Measuring Network Systemic Risk Contributions: A Leave-One-Out Approach

Abstract

Granger-causality measures of interconnectedness between financial institutions are useful indicators of systemic risk (Billio et al., 2012) [Journal of Financial Economics], as they help in evaluating how far the distress of one institution is disseminated across the whole of the financial system through networks. This article provides a critical assessment of Granger-causality networks, showing that they can lead to inconsistent measures of systemic risk contributions because of the presence of spurious causalities arising from indirect contagion effects. The traditional solutions for controlling for these effects using inference on conditional Granger-causality lead to the curse of dimensionality though. To solve this, we provide a measure of financial network systemic risk contributions that is based on the leave-one-out (LOO) concept. For a given financial institution, the new measure evaluates how far the total number of significant Granger-causalities breaks down when this institution is excluded from the system. We control for spurious causalities between the remaining institutions due to the indirect contagion effect of the excluded financial institution using a conditional Granger-causality test, which is free of the curse of dimensionality. Empirical applications are conducted using daily market returns for a sample of the world’s largest banks. The results show that our measure gives a meaningful ranking of the systemic importance of financial institutions that is consistent with the ranking of global systemically important banks (G-SIBs) provided by the Financial Stability Board (FSB). Moreover, our measure is shown to be a robust and significant early-warning indicator of large losses from a systemic event, and is strongly driven by balance-sheet variables related to size, business model and profitability.