We document a unique determinant of financial systemic risks in China, share pledge financing (SPF) network, by studying all listed Chinese financial institutions that provide SPF business. As one of the most popular refinancing tools in China, SPF formulates a network among financial institutions with common collaterals. We propose a centrality measure to quantify such network effects, and document that banks are more important than securities in the SPF network before 2018, but this pattern has reversed afterwards. SPF network effects, rather than SPF margin closeout risks, significantly increase systemic risks of China's financial institutions, by increasing both individual risks of financial institutions and connectedness among them. Such an impact of the SPF network on systemic risks is more pronounced in OTC markets than in exchanges. The SPF network affects systemic risks in more extreme cases, while SPF margin closeout risks affect systemic risks in more moderate cases.

This paper defines asymptotic marginal expected shortfall (AMES) for banks within a financial system and provides corresponding estimation method based on multivariate extreme value theory. The estimation method does not assume a specific dependence structure among bank equity returns. Both theoretical AMES and the estimator possess additive property and thus can serve as a tool to allocate system-wide risk to individual institutions. We apply the AMES to 30 global systemically important financial institutions (G-SIFIs). We show that the AMES outperforms the MES in predicting extreme losses during extreme systemic events. By taking the AMES as the reference point for allocating systemic risk to individual institutions, we show that an allocation according to simple bank characteristics such as size and individual risk can be imperfect. The allocation unfairness of individual risk or size across all the G-SIFIs has increased since 2008.