A Bayesian quantile-regression preprint reports stronger resistance to extreme observations, faster computation and 19 wins in 20 prediction comparisons.