Single-Visit Policies for Allocating a Single Resource in a Stochastic Environment

Conventional analyses of stochastic resource allocation problems based on Gittins' indices frequently yield policies which involve an unacceptable amount of switching of the resource from one option to another. This paper discusses a variety of methodologies aimed at solving this problem. They principally involve the development and analysis of stochastic resource allocation models incorporating switching costs together with a consideration of a new class of single-visit policies for which each option has a single (random) period during which it is in receipt of the key resource.