Explaining the Negative Coefficient Associated with Human Capital in Augmented Solow Growth Regressions

In this paper we consider different explanations for why the coefficient associated with human capital is often negative in growth regressions once country-specific effects are controlled for whereas the coefficient in question is strongly positive in cross-sectional or panel results based on the pooling estimator. In turn, we explore: (i) additional sources of unobserved heterogeneity stemming from country-specific rates of labor-augmenting technological change, (ii) measurement error in the human capital series being used, and (iii) the lack of variability in the human capital series once the usual covariance transformations are implemented. Remaining unobserved country-specific heterogeneity and measurement error alone are shown to be inadequate explanations. The lack of variability in the human capital series is tackled using a new GMM-based estimator that combines the Hausman-Taylor (1981) approach, in which the impact of time-invariant covariates can be identified through use of covariance transformations of the variables themselves as instruments, with the orthogonality conditions of the Arellano-Bond (1991) estimator.