DeepEye:一个基于深度学习的程序化交易识别与分类方法  

DeepEye:a deep learning-based method of recognition and classification of program trading

作　　者：徐广斌[1] 张伟 XU Guangbin;ZHANG Wei(Capital Market Institute of Shanghai Stock Exchange,Shanghai 200120,China;Product Innovation Center of Shanghai Stock Exchange,Shanghai 200120,China)

机构地区：[1]上海证券交易所资本市场研究所,上海200120 [2]上海证券交易所产品创新中心,上海200120

出　　处：《大数据》2018年第5期94-102,共9页Big Data Research

摘　　要：基于沪市A股交易数据,对A股市场程序化交易行为进行系统分析,构建程序化交易识别及分类特征指标体系,结合深度学习技术提出A股市场程序化交易的智能识别与分类方法——DeepEye,该方法可对程序化交易进行识别并分类。在真实交易行为数据集上的实验表明,所提出的方法在识别和分类上取得了较高的准确率,验证了将深度学习用于证券市场行为监管的可行性和有效性。该方法已辅助用于资本市场投资者画像及市场一线行为监管。Program trading behavior in A-share market has been systematically analyzed based on the Shanghai Stock Exchange’s latest trading data and a feature indictor system has thus been built up for characterizing and classifying the program trading in the market. Furthermore, based on the deep learning technology, the A-share program trading intelligentized recognition and classification method, DeepEye, has been proposed, which enables program trading behavior in the market to be recognized and classified. The accuracy of the pilot implementation got about 70% which verified the feasibility and effectiveness of the new method. The proposed method can serve as an auxiliary measure to existing investor portraits and behavior supervision analysis for market regulation and can be a reference for improving the existing program trading regulatory rules.

关 键 词：投资者行为 程序化交易 行为监管 深度学习 分类 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]