EmpiricalPriceModelingfor.PDFVIP

Empirical Price Modeling for Sponsored Search Kuzman Ganchev University of Pennsylvania Philadelphia PA, USA kuzman@ Alex Kulesza University of Pennsylvania Philadelphia PA, USA kulesza@ Jinsong Tan University of Pennsylvania Philadelphia PA, USA jinsong@ Ryan Gabbard University of Pennsylvania Philadelphia PA, USA gabbard@ Qian Liu University of Pennsylvania Philadelphia PA, USA qianliu@ Michael Kearns University of Pennsylvania Philadelphia PA, USA mkearns@ ABSTRACT We present a characterization of empirical price data from sponsored search auctions. We show that simple models drawing bid values independently from a fixed distribution can be tuned to match empirical data on average, but still fail to account for deviations observed in individual auctions. Hypothesizing that these deviations are due to strategic bid- ding, we define measures of “jamming” behavior and show that actual auctions exhibit significantly more jamming than predicted by such models. Correspondingly, removing the jamming bids from observed auction data yields a much closer fit. We demonstrate that this characterization is a revealing tool for analysis, using model parameter values and measures of jamming to summarize the effects of query modifers on a set of keyword auctions. 1. INTRODUCTION Much of the academic literature on sponsored search to date has been theoretical in nature [7, 4, 5, 6], characteriz- ing behavior or payoffs under strong assumptions that may fail in practice—especially when the markets in question are young, bidders are inexperienced, and relevant pieces of information (such as the exact rules of the game) are fre- quently kept secret. To our knowledge there has been no exploratory study of actual bidding data on a large scale to determine how real-world auctions can best be analyzed and understood. This paper provides a simple but needed first look at such questions. We utilize sponsored search data drawn from a wide array of Overture/Yahoo! auctions and examine how b