A reexamination of the equity-premium puzzle: A robust non-parametric approach