On the Linear Programming Model for Dynamic Stochastic Matching and Its Application on Pricing

Abstract: Important pricing problems in centralized matching markets -- such as carpooling and food delivery platforms -- often exhibit a bi-level structure. At the upper level, the platform sets prices for different types of agents (e.g., riders with various origins and destinations, or delivery orders from diverse restaurants to customer locations). The lower level then matches converted agents to minimize operational costs, such as pooling riders with similar itineraries into the same vehicle or consolidating multiple delivery orders into a single trip. Motivated by these applications, we study the optimal value (cost) function of a linear programming model with respect to agent arrival rates, originally proposed by Aouad and Saritac (2022) for cost-minimizing dynamic stochastic matching under limited time. In particular, we investigate the concavity of this cost function. We show that weak concavity holds when there are fewer than or equal to three agent types with identical patience level. For an arbitrary number of agent types, (weak) concavity is more likely to hold when agents have high or similar patience, or when agents are more heterogeneous in their matching efficiencies. Building on these theoretical insights, we develop a Minorize-Maximization (MM) algorithm for the pricing problem that requires little stepsize tuning, and demonstrate substantial performance improvements over projected gradient-based methods on a large-scale, real-world ridesharing dataset. This makes it a compelling algorithmic choice for solving such pricing problems in practice.