The New Economic Geography Model pioneered by Paul Krugman has revolutionized our thinking about the location of factors of production, yet there has so far been little empirical support for this theory. Empirical tests suffer from massive endogeneity problems, and simulations seem to replicate very poorly the data, in part because they use very sparsely the data. But combining both approaches coax out their advantages while not revealing too many of their disadvantages. One attempt was by Kristian Behrens, Giordano Mion, Yasusada Murata and Jens Südekum, who estimate a structural model and then simulate border effects.
Eckhardt Bode and Jan Mutl take a somewhat different approach. They take a fully specified structural model, take a Taylor expansion around the empirical steady-state, and then estimate the resulting reduced form. And the model is soundly rejected on US county data, mostly because migration does not vary in the way the model would want. Not imposing theoretical restrictions improves the estimates considerably, which is not reassuring.
Does this mean the NEG models can be dumped now? Not yet. They still gives us good insight, and if they fail on migration, they appear to be holding rather well with regard to the links between wages and good prices. And the empirical methods can certainly be improved, especially regarding spatial autoregression.