BIGDATA: IA: Collaborative Research: Detecting Financial Market Manipulation: An Integrated Data- and Model-Driven Approach

BIGDATA：IA：协作研究：检测金融市场操纵：一种集成的数据和模型驱动方法

• 批准号: 1741026
• 负责人: David Byrd
• 金额: $ 33万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-09-01 至 2021-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1741026&HistoricalAwards=false
• 关键词: BIGDATA; IA; Collaborative; Research; Detecting

Financial stock market manipulators can profit illegally by misleading investors about market conditions. For example, in several recent incidents, manipulators successfully spoofed markets by inserting orders that deceived investors about supply or demand for the security. This kind of behavior has increased with the prevalence of algorithmic trading. It imposes substantial harm to the economy, by reducing the efficiency of capital allocation, and more seriously, threatening to compromise the integrity and stability of financial markets. Spoofing is difficult to detect because the underlying actions have legitimate purposes as well as nefarious ones. This project will apply innovative approaches to improve detection and deterrence of market manipulation.The project will integrate data-driven methods, including calibration of detectors with normal background activity and extraction of manipulation signatures from enhanced time series, with model-based techniques for characterizing manipulation strategies based on strategic analysis of market microstructure. The key idea is to use simulation and optimization to generate successful manipulation strategies for trading models calibrated from available market data streams. These strategies will then be injected into the trading models, to produce enhanced data streams that include labeled manipulation activity. Having labeled activity enables the application of machine learning techniques to extract signatures of spoofing activity, which can be used to construct surveillance and audit algorithms. Methods produced in this project in conjunction with guidance on market design and regulation policy can contribute to reducing the threat from increasingly capable market manipulators.

金融股市操纵者可以通过误导投资者了解市场状况来非法获利。例如，在最近发生的几起事件中，操纵者通过插入欺骗投资者有关证券供应或需求的订单，成功地欺骗了市场。随着算法交易的盛行，这种行为有所增加。它降低了资本配置的效率，对经济造成了重大损害，更严重的是，有可能损害金融市场的完整性和稳定性。欺骗很难被发现，因为潜在的行为既有合法目的，也有恶意目的。该项目将应用创新方法来改进对市场操纵的检测和威慑。该项目将集成数据驱动的方法，包括校准具有正常背景活动的探测器和从增强的时间序列中提取操纵签名，以及基于模型的技术，用于基于市场微观结构的战略分析来表征操纵策略。关键思想是使用模拟和优化来为根据可用市场数据流校准的交易模型生成成功的操纵策略。然后，这些策略将被注入交易模型中，以生成包括标记的操纵活动的增强数据流。标记活动可以应用机器学习技术来提取欺骗活动的签名，从而可用于构建监视和审计算法。该项目提出的方法与市场设计和监管政策的指导相结合，有助于减少日益强大的市场操纵者的威胁。

项目成果

Fund Asset Inference Using Machine Learning Methods: What’s in That Portfolio?

使用机器学习方法进行基金资产推断：该投资组合中有什么？

• DOI: 10.3905/jfds.2019.1.005
• 发表时间: 2019
• 期刊: The Journal of Financial Data Science
• 作者: Byrd, David;Bajaj, Sourabh;Balch, Tucker Hybinette.
• 通讯作者: Balch, Tucker Hybinette.

Stability Effects of Arbitrage in Exchange Traded Funds: An Agent-Based Model

交易所交易基金套利的稳定性效应：基于代理的模型

• 发表时间: 2021
• 期刊: 2nd ACM International Conference on AI in Finance (ICAIF’21
• 作者: Shearer, Megan;Byrd, David;Balch, Tucker Hybinette;Wellman, Michael P.
• 通讯作者: Wellman, Michael P.

Get real: realism metrics for robust limit order book market simulations

变得真实：稳健的限价订单簿市场模拟的现实指标

• DOI: 10.1145/3383455.3422561
• 发表时间: 2020
• 期刊: ICAIF '20: Proceedings of the First ACM International Conference on AI in Finance
• 作者: Vyetrenko, Svitlana;Byrd, David;Petosa, Nick;Mahfouz, Mahmoud;Dervovic, Danial;Veloso, Manuela;Balch, Tucker
• 通讯作者: Balch, Tucker