Use this URL to cite or link to this record in EThOS:
Title: On assortment optimization under active learning
Author: Schurr, Jochen
Awarding Body: Lancaster University
Current Institution: Lancaster University
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Assortment optimization, also called assortment planning, is the decision process of a retailer of choosing a limited number of products that are to be presented to customers in a show room or an equivalent environment. Assortment optimization has been a subject of active interest in an academic setting since the 1970s and with the emergence of electronic data processing it has become a major driver of success for retail companies. In this thesis, we consider a rather new subfield there of: dynamic assortment planning. Retailers with the ability to adapt quickly to demand observations can beat the market by adapting the assortment towards the products that turn out to sell best. This creates larger revenue and at the same time reduces the costs caused by idle inventory. Having only vague knowledge of the actual demand, the task becomes bifold and an exploration exploitation type trade off between learning about the demand and utilizing this knowledge towards profit maximization has to be faced. We develop various models and appropriate, close- to-optimal, heuristic decision policies in an apparel retailing context. In a first setting, we derive heuristic policies basing on Gittins indices for multi-armed bandit models and develop heuristic methods to apply them in a non- trivial knapsack type constraint situation. Extensive numerical testing demonstrates the outstanding performance strength of our policies and we are able to derive remarkably tight upper bounds to the non- tractable optimal solution. We then extend this model to account for substitution effects, which inflict a tremendous increase in complexity on the problem. ·With the use of stronger simplifications than before, we are still able to develop heuristic policies with active learning. Numerical studies indicate an improvement towards a myopic policy in a similar order as in the previous setting. We close this study by suggesting an improvement on a well- known heuristic method.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available