ICES: Large: Collaborative Research: Towards Realistic Mechanisms: statistics, inference, and approximation in simple Bayes-Nash implementation

点击翻译按钮获取中文摘要

Mechanism design lays the economic foundations for the design and analysis of economic institutions, social and computer protocols, service provisioning, and other applications where participants may act selfishly in their own best interest. A common paradigm for real-world mechanism design is trial and error: a mechanism is proposed, it is executed, then changes are made to it based on its performance. In order to make these changes an econometric analysis must be undertaken, i.e., the participants' actions in the mechanism (assumed to be in equilibrium) must be reverse engineered to obtain the participants' preferences. Using these inferred preferences, potential changes to the mechanism can be evaluated and ranked. While mechanism design theory for the most part relies on knowledge of the market, real world mechanisms tend to do some market analysis on the fly. The PIs research will combine econometric inference with mechanism design theory to investigate the econometric properties of mechanisms and design mechanisms that are simultaneously good at market analysis and exploiting that information to attain an objective specified by the mechanism designer.This research program will introduce econometric techniques to computer science and will bring together topics from computer science and economics that have yet to be studied together. For example, these issues are very important in practice especially in the rapidly growing areas of sponsored search and targeted display advertising where auction mechanisms have been deployed for pricing and placement of advertisements and a major challenge is in adjusting the mechanisms in response to past data.