過去預測市場的交易機制大多都是利用雙方喊價市場進行交易,本文採用 Hanson (2003) 提出的預測方法—市場計分法來進行預測,實驗背景為2012年與2008年的臺灣總統大選,並參考真實世界的預測市場—臺灣的「未來事件交易所」中,各縣市對於總統候選人得票率預測合約的資訊,再加入族群的群聚效果以及反執政黨效果作為參數,最後設計實驗在不同的群聚效果之下,比較模擬市場計分法的預測結果,與預測市場中利用雙方喊價市場機制的預測結果之差異。 研究結果發現,在全體民眾共識一致的社會狀態之下,市場計分法再加入高的群聚效果,會最接近雙方喊價市場機制的預測結果。反之在全體民眾共識較不一致的狀態下,市場計分法加入低的群聚效果,即能使預測結果接近雙方喊價市場機制的預測結果。在2012年與2008年實驗之下,研究還發現高群聚效果主要分布在偏藍的縣市,低群聚效果主要分布在偏綠的縣市,此與當時的社會背景以及不同政黨的選民結構有關。 Instead of using double auction market mechanism, we apply market scoring rule (MSR) to overcome potential liquidity problem (Hanson, 2003). Year 2008 and 2012 Taiwan Presidential Election results have been adopted in agent-based model (ABM). We use ABM to explore the possible belief distributions behind the prediction market in Taiwan, known as xFuture. We assume that the initial belief distribution come from the results of actual vote shares. Then this initial belief distribution will evolve through social networking, controlled by degree of segregation and information radius. In addition, we also add anti-ruling party effect to approximate excess demand of opposition party as evidenced in the trading volume of prediction market. Our goal is to find the best fit setting for the prediction market. We find that under high social consensus, MSR with a higher degree of segregation setting tends to fit xFuture better. In addition, we find that higher degree of segregation settings fit most KMT ruling cities, while lower degree settings tend to fit DPP ruling cities. The results may reflect different society opinion between these two periods.