Bayesian Analysis of Integration at Different Frequencies in Quarterly Data

The paper provides a unifying framework for conducting Bayesian inference on the presence of seasonal and zero frequency unit roots in quarterly data. The main technique used is the analysis of posterior odds ratios. A new parameterization is provided for the model and the prior distributions implemented are discussed and justified. The analysis relies heavily on the application of a Gibbs sampling algorithm. Such techniques render the Bayesian approach more flexible and implementable, giving the applied researcher the possibility of specifying a vast range of prior distributions. The methods are applied to a set of UK quarterly series. Compared to previous studies, less evidence is found to support seasonal integration hypotheses.