Network clearing algorithms and statistical methods for risk assessment

Thesis

This thesis presents methodological contributions for the quantification of systemic risk in financial systems. Clearing algorithms that can be used for computing financial contagion effects are presented and their relation to network centrality measures is discussed. An algorithm that allows computing clearing solutions for networks of financial contracts including debt of multiple seniority classes, equity participations as well as contingent convertible debt and bail-in-able debt is presented. We present a valuation function that allows computing network-based valuations of contingent convertible and bail-in-able debt instruments. An application to systemic risk assessment for bail-in decisions using real-world data is presented, showing that bail-ins have the potential to reduce systemic risk in some crisis situations. An alternative approach to systemic risk assessment based on a metamodel of clearing algorithms is presented. We use quantile panel regression for learning the metamodel and present statistical tests for deciding between different quantile panel estimators as well as goodness-of-fit measures. An application with real-world-data shows how regulators could improve their identification of systemically important banks.

Authors: Siebenbrunner, C

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: mathematical finance; networks; machine learning; systemic risk
• Subjects: Finance; Mathematics