We consider a dynamic pricing problem for a two-sided consumer electronics trade-in program, where a firm acquires and re-sells multiple types of pre-owned (used) products over a finite selling horizon. There are customers trading in their used products for new products at discounted prices and customers buying refurbished products. The firm sets trade-in prices and resale prices to maximize its total expected profit. We first discuss the scenario that the firm knows the choice models of customers. Due to the high-dimensional state space, deriving the optimal policy using dynamic programming is computationally intractable. To circumvent this, we develop simple and provably effective heuristic policies based on the solution to a deterministic upper-bound problem. We design a dynamic policy called the Batched-Adjustment Control (BAC) policy, under which the selling horizon is divided into different consecutive and disjoint batches for different products and the prices in one batch are updated based on the realized uncertainties in the previous batch. The profit loss of BAC relative to the optimal one is in the order of Ō (T^(1/3)). When the firm does not know the choice model parameters of customers, it has to learn while making pricing decisions over time. We develop an algorithm called Parametric-Batched-Adjustment Control (PBAC), in which the firm first uses Maximum Likelihood Estimation to learn the trade-in and demand models’ parameters, and then adopt a similar pricing policy akin to BAC while using the estimated parameters. With carefully chosen algorithm parameters (e.g., length of exploration phase, batch size), we show that PBAC has a regret in the order of Ō (T^(1/2)).

This is based on joint work with Zhuoluo Zhang (Xiamen University), Murray Lei (Queen’s University), and Wenhao Li (SUFE).