1: \begin{abstract}

2:   We study the convergence properties of a collapsed Gibbs sampler for Bayesian

3:   vector autoregressions with predictors, or exogenous variables. The Markov

4:   chain generated by our algorithm is shown to be geometrically ergodic

5:   regardless of whether the number of observations in the underlying vector

6:   autoregression is small or large in comparison to the order and dimension of

7:   it. In a convergence complexity analysis, we also give conditions for when

8:   the geometric ergodicity is asymptotically stable as the number of

9:   observations tends to infinity. Specifically, the geometric convergence rate

10:   is shown to be bounded away from unity asymptotically, either almost surely

11:   or with probability tending to one, depending on what is assumed about the

12:   data generating process. This result is one of the first of its kind for

13:   practically relevant Markov chain Monte Carlo algorithms. Our

14:   convergence results hold under close to arbitrary model misspecification.

15: \end{abstract}

16: