From his blog:
Now, on to geography. A decade after the original new trade stuff, I started thinking about what happens when some (but not all) economic resources, especially labor and capital, can move. In the world of the old trade theory, “factor mobility” was a substitute for trade: if factories and industrial workers can move freely, they’ll spread out to be close to the farmers, and neither food nor manufactured goods will have to be shipped long distances. But in the economies-of-scale world I had been studying, the “centrifugal” effect of widely dispersed resources, which tends to push economic activity into spreading out, would be opposed by the “centripetal” pull of access to large markets, which tend to promote concentration of economic activity.
Think of Henry Ford and his Model T. He could have established many factories, spread across the country, to be close to his customers. Instead, however, he found that it was worth incurring extra shipping costs to achieve the economies of scale of one big factory in Michigan.
And once you’re concentrating production in a limited number of locations, which locations will you choose? Locations where there’s a large market – which will be locations where lots of other producers have also chosen to concentrate their production. If the centripetal forces are strong enough, you’ll get a cumulative process: regions that for historical reason have a head start as centers of production will attract even more producers, becoming the economic “core” while other areas become the “periphery.” Thus for about a century, until the rise of the Sunbelt, the great bulk of U.S. manufacturing was crammed into a fairly narrow belt from New England to the inner Midwest; today, 60 million people live along a narrow stretch of the East Coast. Those 60 million people aren’t there because of the scenery; each of them is there because the other 60 million people are also there.
The same sort of logic explains why particular industries concentrate in certain locations, except that in such cases the logic involves things like a deep labor market for specialized skills and a good market for suppliers of specialized inputs. What determines which industry locates where? Often, accident: Silicon Valley owes its existence in large part to a couple of guys named Hewlett and Packard, who started some stuff in their garage, New York is New York because of a canal that only pleasure boaters use today.
Again, this may seem obvious, and it is now – but it wasn’t before 1991 or so. As with trade, the plain English version was possible only after the mathematical models had been worked out.
You may ask, where’s the policy implication of all this? Actually, the policy morals are fairly subtle – for example new trade theory does suggest a possible role for government interventions, but also suggests bigger gains from trade liberalization. Mainly my work in trade and geography was about understanding the world, not driving a political agenda.
He's also posted his 1991 paper, "Increasing Returns and Economic Geography" (PDF), which lays out the mathematical model.
