An Analysis of Modularity and the Occurrence of Extreme Events in Banking Networks

Abstract

In 2007-2008, the global financial system experienced the most devastating event since the early 20th century. Academics and regulators often cite risky lending practices, highly leveraged trading, and synthesized asset-based securities and derivatives as triggers of the crisis. Regulatory measures have attempted to focus on firm-specific risk to determine appropriate capitalization and leverage to prevent firm insolvency. However, these atomistic risk management tools have not historically prevented financial crises and may struggle to successfully prevent them in the future.

Another strand of research has argued that the financial system's propensity for collapse is a fundamental property of its structure. The financial system, like other complex adaptive systems, is vulnerable to contagion under distress. Financial network structure has created a dual-natured system in response to shocks. When faced with less severe shocks, the network structure facilitates risk-sharing between institutions. However, when faced with shocks above a certain size threshold, the network structure amplifies the shock, spreading contagion between institutions. Faced with an extremely large shock, the amplification mechanisms in the network ultimately generate a catastrophic event.

Research has demonstrated that shocks in financial networks typically follow a heavy-tailed distribution, so shocks are frequently large. Under these distributions, shocks are more likely to cross the network's critical threshold and trigger contagion. In other complex adaptive systems, network structure is modified in last-resort scenarios to stop the spread of contagion and prevent a system-wide collapse.

Modularity utilizes network structure to create a preventive, last-resort protection against financial crises. This thesis analyzes the impact of various types of modularity on the stability of the banking network. Vine copulas are used to model different network structures and their resulting return distributions. Various types of network and dependency structures are considered by varying both the number of modules within the network and by varying the copula families and parameters observed. Critically, copula families accounting for higher degrees of tail dependence are used to model realistic financial systems.

The results of this thesis show that network structure is a key determinant of the frequency of extreme events. While the specific copula families and parameters observed impact the resulting return distribution, modular network structures are consistently more stable than complete network structures. In addition, the results imply a trade-off between minimizing the frequency of extreme events and minimizing volatility. Ultimately, modular network structures provide a potential method to increase the robustness of the financial network and protect against financial crises.