Systemic Risk and Contagion in Interbank Networks [before doctoral defense]

Kivonat, rövid leírás

This dissertation develops three stylized, mechanism-focused models of default contagion in interbank liability networks, all built on the Eisenberg–Noe clearing framework. Each chapter isolates a distinct contagion mechanism while sharing a common balance-sheet structure of fixed nominal bilateral obligations, exogenous external assets, and chapter-specific default or valuation rules. Chapter 2, co-authored with P. Csóka, proposes a cooperative game-theoretic framework for measuring systemic importance via minimally required capital injections. For each shock scenario and coalition of banks, the minimum injection restoring solvency is com-puted, aggregated by Expected Shortfall, and allocated via the Shapley value. We prove that restricted and unrestricted injections are equivalent and that the resulting game is superadditive. A simulation demonstrates that capital-injection-based and loss-based indicators yield materially different systemic-importance measures. Chapter 3 introduces correlated external asset shocks through a one-factor Gaussian cop-ula and develops two complementary analytical approaches—a Bayesian network for-mulation and a mean-field approximation—to derive default probabilities without Monte Carlo simulation. The central finding is a non-monotonic, hump-shaped relationship be-tween pairwise asset correlation and default probability, peaking in the empirically relevant range. The underlying mechanism, termed heterogeneity in asset values at clearing (HAVC), arises from the tension between weakening diversification and disappearing contagion states as correlation increases. Chapter 4 extends the framework to multiple time periods with interim mark-to-market revaluation of interbank claims. A temporal-difference learning approach computes strongly self-consistent, state-dependent, forward-looking valuations, sidestepping the curse of di-mensionality. This chapter identifies a novel contagion mechanism—the synchronous valuation spiral (SVS)—whereby anticipatory markdowns cause defaults before maturity, potentially shifting the minimal default probability away from perfect correlation.