Core Concepts
The seller can partially reveal information about bidders' click-through rates to maximize revenue in click-through auctions, while maintaining calibration constraints.
Abstract
The paper studies the problem of optimal information design in click-through auctions, where the seller has private information about bidders' click-through rates (CTRs) and can partially reveal this information to maximize revenue. This is a Bayesian variant of the "calibrated click-through auctions" studied by Bergemann et al. [11].
The key insights are:
Information design in click-through auctions is different from previous studies, as the revealed information affects the auction's allocation and payment rule, but not the bidders' bidding behaviors.
For the general case with a constant number of bidders, the authors provide an FPTAS to compute an approximately optimal signaling scheme, leveraging the Lipschitz continuity of the revenue function.
For the symmetric two-bidder case, the authors characterize the optimal signal ratio and construct a simple, prior-free signaling scheme that achieves a 0.995 approximation ratio, as long as the bidders' value density functions do not fluctuate much.
The technical contributions include novel discretization techniques to handle the calibration constraints, and a connection between the optimal signaling scheme under unknown value distributions and the uniform distribution.