Track: Inventory Management

Abstract

We study a set of inventory control problems with correlated demands over different time periods. On the other hands, we relax the assumption of fully observation of the demand at the end of each time period. In other words, we consider the case of partially observed (censored) demand in the context of a multi-period inventory problem. If the demand in a period is larger than the inventory level, we don’t observe the unmet demand. Otherwise, the demand is fully observed and the leftover inventory is carried over to the next period. Formulating the problem as a Partially Observable Markov Decision Process provides a dynamic program (DP) to minimize the total expected cost. Unfortunately, the corresponding DP is defined on an uncountable state space, with little hope for a computationally feasible solution. We present an interesting heuristic policy with a percentile threshold structure which outperforms the myopic policy and performs close to the optimal policy. We derive its performance guarantee and evaluate it using numerical simulations.