Approximation Schemes for Capacity-Constrained Assortment Optimization under the Nested Logit Model

Abstract

The main contribution of this paper resides in proposing a novel approximate dynamic programming approach for capacitated assortment optimization under the Nested Logit model in its utmost generality. Specifically, we show that the optimal revenue can be efficiently approached within any degree of accuracy through purely combinatorial techniques, synthesizing ideas related to continuous dynamic programming, state space discretization, and sensitivity analysis of modified revenue functions. These developments allow us to devise the first fully polynomial-time approximation scheme in this context, thus resolving fundamental open questions posed in previous papers.