Over the last quarter century, less productive states in America had been catching up with the rich states. I’m talking about unconditional beta convergence:
GSP PC is Gross State Product per capita. The green line excludes territories colored with red (see “Technical details” for the reasons and OLS estimates). Otherwise, less developed states grew faster than the rest.
However, most of this convergence occurred before 2000:
Slopes after 1992 are not statistically significant. Including the post-2008 recovery:
And after the 1990s, the dispersion of GSP (another perspective on the catch-up process) no longer declines in the US:
Unconditional convergence is not an empirical fact, but something we expect in the workhorse models of economic growth. It exists in the data (especially for developed countries) and in some intra-country studies. Its disappearance from the American economy is not a good sign, however. Unconditional growth acceleration leaves a possibility that regions can catch up with the best neighbors simply by waiting long enough. Now, it’s not the case: states must do something.
A return to the pre-2000 conditions (whatever it means) would not necessarily help. One confirmed source of unconditional convergence is manufacturing (see Dani Rodrik’s 2013). But until the costs of getting manufacturing jobs back to the US remain unclear, that’s not a plan.
I’m going to write more on this problem later, but you can explore the GSP dataset yourself:
Data and code
- Dataset and replication files on GitHub
The GSP data is a straightforward extension of SIC-NAICS files from the BEA. See “usa_gdp_state_convere_data.do” for details.
Key charts exclude three US regions. District of Columbia is out because it’s small and not a state. North Dakota and Alaska have been heavily influenced by oil production and prices. Their inclusion would be distortive. I kept Texas because its oil sector is relatively small compared to the entire state’s economy.
Anyway, convergence is still disappearing in the 2000s, even if we keep all 50 states:
“Log of real GSP” refers to the level of GSP at the beginning of the period (e.g. for 1987, it’s 1987-1992). On the left-hand side of the regression is cumulative 5-year growth.
With 48 states:
So the overall convergence looks deceptive: