Speed traps: algorithmic trader performance under alternative market balances and structures
通过人类与算法交易者的双重拍卖实验，发现优先低延迟的算法交易者在多种市场结构中常导致收益下降，仅在平衡市场中表现优于人类；而在非竞争市场中则陷入速度陷阱，捕获的剩余极小。
Abstract Using double auction market experiments with both human and agent traders, we demonstrate that agent traders prioritising low latency often generate, sometimes perversely so, diminished earnings in a variety of market structures and configurations. With respect to the benefit of low latency, we only find superior performance of fast-Zero Intelligence Plus (ZIP) buyers to human buyers in balanced markets with the same number of human and fast-ZIP buyers and sellers. However, in markets with a preponderance of agents on one side of the market and a noncompetitive market structure, such as monopolies and duopolies, fast-ZIP agents fall into a speed trap. In such speed traps, fast-ZIP agents capture minimal surplus and, in some cases, experience near first-degree price discrimination. In contrast, the trader performance of slow-ZIP agents is comparable to that of human counterparts, or even better in certain market conditions.