Essays in GMM estimation of dynamic panel data models

The aim of the work is twofold. First, we investigate the properties of the dynamic panel data (DPD) GMM estimator when the instrument count is high. We introduce the extraction of principal components fromthe instrument matrix as an effective strategy to reduce the number of instruments. Through Monte Carlo experiments, we want to compare the performances of the GMM estimators when the instrument set is factorized, collapsed or limited. Second, we estimate fiscal response functions on simulated panels and on real data to identify the best-performing estimator in this context, where endogeneity and instrument proliferation issues are unavoidable. The dissertation consists of three chapters. The first reviews the literature of DPD estimation and presents the issue of instrument proliferation in DPD GMM estimation. The second introduces the principal component analysis (PCA) to reduce the dimension of the instrument matrix and compares the performances of the factorized, limited and collapsed GMM estimators, finding them similar. Though the simulated models are extremely simplified, the PCA seems to be promising. The third chapter simulates fiscal response functions and investigates the properties of DPD estimators in fiscal rules estimation; the fiscal rules are then estimated on real data for EMU Countries. The system GMM estimator is the best-performing here. Instrument proliferation does not bias the estimates; collapsing and lag truncation of the instrument matrix can lead to misleading results, while the factorized estimator performs well. Discretionary policies within the EMU are systematically found a-cyclical.