Multivariate Score-Driven Models for Count Time Series To Assess Financial Contagion

Abstract

This paper develops a multivariate model for count time series, in which the time-varying intensity parameter determining the probability that an event occurs evolves according to general autoregressive score (GAS) models (see Creal et al., 2013; Harvey, 2013).

The model is particularly suitable to study the spread of financial contagion between different economic sectors or markets. Indeed, the interdependence between extreme market event counts arises from the effect of shocks in a sector on the probability of shock occurrence in the others.

By applying the model to daily CDS data relative to a sample of European companies, we find significant cross-sector effects. In particular, the Financial and Energy sectors are those whose shocks impact others the most, while the sectors most affected by extreme events in other markets turn out to be ICT and Trade.

Keywords: Score-driven models, Count time series, Contagion, Credit risk

JEL Classification: C10, C32, G32