我們以零智慧代理人,建構了一個簡單的政治預測模擬市場,其中代理人參考周圍鄰居的政治傾向,產生自身的信念,再將信念反映到出價上,並以雙向拍賣的方式在市場中交易,這隱含著預測市場統合訊息的功能,最後產生平均價格,該價格與標準差是很好的預測指標。我們研究鄰居的範圍和居住的群聚效果對預測精確性的影響。此外,我們的模型不但可以精確地複製出冷熱門馬偏誤(favorite-longshot bias),在相關條件成立下,更產生逆轉的冷熱門馬偏誤(reverse favorite-longshot bias)。最後,引入小世界網路改變代理人獲取資訊的方式,發現網路效果取代群聚效果,並使逆轉現象消失,結果恢復為傳統在預測市場中的現象-冷熱門馬偏誤。 We model a basic agent-based political prediction market with zero-intelligence (ZI) agents. The ZI agents form their belief by referring to neighbors’ political preference and then make their orders in a double auction future market. The market efficiently aggregates information and dynamically forms trading prices. In a market without further information, average price will be a good predictor. We investigate the market efficiency by setting various radii of neighbor and effects of clustering. Our model replicates favoriate-longshot bias and reverse favoriate-longshot bias under specific conditions. When radius of neighbor is replaced by social network, the reverse bias turns into typical favoriate-longshot bias.