We propose estimators for the parameters of a linear median regression without any assumption on the shape of the error distribution including no condition on the existence of moments allowing for heterogeneity (or heteroskedasticity) of unknown form, noncontinuous distributions, and very...