An adaptive algorithm for the optimal sample size in the non-stationary datadriven newsvendor problem

Abstract

We investigate the impact of the sample size in the non-stationary newsvendor problem when the underlying demand distribution is not known, and performance is measured by the decision-maker's average regret. The approach we propose is entirely data-driven, in the sense that we do not estimate the probability distribution of the demand and instead rely exclusively on historical data. We propose an iterative algorithm to determine the number of past observations that should be included in the decision-making process, provide insights into the optimal sample size and perform extensive computational experiments. © 2007 by Springer Science+Business Media, LLC.