Of needles and haystacks: revisiting growth determinants by robust Bayesian variable selection

This paper proposes a Bayesian variable selection technique that is robust to model uncertainty and heteroscedasticity of unknown forms in cross-country growth regressions. In particular, we adopt a spike-and-slab prior for the regression coefficients to sidestep the potential complications of highly collinear covariates in the variable selection problems. We find a number of important growth determinants that are distinct from the previous works in the ranking and sign of estimated coefficients, partly because of the presence of highly correlated covariates. Importantly, we find that our results from variable selection are not qualitatively changed when various degrees of freedom for t-errors are considered. The implications of this work highlight the caution needed in empirical growth research when using g-priors under collinearity.