Bayesian graphical models with economic and financial applications

Recent advances in empirical finance has seen a considerable amount of research in network econometrics for systemic risk analysis. The network approach aims to identify the key determinants of the structure and stability of the financial system, and the mechanism for systemic risk propagation. This thesis contributes to the literature by presenting a Bayesian graphical approach to model cause and effect relationships in observed data. It contributes specifically to model selection in moderate and high dimensional problems and develops Markov chain Monte Carlo procedures for efficient model estimation. It also provides simulation and empirical applications to model dynamics in macroeconomic variables and financial networks. The contributions are discussed in four self contained chapters. Chapter 2 reviews the literature on network econometrics and presents a Bayesian graph-based approach as an alternative method. Chapter 3 proposes a Bayesian graphical approach to identification in structural vector autoregressive models. Chapter 4 develops a model selection to multivariate time series of large dimension through graphical vector autoregressive models and introducing sparsity on the structure of temporal dependence among the variables. Chapter 5 presents a stochastic framework for financial network models by proposing a hierarchical Bayesian graphical model that can usefully decompose dependencies between financial institutions into linkages between different countries financial systems and linkages between banking institutions, within and/or across countries.