Abstract:
|
We present a prior probability model for time-varying covariance matrices. Our approach utilizes the separation strategy presented by Barnard, McCulloch and Meng (2000) to separate covariances and correlations. Modeling the covariances is easy and it can be done simply by utilizing a link that ensures positiveness of the covariances. This, for instance, can be the log link. The novelty of our approach is in the prior for the correlation matrix that has to satisfy the more complex requirement of positive definiteness. We present the model, methods for posterior simulation and an application to a real dataset.
|