Searching for Optimal Prices in Two-Sided Markets

A new arXiv paper investigates the fundamental limits of dynamic pricing algorithms in two-sided markets like ride-sharing or e-commerce platforms. The research proves that in many-to-many markets, common Two-Price Mechanisms inevitably suffer linear regret (Ω(T)), meaning they perform poorly over time. To overcome this, the authors introduce Segmented-Price Mechanisms, achieving O(n² log log T + n³) regret for maximizing gains-from-trade. The work delineates sharp boundaries between learnable and unlearnable pricing regimes.