New mechanism for repeated posted price auction with a strategic buyer without discounting

Abstract

On ad exchange platforms the place for advertisement is sold through different kinds of auctions. However, it is not uncommon the situation where the seller repeatedly encounters only one buyer, thus the posted price auction degenerates into a monopoly-monopsony game with asymmetric information and nearly an infinite number of rounds; on each round the seller proposes the price and the buyer accepts or rejects it. I learned this problem from a discussion with members of Yandex research team and my main motivation was to find an incentive-compatible seller's strategy. In this short paper such a strategy is proposed and a corresponding distortion at the top type lower bound (Spence-Mirrlees property, actually) for the surplus of the buyer is established; this shows that the proposed strategy is the best possible. The key ingredients are the following. The main leash that the buyer has is the frequency of accepted deals. Once this frequency (as a function on the buyer's type) is fixed, the strategy randomly chooses between the {\it rewarding} price which incentivises the buyer to reveal his type (the higher the type, the more average surplus the buyer has), the {\it adaptation} price which allows the buyer to communicate that his type is higher then the current guess of the cook, and the {\it type confirmation} price which disincentivises the buyer to pretend that his type is higher than it is.