Agent-Based Modeling of the Prediction Markets for Political Elections

Abstract

We propose a simple agent-based model of the political election prediction market which reflects the intrinsic feature of the prediction market as an information aggregation mechanism. Each agent has a vote, and all agents' votes determine the election result. Some of the agents participate in the prediction market. Agents form their beliefs by observing their neighbors' voting disposition, and trade with these beliefs by following some forms of the zero-intelligence strategy. In this model, the mean price of the market is used as a forecast of the election result. We study the effect of the radius of agents' neighborhood and the geographical distribution of information on the prediction accuracy. In addition, we also identify one of the mechanisms which can replicate the favorite-longshot bias, a stylized fact in the prediction market. This model can then provide a framework for further analysis on the prediction market when market participants have more sophisticated trading behavior.