A Gibbs sampling approach to cointegration (Q1297860)

The authors consider the cointegrated VAR system \[ \Delta x_t=\Psi D_t+\sum_{i=1}^p \Pi_i\Delta x_{t-i}+\Pi x_{i-1} +\varepsilon_t, \] where \(x_t\) is a vector-valued time series, stationary after first differencing, \(\Delta x_t\) are the first differences, \(\varepsilon_t\) are vectors of deterministic components, and \(\varepsilon_t\) are \(N(0,\Omega)\) (i.i.d.). The matrices \(\Psi\), \(\Pi_i\) and \(\Omega\) are assumed to be unrestricted. Some linear restrictions exist for \(\Pi\). A Gibbs sampling algorithm is proposed for the Bayesian estimation of the parameters of this model. The authors describe the general idea of Gibbs sampling and the conditional posterior densities of the estimated parameters (under noninformative priors). This model is applied to the analysis of Belgian (1970-1990) economic data.