Sparse BGVAR models for Systemic Risk Analysis

Measuring systemic risk requires the joint analysis of large sets of time series which calls for the use of high-dimensional models. In this context, inference and forecasting may suffer from lack of efficiency. In this paper we provide a solution to these problems based on a Bayesian graphical approach and on recently proposed prior distributions which induces sparsity in the graph structure. The application to the European stock market shows the effectiveness of the proposed methods in extracting the most central sectors during periods of high systemic risk level.