Can Bank-Specific Variables Predict Contagion Effects?

Assessing the systemic risk a bank poses to the system has become a central part in regulating its capital requirements (e.g. the buffer for global or domestic systemically important banks). As with conventional risk types, systemic risks need to be quantified. Currently global regulators propose a range of bank-specific indicators that measure size and interconnectedness to proxy systemic risk. In this study we gauge the capacity of such indicators to explain contagion losses triggered by realizations of sizeable idiosyncratic shocks. We study contagion impact through different channels, separating these effects into first-round, nth-round, asset fire sale and mark-to-market losses. We provide constructive proofs for the existence of clearing payment vectors and associated market equilibria for these contagion channels in a model of interlinked balance sheets. We provide algorithms that converge to the greatest market equilibrium in a finite number of steps. We evaluate the predictive power of models selected by best-subset selection and Lasso by applying 10-fold panel cross validation. Our empirical results suggest that the Basel III indicator set performs well in comparison to alternative data sets of bank-specific indicators. We also find, however, that the proposed data sets without bank dummies do not perform well in capturing the relevance of the average network position for predicting contagion effects