In this paper we propose a Bayesian variable selection method in quantile regression based on the Savage–Dickey density ratio of Dickey (1976). The Bayes factor of a model containing a subset of variables against an encompassing model is given as the ratio of the marginal posterior and the marginal prior density of the corresponding subset of regression coefficients under the encompassing model. Posterior samples are generated from the encompassing model via a Gibbs sampling algorithm and the Bayes factors of all candidate models are computed simultaneously using one set of posterior samples from the encompassing model. The performance of the proposed method is investigated via simulation examples and real data sets.
- Asymmetric Laplace distribution
- Bayes factor
- Bayesian model selection
- Markov chain Monte Carlo