Optimal policy for the evaluation of partially observable objects with normal distributed rewards

Publication: Working/Discussion PaperWorking Paper/Preprint


This paper presents an optimal stopping policy for searching and/or testing. Possible actions are considered to be independent selections from a large population.The joint distribution in this population of the actual value, the preliminary information and the additional information gained by (additional) testing - is assumed to be 3-dimensional normally distributed and known. The person searches for an action at cost c1 and gets some preliminary information of the value of this action. He can take the action, or get some additional information about its value by testing at the cost c2. If he decides not to test and/or stop he can look for another action again with cost c1. The problem is to maximize the expected value of the value of the action minus the expected total cost of searching and testing. The solution is applicable in many applications of optimal stopping methodology like consumer search, information retrieval, web- search and job search. The solution is also applicable to the case where many possibilities are selected by 2-stage testing subject to budget constraint on search and testing for a reasonably large budget.
Original languageEnglish
Publication statusPublished - 1 Jul 2011

Cite this