Income inequality may not converge after all: Testing panel unit roots in the presence of cross-section cointegration

This paper empirically examines whether there exists stochastic convergence of income inequality among 48 contiguous US states from 1916 to 2012. To control long-run cross-section dependency in panel data, we apply the orthogonal instrument generating approach of Chang and Song (2009) to test unit root. Our results, unlike the literature, do not support the convergence hypothesis. Moreover, we confirm that long-run cross-section correlation has substantial impacts, which are robust to all inequality measures as well as quantile differentials. In addition, although there is a given rising trend, the income distribution of the United States is state-specific and does not converge to either the national level or the state-average.