"In recent years, there has been an increased availability of economic datasets that are characterised by a non-negligible amount of time-series information, as well as a large cross-section dimension. Unfortunately, in many cases, although the time-series dimension is not small anymore, it is not large enough for accurate estimation either. Examples of such datasets can be found in a variety of literatures: growth data, firm data (e.g. studies of insider trading activity), earnings studies (the popular Panel Study of Income Dynamics) and hedge fund returns, to name a few. Such panels are generally designated as short panels in the literature. Intuitively, the problem with short panels is a time-series finite-sample bias. This project’s main aim is to conduct a systematic theoretical and empirical research programme on modelling short panels in the presence of unknown and complex types of dependence across both time and cross-section. Inclusion of cross-section dependence is the central novelty of this project. This is motivated by my doctoral research where I showed that this type of dependence introduces new bias terms. The central theoretical objectives are to develop bias correction methods for (i) the Dynamic Autoregressive and Dynamic Probit/Logit models which are quite popular in applied research and (ii) the multivariate nonlinear and dynamic panel data models. The main empirical objectives are to a large extent concerned with univariate and multivariate volatility modelling using panels of financial data. As such, this project will contribute to the theoretical panel data and volatility modelling literatures, as well as the statistics literature, where the outlined small-sample bias has a long research tradition. In addition, as this project is concerned with modelling and understanding macro and financial panels, this project also has a significant potential to contribute to the key policy area “Connect to compete: building tomorrow’s networks today.”"