A Bayesian analysis of the parameters in the vector autoregressive model with a single cointegrating relationship is presented. A mixture distribution is used as a prior for the cointegration vector to allow for a proper representation of prior beliefs. The traditional normalization coupled with a, intendedly uninformative, flat prior on the free elements in the cointegration vector are shown to add unwanted information into the analysis. Parametrization of the cointegration vector in polar coordinates is argued to be a better alternative and a suitable procedure is developed.
We illustrate the method on a simple example with two interest rates.

 Key words: Bayesian, Cointegration, Mixture prior, Polar coordinates. 

Close this Window

Last update: 1998-05-28 / KH