Panel Growth Regressions with General Predetermined Variables: Likelihood-Based Estimation and Bayesian Averaging

In this paper I estimate empirical growth models simultaneously considering endogenous regressors and model uncertainty. In order to apply Bayesian methods such as Bayesian Model Averaging (BMA) to dynamic panel data models with predetermined or endogenous variables and fixed effects, I propose a likelihood function for such models. The resulting maximum likelihood estimator can be interpreted as the LIML counterpart of GMM estimators. Via Monte Carlo simulations, I conclude that the finite sample performance of the proposed estimator is better than that of the commonly used standard GMM. In contrast to previous consensus in the empirical growth literature, once endogeneity and model uncertainty are accounted for, empirical results indicate that the estimated convergence rate is not significantly different from zero. Moreover, there seems to be only one variable, the investment ratio, that robustly causes long-run economic growth. JEL Codes: C11, C33, O40