A Time Series Test of Regional Convergence in the USA with Dynamic Panel Models, 1972-1998

A good deal of controversy surrounds the empirical regularity of convergence. If capital’s share is taken to be 1/3, as in national accounts, then convergence should occur at a much faster rate than observed. Problems are worse if the economy is open. With perfect capital mobility convergence should occur at an infinite rate. Convergence estimates appear to be as slow for state economies as for national economies, even though the assumption of perfect capital mobility is a closer approximation of reality for these economies. Some argue that other variables, most prominently human capital, must be included in any cross sectional estimation of convergence. Supposedly, this addition of variables can bring the implied rate of convergence in line with empirical estimates by controlling for differences in the steady state level of per capita income. This paper extends the analysis of Islam (1995) to US states by estimating dynamic panel data models. This is a more appropriate method of allowing for different steady states. We find that the data suggests states converge very quickly, implying a high degree of capital mobility, if each state economy is allowed to have its own steady state captured through its own fixed effect. These results demonstrate the pitfalls of applying closed economy models to study growth in very open economies and the dangers of adding variables to the estimation which have, at best, only a weak relationship to differential steady states.