This thesis develops methods for accurately estimating individual-specific coefficients in heterogeneous panel data models. Panel data combine time series and cross-sectional information, offering a powerful framework for empirical analysis, but they also raise important methodological challenges. In particular, standard panel methods often rely on assumptions of homogeneity across units. When these assumptions are violated, conventional estimators can produce biased results, especially when the goal is to understand or predict the behavior of a specific unit. The thesis focuses on addressing the fundamental trade-off between time series estimators, which are unbiased but inefficient, and panel estimators, which are efficient but may be biased when heterogeneity is present. To overcome this problem, the thesis proposes three hybrid methodologies that combine information across units while preserving the ability to estimate unit-specific effects. The proposed methods determine how much information from the rest of the panel should be used for estimating the parameters of a particular unit. By selectively weighting or pooling information from similar units, the estimators achieve greater precision without imposing restrictive homogeneity assumptions. By explicitly accounting for individual differences, these methods support more accurate inference, and policy analysis, providing a robust framework for targeted estimation across heterogeneous panel settings.
More information on the thesis