Advances in panel and network econometrics

In most of the chapters of this dissertation, I contribute to the literature on the econometrics of network models. While network data has some similarities to traditional panel data, estimating network models poses added challenges demanding new econometric techniques. In particular, part of the network dependence structure is often considered by including unit-specific effects for each pair of units, which imposes new challenges in estimating non-linear models. While most available methods provide consistent estimates and valid inference under settings of dense networks, there are still a lot of open questions regarding the estimation for sparse settings. The second chapter of this dissertation illustrates how to obtain the asymptotic properties of estimators for network (dyadic) models using tools from the literature on U-statistics, and the third chapter proposes a model and an estimation method for the conditional cumulative distribution function of outcomes generated by a sparse network.
Finally, the fourth chapter of this dissertation focuses on a different topic in econometrics of panel data models, namely, the estimation of treatment effects and the use of the synthetic control (SC), demeaned synthetic control (DSC), and synthetic difference-in-differences (SDID) methods. More specifically, this chapter re-evaluates the effect of the Brexit referendum on the UK's GDP by considering these different estimation methods. Even though when initially proposed in the literature, the DSC and the SDID do not allow for matching on covariates, we fill this gap by providing an estimation method that takes covariates into account. Moreover, we show that the SDID estimator minimizes both interpolation and extrapolation biases, while the SC method only minimizes the latter.