A Projection Pursuit Approach to Cross Country Growth Data

The empirical modeling of the cross-country differences in growth behavior is undoubtedly one of the most predominant research topics in applied macro-econometrics. However, despite the vast research effort it seems that there are only a few firm conclusions on the sources of cross-country differences. Unlike the bulk of the literature which focuses on linear parametric models this paper studies a semi-parametric way of modelling. In particular, it employs projection pursuit regression (PPR) to model the mean regression function of the growth process by a sum of unknown ridge functions (functions of linear combinations of covariates). PPR model was proposed by Friedman and Stuetzle (1981) to approximate high dimensional functions by simpler functions that operate in low dimensional spaces-typically one-dimensional. My findings identify non-linear relationships among the basic Solow-type variables. In particular, initial income and human capital affect growth in a very nonlinear way. Furthermore, there is evidence of interaction effects between human capital and initial income as well as between initial income and population growth rates. The findings suggest the presence of two steady-state equilibria that classify countries into two groups with different convergence characteristics.