A hybrid approach to assess systemic risk in financial networks

We propose a credit risk approach in which financial institutions, modelled as a portfolio of risky assets characterized by a probability of default and a correlation matrix, are the nodes of a network whose links are credit exposures that would be partially lost in case of neighboursu0027 default. The systemic risk of the network is described in terms of the loss distribution over time obtained with a multi-period Montecarlo simulation process, during which the nodes can default, triggering a change in the probability of default in their neighbourhood as a contagion mechanism. In particular, we have considered the expected loss and introduced new measures of network stress called PDImpact and PDRank. They are expressed in monetary terms as the already known DebtRank and can be used to assess the importance of a node in the network. The model exhibits two regimes of u0027weaku0027 and u0027strongu0027 contagion, the latter characterized by the depletion of the loss distribution at intermediate losses in favour of fatter tails. Also, in systems with strong contagion, low average correlation between nodes corresponds to larger losses. This seems at odds with the diversification benefit obtained in standard credit risk models. Results suggest that the credit exposure network of the European global systemically important banks is in a weak contagion regime, but strong contagion could be approached in periods characterized by extreme volatility or in cases where the financial institutions are not adequately capitalized.