Essays in Dynamic Panel Data Econometrics

This thesis consists of three essays in dynamic panel data econometrics. It studies the estimation of persistent dynamic relationships in panel models with unobserved heterogeneity and predetermined regressors, combining methodological contributions with an application in growth empirics.

The first paper analyzes structural breaks in persistence in a dynamic panel first-order autoregressive model when the time dimension is fixed and the cross-sectional dimension tends to infinity. A least squares breakpoint estimator is proposed and shown to be consistent without requiring increasing the number of time observations, allowing rapid detection of changes in dynamic behavior across stationary, unit root, and explosive regimes.

The second paper develops a generalized first-difference least squares (FDLS) estimator for dynamic panel autoregressive and vector autoregressive models. The estimator is consistent under both stationary and unit root processes, regardless of panel dimensions, and performs well in finite samples, outperforming general methods of moments estimators in highly persistent settings.

The third paper applies the FDLS framework to study the relationship between redistribution, inequality, and economic growth. The results show that redistribution is negatively associated with growth, but this effect disappears once investment is included, indicating that redistribution primarily affects growth through capital accumulation. A mediation framework is introduced to distinguish direct and indirect effects.