteroskedacity, but I wasn't really trying to make any particular quantitative estimate of the trend, just make the point that there was one. 1950 was just the first year that gives reasonable x-axis labels...

Michael Cain said...


But if you correct for heteroskedacity in some fashion, and for the substantial autocorrelation, and add in the earlier data, the resulting regression line is much less impressive. The end point in 2010 is little changed, but the starting point in 1950 is a lot lower -- about 3.2% rather than 4.9%. Plus the last 10 quarters have had a rather outsized effect; drop those from the data set and make the other corrections and the negative trend coefficient misses being significant at the 90% level (by a little).

It's a critically important topic; lots of things depend on continued productivity gains in the future, and energy constraints make it much less likely that those gains can be maintained. And while my general outlook is pessimistic, I try not to make things worse than they really are.

Stuart Staniford said...


No way can you justify chopping off the great recession - after all, it was a direct result of the same things that led to the "great moderation" previously, so counting all the positives from those years but then chopping off the negatives from the recession is not legit.