Combinatorial Arbitrage in Prediction Markets

本文构建了预测市场组合套利框架,以逻辑关系生成联合收益状态,通过凸包一致性、整数规划与全深度订单簿模拟识别可执行套利。基于2026年6月23日至8月3日的Polymarket数据,研究恢复约190万条逻辑关系,覆盖超过25.5万个合约,并发现结算组合在扣除点差和费用后仍可获得显著无风险收益;关系越依赖语义解析和LLM推断,套利机会持续时间越长、实际参与越少。

本文构建了预测市场组合套利框架,以逻辑关系生成联合收益状态,通过凸包一致性、整数规划与全深度订单簿模拟识别可执行套利。基于2026年6月23日至8月3日的Polymarket数据,研究恢复约190万条逻辑关系,覆盖超过25.5万个合约,并发现结算组合在扣除点差和费用后仍可获得显著无风险收益;关系越依赖语义解析和LLM推断,套利机会持续时间越长、实际参与越少。

This study applies recent time-series causal discovery (LPCMCI) to ERCOT day-ahead prices (2019–2024), showing West Texas wind forecasts have become the dominant causal driver of the system-wide price (lambda), with effects on average ~3.4x larger than natural gas price effects; these causal relationships vary systematically by season, time-of-day and region, and the dominant gas benchmark shifts from Waha to Henry Hub over the study period .