Conditional graphical models for systemic risk measurement

Financial network models are a useful tool to model interconnectedness and systemic risks in financial systems. They are essentially descriptive, and based on highly correlated networks. In this paper we embed them in a stochastic framework, aimed at a more parsimonious and more realistic representation. First we introduce Gaussian graphical models in the field of systemic risk modelling, thus estimating the adjacency matrix of a network in a robust and coherent way. Second, we propose a conditional graphical model that can usefully decompose correlations between financial institutions into correlations between countries and correlations between institutions, within countries. While the former may be further explained by macroeconomic variables, the latter may be further explained by idiosyncratic balance sheet indicators. We have applied our proposed methods to the largest European banks, with the aim of identifying central in situations, more subject to contagion or, conversely, whose failure could result in further distress or breakdowns in the whole system. Our results show that, in the transmission of the perceived default risk, there is a strong country effect, that reflects the weakness and the strength of the underlying economies. In addition, each country reveals specific idiosyncratic factors, with communalities among similar countries