We consider optimal reserve prices in BrightRoll Video Ex-change when the inventory opportunity comes from other exchanges (downstream marketplaces). We show that the existence of downstream auctions impacts the optimal floor. Moreover, it renders the classical derivation of the floor set by a monopolist inadequate and suboptimal. We derive the new downstream-corrected reserve price and compare its performance with respect to existing floors and the classical optimal monopoly price. In our application, the downstream-corrected reserve price proves superior to both. The proposed model also deals with data challenges commonly faced by exchanges: limited number of logged bids in an auction, and uncertainty regarding the bidding behavior in other exchanges.

The relevance of this study transcends its particular context and is applicable to a wide range of scenarios where sequential auctions exist, and where marketplaces interact with each other.

Filed under: Deep Learning