A New Approach for Detecting High-Frequency Trading from Order and Trade Data

Abstract

Abstract We suggest a two-step approach in detecting HFT activity from order and trade data. While the first step focuses on multiple actions of an order submitter in low latency, the second searches for the surroundings of these orders to link related orders. On a sample of 2015 data from Borsa Istanbul, we estimate that average HFT involvement is 1.23%. HFT activity is generally higher in large cap stocks (2.88%). Most HFT orders are in the form of very rapidly canceled order submissions. A robustness check reveals a mean accuracy rate of 97% in the linkage of orders.