Structural Models of Volatility

点击翻译按钮获取中文摘要

The development of multivariate volatility or correlation models has become a rapidly growing branch in finance, both in theory and in the applied fields of portfolio allocation and optimization, portfolio risk evaluation, and asset pricing. This development has been spurred especially by the introduction of multivariate models featuring conditional heteroskedasticity (MGARCH). While conveying insightful information about the underlying volatility dynamics, MGARCH models are, however, limited in the sense that in most studies the underlying model of shock transmissions lacks identification in a strictly structural sense.In this international project we build upon recent advances in identifying macroeconometric structural vector autoregressive models, and develop two alternative ways to identify structural stochastic volatility models of the multivariate GARCH type. We proceed from two perspectives. On the one hand, we study a purely statistical approach that proceeds from the assumption that second order dynamics of speculative returns can be traced back to unique and independent structural shocks. From this assumption, we derive moment conditions that identify the structural MGARCH model. We study the method in a static setting and dynamically, taking into account potential structural shifts. On the other hand, we approach identification by exploiting the information inherent to news analytics data. In a third step, both applicants and their research groups will integrate their insights in applying the two schemes to two major fields of empirical research: oil price shocks and the banking crisis in 2008/2009. On the one hand external information is expected helpful for the economic labeling of statistically identified shocks. On the other hand a systematic comparison of independent and instrumental shocks could add a solid conceptual support for the latter when it comes to the descriptive analysis by means of common impulse responses which, by construction, rely on the assumption of isolated (i.e. independent in the non-Gaussian case) unit shocks. We plan to provide our research via an R framework for the analysis of structural volatility models to other researchers. Based on this, an interactive web application demonstrates the detection, identification and visualization of structural changes in volatility transmissions with real-time data. By developing open source research software, we aim to disseminate our research and open up a new communication channel with the scientific community.