Spillover dynamics for systemic risk measurement using spatial financial time series models

A new model for time-varying spatial dependencies is introduced. It forms an extension to the popular spatial lag model and can be estimated conveniently by maximum likelihood. The spatial dependence parameter is assumed to follow a generalized autoregressive score (GAS) process. The theoretical properties of the model are established and its satisfactory finite sample performance is shown in a small simulation study. In an empirical application, spatial dependencies between nine European sovereign CDS spreads are estimated for the time period from November 2008 until October 2013. The empirical model features a spatial weights matrix constructed from cross-border lending data and regressors including country-specific and Europe-wide risk factors. The estimation results indicate a high, time-varying degree of spatial spillovers in the spread data. A spatial GAS model with t-distributed errors provides the best fit. There is evidence for a downturn in spatial dependence after the Greek default in winter 2012, which can be explained economically by a change in bank regulation.