Bayesian Tests for Co-Integration in the Case of Structural Breaks : An Application to the Analysis of Wage Moderation in France

This paper considers a special non-linear time series problem, that of testing for co-integration in a Bayesian framework when there is a break in the co-integrating relationship. It is shown that a partial linearization of the likelihood function solves many puzzling questions, in particular identification and common factor restrictions which are originally imbedded in the model. A generalization of the Jeffreys' prior is derived for the dynamic parameter which monitors co-integration. The procedure is applied to a one time much debated question in France which concerns the wage regulation policy implemented at the beginning of the eighties.