An agent-based prediction market: a case study of xFuture in Taiwan

Abstract

This study adopted an agent-based modelling approach to investigate results of the world's leading Chinese prediction market, xFuture; in particular, Taiwan's 2008 and 2012 presidential elections. Real transaction results of xFuture were used as the base model. We employed double auction mechanism and Schelling's segregation model, and attempted to reconstruct the networking structures of the prediction market in 2008 and 2012. Purpose of this study is to discuss whether networking structures have any deterministic influence on how individual and hence joint belief distribution of participants could be formed. It is found that certain belief distribution properties, including shape, spread and location, could be critical factors for market efficiency/inefficiency. Our analysis suggested that in 2012 xFuture participants' belief distributions for the main candidates were associated with high kurtosis, small variance and approximate means. In other words, it might have a higher probability to deviate from the true outcome.