Estimation of grouped, time-varying convergence in economic growth

Classical growth convergence regressions fail to account for various sources of heterogeneity and nonlinearity. Recent contributions advocating nonlinear dynamic factor models remedy these problems by identifying group-specific convergence paths. Similar to statistical clustering methods, those results are sensitive to choices made in the clustering/grouping mechanism. Classical models also do not allow for a time-varying influence of initial endowment on growth. A novel application of a nonparametric regression framework to time-varying, grouped heterogeneity and nonlinearity in growth convergence is proposed. The approach rests upon group-specific transition paths derived from a nonlinear dynamic factor model. Its fully nonparametric nature avoids problems of neglected nonlinearity while alleviating the problem of underspecification of growth convergence regressions. The proposed procedure is backed by an economic rationale for leapfrogging and falling-back of countries due to the time-varying heterogeneity of number, size, and composition of convergence groups. The approach is illustrated by using a current Penn World Table data set. An important aspect of the illustration is empirical evidence for leapfrogging and falling-back of countries, as nonlinearities and heterogeneity in convergence regressions vary over time.