Repeated Auctions in Incomplete Information Settings with Learning Bidders

最近，电子市场的数量有了巨大的增长，迎合了各种各样的参与者，如广告商、零售商、出版商、贸易商和自由职业者。这类市场的一个统一特征是，参与者不具备与收益相关的市场特征的完整信息，这些信息是他们必须在与市场的多次互动中学习的。本研究的目的是研究竞标者对被拍卖物品信息不完全的重复拍卖市场，并分析由此产生的动态竞价行为。这种拍卖出现在在线广告市场，在维持互联网上提供的广泛服务方面发挥着重要作用。从这项研究中获得的见解将用于评估和改进这些市场的设计和运作。PI致力于让代表性不足的群体的学生参与这项研究项目，并将通过康奈尔大学的各种外展项目开展外展工作。重复拍卖市场固有的学习激励要求每个竞标者评估他或她的行为以及他或她的竞争对手的行为对市场状态未来演变的影响。为了了解竞标者在这种情况下的行为，本研究将首先使用动态规划工具来了解竞标者在给定其竞争对手的固定策略下的最优策略结构。利用这些结构性见解，研究人员将描述市场中每个竞标者对其他竞标者做出最优反应的最终均衡。我们将开发基于迭代方案的算法来近似计算这种均衡，并数值分析市场设计如何影响其均衡特性。我们将使用数值调查来确定改进重复拍卖市场的设计和运作的实用指导方针，以最大化其收入和整体效率。

Recently, there has been a tremendous growth in the number of electronics markets that cater to a diverse set of participants such as advertisers, retailers, publishers, traders, and freelance workers. A unifying feature of such markets is that the participants do not possess complete information about payoff-relevant market features, which they must learn during their multiple interactions with the market. The goal of this research activity is to study repeated auction markets where the bidders have incomplete information about the items being auctioned, and analyze the resulting dynamic bidding behavior. Such auctions arise in online advertising markets and play a significant role in sustaining a wide range of services available on the Internet. The insights obtained from this research will be used to evaluate and improve the design and operation of such markets. The PI is committed to involving students from underrepresented groups into this research program, and will engage in outreach efforts through various outreach programs at Cornell University.The learning incentives inherent in repeated auction markets require each bidder to evaluate the effects of his or her actions and those of his or her competitors on the future evolution of the market state. In order to understand how a bidder behaves in such settings, this research will first use the tools of dynamic programming to obtain insights about the structure of the bidder's optimal strategy for given fixed strategies of her competitors. Using these structural insights, the researcher will then characterize the resulting equilibrium in the market where each bidder reacts optimally to others. We will develop algorithms based on iterative schemes to approximately compute such equilibria and numerically analyze how the design of the market affects its equilibrium properties. We will use the numerical investigations to identify practical guidelines for improving the design and operation of repeated auction markets, in terms of maximizing its revenue and overall efficiency.