Statistical Inference in Multivariate Time Series Models

Specialeforsvar: Anne Jensen

Titel: Statistical Inference in Multivariate Time Series Models

Resume: The interest of analysing heavy-tailed data has evolved a lot since computers were developed. Further, a lot of facts about heavy-tailed data, e.g. financial data, have seen the light after the possibility to handle large data frames and create graphical outputs became a reality. It is now a known fact that the development of financial time series is based on their past behaviour. Therefore they can be explained in terms of
stochastic recurrence equations. It is of great interest to determine models that can take this behaviour into account. Modelling the volatility of these models is difficult in higher dimensions, as more complicated models are needed. However, in recent time, multivariate volatility models have been introduced to handle the heteroscedastic behaviour of financial time series through stochastic recurrence equations. Some multivariate volatility models are presented in this thesis, and a study based on 50 stocks from the S&P-500 index has been performed. The study includes a univariate estimation approach using the Auto-Regressive Conditional Heteroscedasticity (ARCH) model, and a multivariate estimation procedure based on the multivariate BEKK-ARCH (and rotated BEKK-ARCH) model, including variance targeting for optimization.

Vejleder: Olivier Wintenberger
Censor: Mette Havning