Bayesian quantile inference

The paper proposes a Bayesian interpretation of quantile regression that is shown to be equivalent to scale mixtures of normals leading to a skewed Laplace distribution. This representation of the model facilitates Bayesian analysis by means of Gibbs sampling with data augmentation, and nests regression in the L-1 norm as a special case. The new methods are applied to an analysis of the patents - R&D relationship for U.S. firms and unit root inference for the dollar-deutschemark exchange rate.