表題番号:2017B-106 日付:2018/03/26
研究課題株式市場の超高速取引環境の解明と市場へのインパクト
研究者所属(当時) 資格 氏名
(代表者) 商学学術院 大学院経営管理研究科 教授 宇野 淳
(連携研究者) ビジネス・ファイナンス研究センター 招聘客員研究員 五島圭一
(連携研究者) ビジネス・ファイナンス研究センター 客員次席研究員 戸辺玲子
研究成果概要

Cluster Analysis on Trading Behavior of HFT

 

Jun Uno, Keiichi Goshima, Reiko Tobe 

March 26, 2018  

We apply a method of cluster analysis to identify a group of traders who exhibit similar trading patterns in the Tokyo Stock Exchange. In the literature on HFT (high frequency trading), cancelation ratio and inventory ratio are popular proxies to identify traders as HFT, but what should be the threshold for each proxy is difficult to determine objectively. Cluster analysis can overcome such subjective classification of traders. We use DBSCAN (Density-Based Spatial Clustering of Applications with Noisein this study. 

We identify a group of trades with low inventory and exclusive usage of limit orders as a HFT, then examine their behavior in the case of market crash. More specifically, we focus on orders and trades submitted by HFT on five days between May 23 and May 29, 2013 as a crash period, comparing those on prior five days as a benchmark period. We find that they increase presence at the time of market crash and participate as limit order traders. Our classification reveals that the usage of specific threshold such as 20% of cancellation ratio can be misleading due to exclusion of traders from HFT who show very similar behavior but slightly higher cancellation ratio.