Mechanism Design for Correlated Valuations: Efficient Methods for Revenue Maximization
Michael Albert, Vincent Conitzer, Giuseppe Lopomo, Peter Stone
Operations Research · 2022 · Volume 70 · Issue 1
Abstract
Traditionally, the mechanism design literature has been primarily focused on settings where the bidders’ valuations are independent. However, in settings where valuations are correlated , much stronger results are possible. For example, the entire surplus of efficient allocations can be extracted as revenue. These stronger results are true, in theory, under generic conditions on parameter values. However, in practice, they are rarely, if ever, implementable because of the stringent requirement that the mechanism designer knows the distribution of the bidders types exactly. In this work, we provide a computationally efficient and sample efficient method for designing mechanisms that can robustly handle imprecise estimates of the distribution over bidder valuations. This method guarantees that the selected mechanism will perform at least as well as any ex post mechanism with high probability. The mechanism also performs nearly optimally with sufficient information and correlation. Furthermore, we show that when the distribution is not known and must be estimated from samples from the true distribution, a sufficiently high degree of correlation is essential to implement optimal mechanisms. Finally, we demonstrate through simulations that this new mechanism design paradigm generates mechanisms that perform significantly better than traditional mechanism design techniques given sufficient samples.
Keywords
Machine Learning and Data Science, mechanism design, robust optimization, revenue maximization, correlated valuations
Enable JavaScript for citation tools and additional paper interactions.