In this paper, we propose the spatial state-space model as a novel framework to quantify time varying spillover effects in economic networks. Similar to spatial econometric models, this framework revolves around measuring the spatial autoregressive parameter, which we interpret as spillover intensity for a given network. However, this parameter is often assumed to be constant, which can become restrictive for the analysis of time-variant systems. The key innovation of spatial state-space models is that they model this parameter as a dynamic latent variable with a stochastic component. For the resulting nonlinear state-space model with stochastic volatility, we provide an estimation methodology based on the extended Kalman filter and numerically accelerated importance sampling for likelihood evaluation, and particle filter for signal extraction. The good finite sample properties of our estimation method are demonstrated in a simulation study. As an application, we examine financial contagion between the 24 largest banks in the eurozone using daily credit default swaps (CDS) data for the period 2014Q1-2016Q2. Based on confidential supervisory data, we construct banking networks on a quarterly basis to capture two important risk channels: The interbank lending channel and the portfolio overlap channel. Preliminary findings indicate that our model can capture time-varying spillover intensities for both channels. Furthermore, we find that the inclusion of the spillover intensity leads to improved in-sample predictions of CDS spread changes. It performs particularly well during the interbank stress period in 2016Q1.

Discussant: Rob Sperna Weiland (University of Amsterdam)