Content area
Full text
Rev. Econ. Design (2013) 17:95128
DOI 10.1007/s10058-013-0144-z
ORIGINAL PAPER
Received: 29 December 2011 / Accepted: 16 April 2013 / Published online: 28 April 2013 Springer-Verlag Berlin Heidelberg 2013
Abstract The Gates Hillman prediction market (GHPM) was an internet prediction market designed to predict the opening day of the Gates and Hillman Centers, the new computer science complex at Carnegie Mellon University. Unlike a traditional continuous double auction format, the GHPM was mediated by an automated market maker, a central agent responsible for pricing transactions with traders over the possible opening days. The GHPMs event partition was, at the time, the largest ever elicited in any prediction market by an order of magnitude, and dealing with the markets size required new advances, including a novel span-based elicitation interface that simplied interactions with the market maker. We use the large set of identity-linked trades generated by the GHPM to examine issues of trader performance and market microstructure, including how the market both reacted to and anticipated ofcial news releases about the buildings opening day.
Keywords Prediction markets Automated market making Case studies
Market design
JEL Classication D4 D7
An earlier version of this paper was presented at and appeared in the proceedings of the 11th ACM Conference on Electronic Commerce (EC 2010).
A. Othman (B) T. Sandholm
Computer Science Department, Carnegie Mellon University, Pittsburgh, PA, USA e-mail: [email protected]
T. Sandholme-mail: [email protected]
The Gates Hillman prediction market
Abraham Othman Tuomas Sandholm
123
96 A. Othman, T. Sandholm
1 Introduction
Prediction markets are powerful tools for aggregating information (Berg et al. 2001; Wolfers and Zitzewitz 2004; Cowgill et al. 2009; Chen and Pennock 2010). A typical prediction market only generates a single point of interest; for instance, the probability that a certain candidate will win an election, or the percent of the vote that candidate will receive. Over more complex event spaces, however, these point estimates can be inappropriate. Consider a prediction market to estimate the US ination rate over the next 5years. Conceivably, market participants could be split between a very low estimate and a very high estimate. The resulting consensus of a middle value could be an accurate estimate of the expectation, but would be misleading to design policy around.
Recent theoretical work has indicated that...