通过订单级影子跟随实现无模型被动执行
Model-Free Passive Execution via Order-Level Shadowing
浏览论文内容
中文总结 AI 辅助
本文提出Shadow-PPOV，一种通过跟踪第三方订单流实现无模型被动执行的算法，无需订单簿模型或成交预测，并在CME ES期货重放数据上验证其性能，作为被动订单放置的基准。
中文摘要 AI 辅助
自动执行算法分为基于时间表的家族和基于流动性的家族。本文关注前者，其成员——时间加权平均价格（TWAP）、成交量加权平均价格（VWAP）、成交量百分比（POV）和执行缺口（Implementation Shortfall）——都是基于模型的：每个算法都从显式模型、预测、时间表或控制规则中得出其决策。我们引入了Shadow-PPOV，一种被动POV算法，其订单放置率由观察到的订单流而非已交易成交量决定。传统上，放置被动订单以高效成交涉及订单簿模型和成交预测。Shadow-PPOV用跟踪取代了该预测：在观察到第三方挂单时，它可以在同一交易场所的相同价格上发送自己的限价单，并记录观察到的订单的交易所标识符与其自身订单之间的单一关联。随后，取消由标识符驱动——当被跟随的订单结束时，影子订单被撤回，在取消时立即撤回，在成交后经过短暂宽限期撤回。因此，放置决策是无模型的：价格和交易场所从观察到的订单中直接读取。无模型并非无信息。Shadow-PPOV读取每条订单簿消息，并且仅在参与者刚刚投入资本的位置放置订单，同时不根据其读取的内容计算任何东西。信息是从订单流中继承的，而非从模型中推导的。我们在确定性市场重放模拟器中，对芝加哥商品交易所（CME）ES期货的完整日历年的重放数据评估了Shadow-PPOV，报告了其滑点和延迟敏感性，并将其与激进等价POV进行比较。我们提出将其作为被动订单放置的无模型基准，可据此对预测性放置模型进行评分。该算法和订单簿模拟器已在开源项目kaspar-hft中实现。
英文摘要
Automated execution algorithms are organized into schedule-based and liquidity-seeking families. This paper concerns the first, whose members -- Time-Weighted Average Price (TWAP), Volume-Weighted Average Price (VWAP), Percentage of Volume (POV) and Implementation Shortfall -- are all model-based: each derives its decisions from an explicit model, forecast, schedule or control rule. We introduce Shadow-PPOV, a passive POV whose order-placement rate is set from observed order flow rather than from traded volume. Placing a passive order to fill efficiently conventionally involves an order-book model and a fill prediction. Shadow-PPOV replaces that prediction with tracking: on observing a third-party add, it may transmit its own limit order at the same price on the same venue, recording a single association between the observed order's exchange identifier and its own. Cancellation is then identifier-driven -- the shadow is withdrawn when the order it follows ends, at once on a cancel and after a brief grace window on a trade. The placement decision is thus model-free: price and venue are read off the observed order. Model-free is not information-free. Shadow-PPOV reads every order-book message and places only where a participant has just committed capital, while computing nothing from what it reads. Information is inherited from the flow rather than derived from a model. We evaluate Shadow-PPOV on a full calendar year of replayed Chicago Mercantile Exchange (CME) ES futures in a deterministic market-replay simulator, reporting its slippage and latency sensitivity and comparing it against the aggressive equivalent POV. We propose it as a model-free benchmark for passive-order placement, against which a predictive placement model can be scored. The algorithm and the order-book simulator are implemented in the open-source kaspar-hft project.
发表机构
- M2 Technologies
机构由 AI 辅助整理，请以论文原文为准。
