Advances in the specification and the estimation of multivariate GARCH models
(2004)
Files
Rombouts_J.pdf
Restricted Access - Adobe PDF
- 3.52 MB
Details
- Authors
- Supervisors
- Bauwens, Luc
- Abstract
- This thesis puts forward several contributions in the specification and the estimation of multivariate GARCH models. The second chapter provids an up do date survey on multivariate GARCH models. The third chapter proposes an alternative to the common choice of the innovation density that underlies the multivariate GARCH model. A semiparametric model is defined where the variance matrix still has a specific parameterized structure but the data themselves determine the innovation density. Chapter four demonstrates the empirical relevance of the choice of the innovation density underlying the multivariate GARCH model. This choice has consequences for risk measures like Value-at-Risk and it seems that the Value-at-Risk is rather insensitive to the choice of the multivariate GARCH model. Chapter five investigates the estimation of temporally aggregated multivariate GARCH models. It studies whether the quasi maximum likelihood estimator can still be used as an approximately valid procedure and its performance is compared to a nonlinear least squares estimator. It turns out that the first estimator still performs reasonably well for small levels of aggregation. Chapter six develops a Bayesian approach that enables inference on a large number of time series. To achieve parsimony, the procedure classifies the series in a small number of clusters. Within each cluster, the series share the same GARCH model and the same parameter. We make it possible to calculate the marginal likelihood of the model so that Bayesian model comparison techniques can be used to infer the number of components in the model.
- Affiliations
Citations
Rombouts, J. V. K. (2004). Advances in the specification and the estimation of multivariate GARCH models. https://hdl.handle.net/2078.5/73057