Robust Revenue from Correlated Private Values

We study how much revenue a seller can guarantee when bidders know their values but their additional information and equilibrium are unknown. For full-product-support finite priors and everywhere-positive continuous densities, our main result identifies this guarantee with the worst-information revenue potential, approached by finite auctions. With at least two bidders, full extraction is robust exactly when almost every positive-surplus state has a highest-valued bidder at its conditional upper endpoint. This criterion covers ties and arbitrary bounded Borel priors in the specified information class. Files: 45-page main paper and 25-page supplement, Binary Revenue and Withholding.