Coordinated replenishment game and learning under time dependency and uncertainty of the parameters (Q6159521)

This paper presents a multi-item two-layer inventory model under periodic reviews to determine the can-order threshold of a \((S, c, s)\) policy in a coordinated replenishment problem. Each item is managed by an independent decision maker who is assumed to lack information about the demand of any other item. Then, this coordinated replenishment problem can be formulated as a Bayesian-like game, where each item is considered as a player. The players are, contrary the usual assumption, do not have common knowledge of a prior distribution over ``types'' of the items. Thus, players apply distributionally robust optimization procedures to obtain the optimal can-order level that minimizes expected costs. The optimal can-order level obtained corresponds to a Bayes-Nash equilibrium.