Bayesian Testing in Cointegration Models using the Jeffreys' Prior

We develop a Bayesian cointegration test statistic that can be used under a Jeffreys' prior. The test statistic is equal to the posterior expectation of the classical score statistic. Under the assumption of a full rank value of the long run multiplier the test statistic is a random variable with a chi-squared distribution. We evaluate whether the value of the test statistic under the restriction of cointegration is a plausible realization from its distribution under the encompassing, full rank model. We provide the posterior simulator that is needed to compute the test statistic. The simulator utilizes the invariance properties of the Jeffreys' prior such that the parameter drawings from a suitably rescaled model can be used. The test statistic can straightforwardly be extended to a more general model setting. For example, we show that structural breaks in the constant or trend and general mixtures of normal disturbances can be modelled, because conditional on some latent parameters all derivations still hold. We apply the Bayesian cointegration statistic to the Danish dataset of Johansen and Juselius (1990) and to four artificial examples to illustrate the use of the statistic as a diagnostic tool.