Dynamic Panels with Predetermined Regressors: Likelihood-Based Estimation and Bayesian Averaging With an Application to Cross-Country Growth

Abstract

This paper discusses likelihood-based estimation of linear panel data models with general predetermined variables and individual-specific effects. The resulting (pseudo) maximum likelihood estimator is asymptotically equivalent to standard GMM but tends to have smaller finite-sample biases as illustrated in simulation experiments. Moreover, the availability of such a likelihood function allows applying the Bayesian apparatus to this class of panel data models. Combining the aforementioned estimator with Bayesian model averaging methods we estimate empirical growth models simultaneously considering endogenous regressors and model uncertainty. Empirical results indicate that only the investment ratio seems to robustly cause long-run economic growth. Moreover, the estimated rate of convergence is not significantly different from zero.