Dynamic Heterogeneous Panel Data Models with Cross-Section Dependence

This dissertation generalizes the asymptotic theory of dynamic cross-section heterogeneous coefficient panels with interactive fixed effects to allow for the cross-section units to be correlated due to local shocks. Traditional heterogeneous coefficient estimators in the multifactor error structure literature typically only focus on the influence that global shocks have on model estimates; global shocks such as the impact of the COVID-19 pandemic on a panel of world economic growth rates.

We extend this focus to also allow for local shocks that significantly impact only a small subset of the cross-section units in the sample. Local shocks like the impact of a falling crop price on the significant producer countries of that crop or the impact of a drought on the drought-affected nations. We derive the limiting distribution for the cross-section heterogeneous coefficients under √TN → c, 0 ⩽ c < ∞ asymptotics (where N and T are the number of cross-sections and time periods respectively). We observe a bias in the coefficient estimates associated with local shocks to the dependent variable (i.e., associated with the weak cross-section dependence of the idiosyncratic error). We provide sufficient conditions so that the bias and the covariance matrix of the limiting distribution can be consistently estimated in the presence of local and global shocks. Our theoretical findings are accompanied by extensive Monte Carlo experiments demonstrating the often superior finite sample performance of our estimation method over other competing techniques when the idiosyncratic errors are weakly cross-section dependent. We also provide an empirical application of our estimator and evaluate the country-specific long-run impact public debt has on economic growth for 86 countries.