Structural Breaks and Time Variation in High-Dimensional Dependence Structures

• Applicant: Professor Dr. Dominik Wied
• Subject Area: Statistics and Econometrics
• Term: from 2016 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 313610190

In the last 15 years, a lot of work has been done on modeling dependencies, in particular with copula models. These models have advantages compared to alternative models based on the multivariate normal distribution or Gaussian factor models. In particular, they allow for separating dependencies and marginal distributions resulting in considerable flexibility, which permits applications in distinct research fields. Moreover, the models allow for nonlinear dependencies of various forms, in particular dependencies of extremes and asymmetric dependence structures. These models have turned out to be suitable in risk management applications because financial time series are often characterized by joint extreme values.The last few years brought two important developments in the copula literature. First, many models allow for time-varying dependencies. Second, several models allow for flexible dependencies in more than two dimensions. The aim of this project is to develop methods which lie at the intersection of these two developments. In particular, we plan to further advance the analysis of tests for structural breaks with a huge number of variables. The basic question is whether it is necessary to model time-varying dependencies in a complex way or whether, e.g., a model with a single breakpoint is sufficient. To this end, suitable methods have to be developed or modified for specific existing models. Theoretical properties of these methods need to be analyzed and tools for statistical inference will be developed. Applications to risk management problems and financial contagion shall demonstrate the practical usefulness of our methods. We expect the new methods to have considerable advantages compared to existing approaches. In risk management applications, one needs models which are sufficiently flexible, but that can be handled in high-dimensional problems. At the same time, high-dimensional data are expected to be beneficial for detecting structural changes. First and foremost, when one is interested in detecting correlation changes in real time, one wants to use available data as efficiently as possible. Precise detection of breakpoints is also crucial for considering contagion effects on financial markets. The new methods may turn out to be useful for better understanding contagion effects in a historical analysis, but also for the timely detection of contagion risk in real time.

• DFG Programme: Research Grants
• International Connection: Austria
• Cooperation Partner: Professor Dr. Hans Manner