Bayesian Analysis of Stochastic & Deterministic Processes in the Error Correction Model

In this paper we present a method of estimating Bayes factors for determining the number of stochastic trends (s) and for joint estimation of s and the deterministic processes present in a multivariate error correction model. The Bayes factor estimates for s can, in particular cases, be shown to be functions of the classical trace statistic. This approach takes advantage of the Laplace method of numerical integration and the nonordinal restrictions presented by Strachan (2000) and follows the classical approach of Anderson (1951) and Johansen (1988 and 1991). The technique performs comparably with the classical trace test for s and with information criteria in finite samples in both Monte Carlo experiments and in applications to real data.