An important task in the analysis of multivariate time series is to study its dependence structure, i.e. serial, cross-serial and cross sectional dependence. The copula functions introduced by Sklar (1959) are a nice tool to capture these dependences. In one approach, Smith (2015) modeled the dependence structure in multivariate time series using D-vines. He introduced products of pair copula's subsets and named them block functional, which capture the serial and the cross-sectional dependence. In another approach, Brechmann and Czado (2015) invented a new regular vine structure called COPAR, based on C-vines, where each tree has a unique node connected to the other. To this purpose, they used the R-vine matrices (Aas et al. (2009)) to store all the relevant information of their model i.e. the COPAR-structure, the copula families and their parameters.
In this thesis we focus on a new vine structure called M-vines created by Beare and Seo (2015). We argue that this regular vine is well suited to model the dependence structure in multivariate higher order Markov chains. Especially, we show that the stationarity can be easily imposed by requiring the equality of some copula functions, while the Markov property can be imposed by assigning some copula functions independence copulas in any M-vine model. For an arbitrary multivariate time series, we derive an algorithm for the R-vine matrices of M-vines model. Especially, our algorithm is very useful for the construction of non-Gaussian VAR(p) models. We propose an algorithm to estimate the parameters of arbitrary copula functions for M-vine model following the semi-parametric estimation method (SSP) proposed by Aas et al. (2009). We test our implementation regarding the quality of estimates with a Monte Carlo analysis. Finally, we apply our implementations on the log-returns of BMW, Daimler and VW. We use a VAR(1)-model as benchmark for our application.
«
An important task in the analysis of multivariate time series is to study its dependence structure, i.e. serial, cross-serial and cross sectional dependence. The copula functions introduced by Sklar (1959) are a nice tool to capture these dependences. In one approach, Smith (2015) modeled the dependence structure in multivariate time series using D-vines. He introduced products of pair copula's subsets and named them block functional, which capture the serial and the cross-sectional dependence....
»