Abstract.

We study dynamic assortment decisions of a firm learning about consumer tastes by observing sales of the products offered. Each period, the firm offers an assortment to maximize expected total profits over a finite horizon given its subjective beliefs on consumer tastes. The consumers choose a product that maximizes their own utility and the firm updates its beliefs on consumer tastes after having observed the sales of each product in the assortment. We model consumer tastes as locations on a Hotelling line and develop a nonparametric Bayesian learning model using Polya tree priors that makes no assumptions on the form of the consumer taste distribution. Our nonparametric learning model results in optimal profits that are robust to misspecification in the consumer taste distribution. We develop upper bounds on the firm’s total profit based on information relaxations and study the performance of various heuristic policies with respect to these upper bounds.