Model-Free Assortment Pricing with Transaction Data

We study a problem in which firm sets prices for products based on the transaction data, i.e., product past customers chose from an assortment and what were historical that they observed. Our approach does not impose model distribution of customers' valuations only assumes, instead, purchase choices satisfy incentive-compatible constraints. The individual valuation each customer can then be encoded as polyhedral set, our maximizes worst-case revenue assuming new are drawn empirical implied by collection such polyhedra. show optimal this setting approximated at any arbitrary precision solving compact mixed-integer linear program. Moreover, we special practical cases where program solved efficiently, design three approximation strategies low computational complexity interpretable. Comprehensive numerical studies synthetic real data suggest pricing is uniquely beneficial when has limited size or susceptible to misspecification.