Conventional tests of convergence hypothesis using linear cross-country growth regressions often report favorable evidence consistent with the hypothesis, namely, poor countries growing at a faster rate than the rich ones so that the differences in the levels of per-capita income disappear over time. However, recent studies based on theoretical background of multiple-regime equilibria, tend to suggest the existence of multiple steady states, i.e., rejecting the convergence hypothesis. Econometric methodology along this line includes the regression tree analysis, semiparametric partially linear method and threshold regression.
In contrast, this paper revisits the issue via a flexible nonlinear approach recently developed by Hamilton [A parametric approach to flexible nonlinear inference. Econometrica, 69 (2001) 537–573]. The novel method not only provides a Lagrange multiplier test for nonlinearity but also allows us to derive a consistent estimation of what the nonlinear relation looks like. A real data set is used to investigate if the cross-country growth regressions exhibit nonlinearity and to see if there are “convergence clubs” and “diverging economies”. Empirical results indicate that the linearity assumption of cross-country growth regressions can be rejected to favor the multiple-regime steady states.