AUTHOR: P.J. Healy
In many environments prediction markets have been used to aggregate information and accurately estimate the probability of future events. The environments where these markets are most successful are typically `simple' environments with few securities and many traders. We test the standard prediction market mechanism in a highly complex environment with several securities and few traders and compare its performance to three alternative mechanisms for aggregating information. In the complex environment the performance of the prediction market is dominated by a simple iterative polling mechanism. We analyze four behavioral conjectures that explain why the poll performs better in the complex setting.