Dynamic modelling of the dependence in multivariate time series
(2010)
Files
REZNIKOVA.pdf
Restricted Access - Adobe PDF
- 5.05 MB
Details
- Authors
- Supervisors
- Hafner, Christian
- Abstract
- (en) Research projects in the area of multivariate financial time-series are of a particular interest for financial institutions, as for all of them the appropriate estimation of their risks is an essential requirement. In this thesis we look at modelling the dependence between financial series from two perspectives: time-varying copula and time-varying correlation. A motivation for this is that in extreme market situations and crashes, correlations typically increase. It seems plausible to assume that in such situations measures of nonlinear dependence also change. In the first part of the thesis we review the existing time-varying copula models that combine simple and parsimonious univariate volatility models with a flexible copula model for the error term. We also introduce a new semiparametric dynamic copula model, where the dependence parameter is a smooth function of time and is estimated in a nonparametric way. In the second part of the thesis we discuss the estimation methods for the popular dynamic conditional correlation model applied to a large number of time series. We analyze the problems of the existing estimation methods and suggest techniques to improve the estimators. Finally, we study the forecasting performance of different time-varying correlation models. We focus on the correlation forecasts only and provide an empirical example for the stock market indices of the G5 countries.
- Affiliations
Citations
Reznikova, O. (2010). Dynamic modelling of the dependence in multivariate time series. https://hdl.handle.net/2078.5/72900