Interconnectedness and systemic risk in financial networks

This thesis investigates three different forms of interconnectedness and their implications for systemic risk in the corresponding financial markets. It comprises three key chapters based on three research papers. 

The first study applies dynamic network analysis to the power sector, examining the relationship between regional spot electricity prices in the Australian National Electricity Market (NEM). In particular, we employ principal component analysis and generate Granger causality networks to examine the degree of interconnectedness of the NEM in a time-varying setting. We find that the derived measures of interdependence can be related to actual market events. In the analysed network, we find that stronger dependence is exhibited by regional markets that are linked through interconnectors, while the direction of Granger causality can be related to interregional trade. We further examine the usefulness of the derived measures for forecasting distributional characteristics of spot prices such as the maximum price, volatility, price spreads, or upcoming periods of price spikes. Our results suggest that the derived network measures have predictive power, albeit limited, for the behaviour of spot electricity prices in the NEM.

The second study focuses on interbank exposure as one of the main channels of risk contagion. We examine the impact of negative shocks on the distribution of banks' default probabilities, by applying simulated shocks to interbank lending matrices, based on values of total interbank assets and liabilities of individual banks. The results show that the structure of interbank lending matrices can remarkably impact on default probabilities, and even on systemic risk. Loss events have an influence on a larger number of banks if the lending network has a connection probability around 0.5. In a complete network, the effect of risk sharing exceeds that of risk contagion. To conclude, interbank exposures not only enable banks to share risk but also provide a channel for spreading risk. This role switch depends on the interaction of the following factors: the shock size, the number of attacked banks, the category of attacked banks, the recovery rate of liquidation, and the structure of the banks' balance sheets.

The third study analyses sectoral similarity of commercial banks' business loans in China and describes cross-sectional characteristics of constructed networks based on different measures as well as the trends of these characteristics over time. The findings show that the similarity has spatial effects where banks that serve similar areas have similar loan structures. Loan similarity is also related to differences in several financial indicators, such assize, profitability, the structure of assets and liabilities, leverage ratio, expected credit risk, and sectoral concentration of loans. All these factors can explain approximately 20% of the cross-sectional variation of loan similarity on average, up to 33% in 2013. Among the considered factors, differences in size, sectoral concentration of loans, the ratio of loans to assets, and expected credit risk significantly account for similarity in business loan structure during our sample period. We also provide evidence on negative externalities of loan similarity, suggesting that while loan similarity might lower individual risk and systemic vulnerability of a bank, at the same time it might increase an individual banks contribution to systemic risk.