A1282
Title: Pricing in the dark: Are partial competitor data sufficient for optimal pricing?
Authors: Minhao Qi - Zhejiang University (China) [presenting]
Abstract: The aim is to investigate personalized pricing in competitive markets where competitors prices are only partially observed in offline data. Traditional approaches that ignore competitor pricing often yield suboptimal policies, while methods assuming full price observability are impractical due to privacy or data limitations. To address this gap, we develop a novel identification strategy that enables recovery of optimal personalized prices under partial observability. We analyze four realistic information scenarios: (i) no access to competitor prices, (ii) competitor prices observed only when the customer chooses a competitor, (iii) competitor prices observed only when the customer chooses the focal firm, and (iv) full price observability. We show that, under mild conditions, the optimal policy under partial observability can match that under full observability by exploiting variation in competitor pricing. This insight enables optimal policy learning even when the true optimal prices are not present in the training data. Based on this, we introduce a new learning framework, PRIME (PeRsonalized prIcing with partially observed coMpEtitors)tailored to each of the four information structures. We provide rate-optimal finite-sample regret bounds for all scenarios. Empirical evaluations using simulations and real-world insurance data demonstrate that PRIME achieves substantial improvements in pricing performance across various data scenarios.
