A0356
Title: Stochastic block network vector autoregressions
Authors: Florian Huber - University of Salzburg (Austria)
Gary Koop - University of Strathclyde (United Kingdom)
M. Marcellino - Bocconi University (Italy)
Tobias Scheckel - University of Salzburg (Austria) [presenting]
Abstract: Commonly used priors in vector autoregressions (VARs) induce shrinkage on the autoregressive coefficients. Introducing shrinkage on the covariance matrix is sometimes done but, in the vast majority of cases, without considering the network structure of the shocks. A prior is proposed on the covariances in a VAR that takes the topological structure between the reduced form shocks of the model into account. The prior resembles a standard Spike and Slab prior. The indicators which govern variable inclusion are modelled through a stochastic block model which clusters shocks into groups. Within groups, the probability of having relations across group members is higher (inducing less sparsity) whereas relations across groups imply a lower probability that members of each group have non-zero covariances. It is shown in simulations that the approach recovers the true network structure well and this translates into more precise estimates of the error covariance matrix. In a real US data macro example, it is illustrated how the approach can be used to cluster shocks together and that this feature leads to better forecasts.
