Measuring Dynamic Spillovers in Networks: Sovereign Contagion, Regulation Effects and Market Sectors

Abstract

In the recent subprime mortgage crisis and sovereign debt crisis, effective threats to and the loss of confidence in the stability of the financial system have increased, which is also denoted as a rise of systemic risk. Systemically important institutions were first categorized only as ``too big to fail''. This was insufficient during the crisis and required an extension by ``too interconnected to fail', reflecting the importance of interconnectedness for safeguarding financial and economic stability. However, respective data on interdependencies are not publicly available or incomplete. Thus, there is a need for econometric techniques that can handle incomplete and complex, large datasets to detect interconnections and to quantify spillovers between entities. While there have been various econometric approaches to measure spillovers, in this thesis quantitative measures are developed and evaluated that shed light on important and so far disregarded aspects. Technically, a connectedness measure is introduced that is based on out-of-sample forecast errors, which responds more quickly to crisis occurrences than already existing measures based on in-sample forecast errors. Furthermore, I explore spillover measures computed from Bayesian time-varying parameter models, leading to dynamic measures which capture changes in the underlying data immediately, which is in contrast to previous static and rolling-window measures. So far, the significance of changes in spillovers has not been evaluated. In this thesis, this deficiency is overcome by constructing confidence intervals for spillover measures, using either a bootstrap methodology or the posterior distribution of Bayesian estimates. Via the confidence intervals and a unique intraday dataset on credit risk of European sovereigns, this thesis delivers clear results concerning the effectiveness of recent policy and regulation interventions. The results show that regulatory changes in the credit default swap (CDS) market effectively reduced systemic risk, whereas policy interventions had a smaller and less sustainable impact. Moreover, in this work it is shown that daily CDS and bond yield spreads contain complementary information for measuring spillovers. To my knowledge, I am the first to identify the differences between volatility-type spillover measures based on these two datasets and thus to demonstrate the importance of considering both data sources for obtaining a complete picture on sovereign connectedness. In this thesis, empirical guidance is provided for the choice of a suitable time-varying parameter estimation approach for measuring spillovers between U.S. stock market sectors. The analysis shows the importance of not only the financial sector, on which most studies focus, but also commodity industries and other sectors of the real economy for a comprehensive understanding of spillover effects.