Systemic Risk and the Variability of SRISK

In the wake of the 2007-2009 financial crisis, many government and academic researchers have turned their focus towards defining and measuring systemic risk. They aim to ensure the soundness of the entire financial system and eliminate the need to bail out individual financial institutions. While there are a variety of systemic risk measures that identify and rank these Systemically Important Financial Insitutions (SIFIs), few have had the impact of the SRISK index created by Brownlees and Engle (2011). SRISK's dependence on publicly available data, its similarities to stress testing, and its straightforward interpretation have made it a leading metric for measuring an individual firm's systemic risk contribution.;While much of the research around SRISK has focused on its ability to monitor systemic risk, the validity of its SIFI rankings, and possible regulatory responses, there has been far less focus on the variability present in SRISK's estimation procedure. The computation of SRISK is neither simple nor straightforward, but instead, relies on a bivariate dynamic process and a simulation procedure to estimate a firm's Long Run Marginal Expected Shortfall (LRMES). By definition, a firm's LRMES is the percentage of equity that firm will lose, conditionally on the market falling into a crisis. In this dissertation, we explore how changes to the various simulation settings and statistical assumptions required to compute LRMES affect its variability. Specifically, we demonstrate that the use of a leptokurtic working likelihood in the GJR-GARCH model produces LRMES differences on the order of 20% for certain firm and date combinations. While these differences vary by firm and date, the typical leptokurtic LRMES estimate is smaller than its Gaussian counterpart, indicating a reduced systemic risk contribution. A study of asymmetric working likelihoods shows that while many firm and date combinations exhibit asymmetry, the corresponding differences in LRMES are negligible.;In order to further comment on LRMES variability, we also propose a new block bootstrapping methodology that allows for the propagation of DCC-GARCH parameter estimation error through the LRMES simulation procedure. Our Block Bootstrap for Estimating Equations (BBEE) methodology is unique in its approach to estimating DCC-GARCH parameter estimation error. Under typical LRMES settings, the BBEE methodology often outperforms asymptotic standard error approximations. Additionally, the use of our BBEE methodology allows us to better quantify the full amount of error present in estimating LRMES. The amount of LRMES variability due to DCC-GARCH parameter estimation is often larger than the LRMES variability due to model selection, and has the potential for billion dollar changes to SRISK estimates. By addressing the variability in LRMES that comes from both model selection and parameter estimation, we provide a better understanding of the SRISK index and systemic risk as a whole.