A0718
Title: Large Bayesian structural matrix autoregression
Authors: Ignacio Moreira Lara - Universitaet Duisburg Essen (Germany) [presenting]
Christoph Hanck - Universität Duisburg-Essen (Germany)
Jan Prueser - University Duisburg-Essen (Germany)
Abstract: The structural identification of a high number of shocks in a vector autoregression (VAR) becomes increasingly challenging as the number of parameters proliferates and the set of restrictions needed for identification grows. These challenges hinder both the feasibility and interpretability of spillover analysis in high-dimensional systems. A large structural Bayesian matrix autoregression is introduced that overcomes these obstacles by exploiting a Kronecker structured autoregressive structure, dramatically reducing both parameter and restriction counts. The algorithm is highly efficient and scales to large panels of time series. Contemporaneous relations are identified using established SVAR methods with minimal adjustments for the framework. To demonstrate its capabilities, the BMAR on monthly macro data is estimated for a broad panel of European countries, simultaneously extracting country-specific supply and demand shocks. The model delivers clear impulse responses and historical decompositions, providing concrete evidence of the asymmetric spillovers and spillbacks of multiple contemporaneous shocks across the euro-area.
