Learning Prices for Repeated Auctions with Strategic Buyers
Venue
Neural Information Processing Systems (2013)
Publication Year
2013
Authors
Kareem Amin, Afshin Rostamizadeh, Umar Syed
BibTeX
Abstract
Inspired by real-time ad exchanges for online display advertising, we consider the
problem of inferring a buyer’s value for a good when the buyer is repeatedly
interacting with the seller through a posted-price mechanism. We model the buyer as
a strategic agent, interested in maximizing her long-term surplus, and are
interested in optimizing seller revenue. We show conditions under which the seller
cannot hope to gain an advantage by learning the buyer’s value – i.e. the buyer can
always manipulate the exchange to hide her value. This result is accompanied by a
seller algorithm that is able to achieve no-regret when the buyer is unable to
incur the short-term costs of such manipulation.
