Distributionally Robust Optimal Auction Design under Mean Constraints

TL;DR

This paper identifies a robust auction mechanism that guarantees optimal revenue under mean constraints and worst-case distributions, revealing that a symmetric second-price auction with a random reserve is optimal.

Contribution

It introduces a distributionally robust auction design framework under mean constraints, showing the optimality of a symmetric second-price auction with a random reserve price.

Findings

Optimal auction guarantees under worst-case distribution are achieved by a symmetric second-price auction with a random reserve.

The optimal reserve price converges to a non-binding level as the number of bidders grows large.

Revenue guarantees improve at a rate of O(1/n) as bidders increase.

Abstract

We study a seller who sells a single good to multiple bidders with uncertainty over the joint distribution of bidders' valuations, as well as bidders' higher-order beliefs about their opponents. The seller only knows the (possibly asymmetric) means of the marginal distributions of each bidder's valuation and the range. An adversarial nature chooses the worst-case distribution within this ambiguity set along with the worst-case information structure. We find that a second-price auction with a symmetric, random reserve price obtains the optimal revenue guarantee within a broad class of mechanisms we refer to as competitive mechanisms, which include standard auction formats, including the first-price auction, with or without reserve prices. The optimal mechanism possesses two notable characteristics. First, the mechanism treats all bidders identically even in the presence of ex-ante asymmetries. Second, when bidders are identical and the number of bidders $n$ grows large, the seller's optimal reserve price converges in probability to a non-binding reserve price and the revenue guarantee converges to the best possible revenue guarantee at rate $O(1/n)$.