Collaborative Research: Model Accuracy and Learning in Revenue Management and Dynamic Pricing

这项研究将解决与决策者（买方和卖方）在收入管理和动态定价背景下使用的模型准确性相关的问题。该研究将开发数学框架，在其中将研究这种定价过程的长期动态行为。重点将放在发展对决策者应用结构不准确模型的积极和消极后果的理解，他们试图使用数据学习模型参数，并随着时间的推移做出一系列定价或购买决定。将分析决策者模型的准确性对一些自然绩效指标（如长期应计收入）的影响。这些研究将包括发展一种描述长期运行性能的理论，以及通过计算机模拟进行实验。如果成功，建议的工作将帮助收入管理专业人员了解使用不准确模型的后果。这是一个具有实际意义的问题，因为在收入管理应用程序中使用的人类行为模型通常是不准确的。可能从本研究中受益的具体环境包括：(1)可替代产品的竞争卖家试图学习价格与需求之间的关系，如移动电话服务的卖家；(2)多个购买者分别试图了解某一特定产品可供购买的概率，例如购买折扣机票的购买者想要决定是早买还是晚买；或者(3)一个单一的可替代产品的销售者，他试图了解买家如何在产品中进行选择，例如一个制造商想知道买家如何在不同的交货时间（交货时间价格的函数）中进行选择。这项研究的结果可能会提醒实践者考虑学习和决策之间相互作用的重要性，以及模型结构对这种相互作用的影响。反过来，这可能会导致模型的发展，这些模型对偏离它们的假设是稳健的，并最终在这些行业中产生更有效的定价方法。

This research will address issues related to the accuracy of models used by decision makers (buyers and sellers) in the context of revenue management and dynamic pricing. The research will develop mathematical frameworks within which the long run dynamic behavior of such pricing processes will be studied. Emphasis will be given to developing an understanding of positive and negative consequences of the application of structurally inaccurate models by decision makers, who attempt to learn about model parameters using data and who make a sequence of pricing or buying decisions over time. The impact of the accuracy of the decision makers' models on some natural performance measures such as long run accrued revenues will be analyzed. These studies will involve the development of a theory characterizing long run performance as well as experimentation via computer simulations.If successful, the proposed work will help revenue management professionals understand the consequences of using inaccurate models. This is an issue of practical interest, since models of human behavior used in revenue management applications are often inaccurate. Specific settings that may benefit from this research include those with: (1) competing sellers of substitutable products who attempt to learn the relationship between prices and demand, such as sellers of mobile phone services; (2) multiple buyers who individually attempt to learn the probability that a particular product will be available for purchase, such as buyers of discount airline tickets who want to decide whether to purchase a ticket early or wait until later; or (3) a single seller of substitutable products who attempts to learn how buyers choose among the products, such as a manufacturer who wants to know how buyers will choose among various lead times as a function of lead times' prices. Results of this research may alert practitioners of the importance of considering interactions between learning and decision-making, and the effect of a model's structure on this interaction. This, in turn, may lead to the development of models that are robust to departures from their assumptions, and ultimately to more effective methods of pricing in these industries.