Efficient two-sided markets with limited information

Many important practical markets inherently involve the interaction of strategic buyers with strategic sellers. A fundamental impossibility result for such two-sided markets due to Myerson and Satterthwaite [33] establishes that even in the simplest such market, that of bilateral trade, it is impossible to design a mechanism that is individually rational, truthful, (weakly) budget balanced, and efficient. Even worse, it is known that the “second best” mechanism—the mechanism that maximizes social welfare subject to the other constraints—has to be carefully tailored to the Bayesian priors and is extremely complex. In light of this impossibility result it is very natural to seek “simple” mechanisms that are approximately optimal, and indeed a very active line of recent work has established a broad spectrum of constant-factor approximation guarantees, which apply to settings well beyond those for which (implicit) characterizations of the optimal (second best) mechanism are known. In this work, we go one step further and show that for many fundamental two-sided markets— e.g., bilateral trade, double auctions, and combinatorial double auctions—it is possible to design near-optimal mechanisms with provable, constant-factor approximation guarantees with just a single sample from the priors! In fact, most of our results in addition to requiring less information also improve upon the best known approximation guarantees for the respective setting.