A1292
Title: GVAR models and linear transformations of VAR processes
Authors: Alexandros Konstantopoulos - University of Klagenfurt (Austria) [presenting]
Christian Zwatz - University of Klagenfurt (Austria)
Martin Wagner - University of Klagenfurt, Bank of Slovenia and Institute for Advanced Studies, Vienna (Austria)
Abstract: Multi-country, multi-regional or multi-sectoral data have become increasingly available over the last 20 years and are currently used in e.g. forecasting and policy analysis. The expanding cross-sectional dimension of these data sets necessitates complexity reduction techniques to tackle the curse of dimensionality. A prominent approach is the so-called global vector autoregressive (GVAR) model. From the perspective of an unrestricted and infeasible model, this approach relies upon different types of restrictions. The complexity reduction in GVAR models consists of imposing restrictions that collapse a joint, e.g., VAR model to a set of country-, region- or sector-specific VAR models with exogenous variables, with the impact of all other variables of the joint system condensed to a small number of variables constructed from the large set of variables and considered to be exogenous. This approach, thus, consists of imposing a large number of parameter restrictions and exogeneity restrictions. The issues concerning the conditions under which the country-specific VARX models a correct description of the subset of variables and under which exogeneity conditions prevail are addressed.
