Testing for the Validity of W in GVAR models

Abstract

Global vector autoregressive (GV AR) models are one of the most established econometric frameworks to measure shock transmission in economics and finance. Nevertheless, they are based on ad hoc matrices of distance W, proxying for the tightness of the interconnections among nodes/countries/markets. This paper shows that GV AR can be restated as a specific form of structural VAR and derives the corresponding restrictions. Therefore, it elaborates a simple likelihood ratio test for the validity of the proposed W. Simulation studies show satisfactory asymptotic size and power, and we propose the use of the wild bootstrap version of the test to avoid finite sample distortions. The empirical exercise addresses euro area sovereign bond spread modeling on the onset of the euro area debt crisis. The new test rejects the validity of the W matrix used by Favero (2013) based on fiscal distance and that estimated as in Gross (2019). The nonrejected W matrix however yields important insights. Apartfromthetightnessofinterconnection, thismatrixtakesintoaccountthesplit between foreign countries having a positive (contagion) and negative (flight-to-safety) effects on domestic sovereign yield spreads, highlighting the asymmetric shock transmission in Euro Area government bond markets even before the occurrence of the 2010 sovereign debt crisis.