Abstract/Summary

We investigate nine data series previously identified as containing bubbles using Bayesian Markov Switching models. Nearly all of the series appear to display strong regime switching that could possibly be induced by `bubble' processes, but in each case the type of model that best describes each price differs substantively. We pay particular attention to whether these series contain transient explosive roots, a feature which has been suggested to exist in several bubble formulations. Bayesian model averaging is employed which allows us to average across a range of submodels, so that our empirical findings are not based on only one well performing model. We show that explosive regimes may exist in many submodels, but only when the flexibility of the model is limited in other important respects. In particular, when Markov Switching models allow for switching levels of error variance, explosive root regimes occur in only a minority of the series.