Abstrakt
The aim of the contribution is to introduce an innovative approach to conditional covariance and correlation modelling. This can be obviously useful in multivariate financial time series analysis, e.g. in the multivariate GARCH context.
The proposed method consists of two steps. The first one is based on the LDL factorization of the conditional covariance matrix, state space modelling and associated Kalman recursions.
Moreover, it is able to deliver a dynamic orthogonal transformation of given stochastic vector data. The second step of the suggested technique analyses conditional covariances of transformed time series which is indeed simplified due to its simultaneously uncorrelated components.
In the paper, performance of the introduced procedure is tested in an empirical financial framework. Namely, the daily correlation links between logarithmic returns on stocks and bonds are investigated and compared with other estimated dynamic correlations gained by several common methods, e.g. the moving averages, the diagonal BEKK model or the dynamic conditional correlation (DCC) models.