OF THE UNIVERSITY OF MINNESOTA BY

I am deeply indebted to my advisor, Patrick Bajari, for his advice, guidance, and encour-agement during all stages of the completion of this dissertation. I am greatly inspired by his keen insights on interesting economic topics, and his complete devotion to the profession of economics. I am very grateful to Kyooil Kim, Minjung Park, Amil Petrin and Robert Town, for their advice, continuous encouragement and support. I also benefited from the participants of the Applied Micro Workshop at the University of Minnesota, and many comments from outside conferences and seminars. i Dedication To my parents, Quansheng Yuan and Jingpu Li, who have supported me all the way to realize my dream. To my wife Ying Li, whose love and sacrifices have sustained my entire graduate school career. ii This dissertation is a collection of two essays that deal with sponsored search auctions. Chapter 1 investigates and evaluates the performance of different types of sponsored search auction mechanisms. For the two popular sponsored search auctions—the General-ized First Price (GFP) auction and the Generalized Second Price (GSP) auction—current consensus in both the industry and academia is that the GSP auction is more stable and more efficient than the GFP auction. Specifically, in the GSP auction, bidders are less likely to “game the system”, meaning that an individual bidder will change his bid less frequently; his bid range will be smaller; and a bidder with a higher value will be more likely to win a higher and better slot. This paper examines this prevailing belief using a Regression Discontinuity Design (RDD) approach and finds that after bidders switch to the GSP auction, they actually bid 36 % more frequently and increase their daily bid range by $1.31. To compare efficiency differences, this paper constructs an efficiency index and shows that the GSP auction mechanism is at least 4 % more efficient. Chapter 2 examines how different automated bidding strategies impact advertisers ’ bid-ding performances. It backtests and simulates the following simple strategies: targetin