Hybrid Dual Censored Joint Learning of Reserve Prices and Bids for Upstream Auctioneers
Piyush Paliwal and Lampros Stavrogiannis
Open Marketplace is an environment where buyers and sellers come together to openly exchange ad inventory. A discerning characteristic of this marketplace is that inventory is traded in a sequence of auctions, known as the supply path. We consider the profit maximization problem from the standpoint of an upstream marketplace (exchange) for opportunities coming from downstream marketplaces. To make a profit, the exchange extends the reserve price originated from the downstream marketplace by a reserve multiplier in its auction for its demand partners and reduces the clearing price of that auction by shading its bid when bidding downstream. We introduce, for the first time, a framework for jointly learning the reserve multiplier and bid shading factor for first-price auctions that generalizes to second-price auctions too, taking into account the censoring of prices on both sides of the market. We also provide an elegant strategy based on the Revenue Equivalence Theorem to deal with the co-existence of both auction types for the same inventory. A/B tests in LoopMe exchange demonstrate the effectiveness of our framework in practice.