Convergence revisited

The conventional approach to testing whether economies converge examines the cross-sectional relationship between the growth rate of per capita output over some time period and the initial level of per capita output. This paper shows that this approach is valid only if economies have identical first-order autoregressive dynamic structures and all permanent cross-economy differences are completely controlled for, conditions that are grossly violated for the data sets considered here. The paper also develops an alternative approach that is valid under much less restrictive conditions. Strong evidence is found for conditional convergence of the 48 contiguous U.S. states and a group of 54 countries. Our evidence does not support theories in which trend growth rates differ across economies endogenously.