Efficient algorithms for online ad markets with time constraints

算法交易是当今经济的重要组成部分。该提案涉及在无线移动环境中的广告拍卖环境中的算法交易的设计和实现。这项工作是与我们位于多伦多的工业合作伙伴Additive Mobility合作完成的，该公司通过在广告拍卖交易所进行交易，为移动无线系统背景下的广告商提供尖端的广告空间解决方案。拟议的研究旨在开发先进的算法，以帮助提高广告拍卖背景下的算法交易的收益并最大化利润。关键问题是竞价策略必须在100毫秒左右的非常快的时间尺度上运行。这项研究将由三个主要部分组成：1)在严格的时间限制下，广告空间竞价的模型和技术的发展；2)算法交易策略的发展；3)战略的实施。我们将在第二次和第三次推进中与我们的工业合作伙伴密切合作。这项研究将涉及许多组成部分：1)使用随机交换理论的思想建立随机交互模型；2)了解预算和时间约束在执行时间和活动时间范围方面的作用；3)了解信息在行动环境中的作用；4)开发价格发现技术和算法；5)开发实时竞价算法，可以在拍卖持续时间的限制内对信息历史进行操作。这些技术将涉及马尔科夫模型、平均场理论、随机决策理论、学习和算法的发展。

Algorithmic trading is an essential part of today's economy. This proposal in on the design and implementation of algorithmic trading in the context of ad auctions in the wireless mobile context. This work is being done in collaboration with our industrial partner Addictive Mobility, Toronto that provides cutting edge solutions for providing ad space to advertisers in the context of mobile wireless systems by trading on ad auction exchanges.The proposed research is to develop advanced algorithms that help improve the yield and maximize the profitability for algorithmic trading in the ad auction context. The key issues are bidding strategies that must operate at very fast time scales of around 100 milliseconds. The research will consist of three main components: 1) Development of models and techniques for ad space bidding given strict time constraints.2) Development of algorithmic trading strategies, and 3) Implementation of strategies. We will work closely with our industrial partner in the 2nd and 3rd thrusts.The research will involve many components: 1) Building models of random interactions using ideas from random exchange theory, 2) Understanding the role of budget and time constraints both in terms of execution time as well as campaign time horizon, 3) Understanding the role of information in the context of actions, 4) Develop techniques and algorithms for price discovery, and 5) Development of real-time bidding algorithms that can operate on the information history within the constraints of the auction durations.The techniques will involve Markov models, mean field theory, stochastic decision theory, learning, and the development of algorithms.