Evaluating the Markov Property in Studies of Economic Convergence

Abstract

Markov chain theory, which has frequently been applied to analyze income convergence, imposes restrictive assumptions on the data-generating process. In most empirical studies, it is taken for granted that per capita income follows a stationary first-order Markov process. To examine the reliability of estimated Markov transition matrices, the authors propose Pearson X2 and likelihood ratio tests of the Markov property, spatial independence, and homogeneity over time and space. As an illustration, it is shown that per capita income in the forty-eight contiguous U.S. states did clearly not follow a common stationary first-order Markov process from 1929 to 2000.