Efficient estimate of Bayes factors from Reversible Jump output

We exend Meng and Wong (1996) identity from a fixed to a varying dimentional setting. The identity is a very powerful tool to estimate ratios of normalizing constants and thus can be used to evaluate Bayes factors. The extention is driven by the reversibler jump algorithm so that the output from the semplar can be directly used to efficiently estimate the required Bayes factor. Two applications, involving linear and logistic regression models, illustrate the advantages of the suggested approach with respect to alternatives previously proposed in the literature.