2012/12/05_Elisabeth Waldmann (University of Göttingen)

(Bayesian) Quantile Regression

2012/12/05 – 12:00 h | Elisabeth Waldmann, University of Göttingen

Abstract

Quantile regression describes the impact of covariates on the conditional quantiles instead of the mean and thus provides more detailled information about the conditional distribution without forcing the statistician to assume a specific error distribution. The estimation of the parameters can be performed by minimizing weighted absolute deviances, which is not easy to implement when using more complicated effects than in simple linear regression. One way to avoid the difficulties of linear programming is to use Bayesian inference. The latter can be conducted via Markov Chain Monte Carlo (MCMC) techniques or by using Variational Approximations (VA). The concept of quantile regression, how the estimations work, and their advantages and problems will be explained based on simulations and a data example stemming from English farm efficiency analysis.