算法交易对金融市场的影响: 基于网络博弈理论的研究

The study on influence of algorithmic trading on financial markets will assist financial practitioners and regulators to understand the microscopic mechanisms of financial markets and control the potential systemic risk from algorithmic trading. This project aims to study this influence of algorithmic trading on continuous double auction markets through network games. Firstly, Partially Observable Markov Decision Process is applied to identify if a trader is an algorithmic trader, and if an algorithmic trader is one high frequency trader via trading records of traders and market information. The properties of network games and the attributes of traders are used to model the network formation of traders in the markets, when they are cooperative or noncooperative. Further, the diffusion process of traders is built. Decision-making processes of traders are described by deep learning methods. Based on studies described above, the microscopic mechanism for influence of algorithmic trading on continuous double auction markets is completed. Secondly, the microscopic market mechanism is applied to identify and evaluate the potential systemic risk by collecting and mining the behavioural data of the markets and traders using big data and artificial intelligence techniques. Then, a new monitoring and control system for the systemic risk is built. Historical events with systemic risk can be replicated based on this system. Finally, we investigate empirical studies on influence of algorithmic trading on the whole market performance and individual traders’ behaviours in terms of algorithmic trading strategies, types of algorithmic traders, and market mechanisms.

研究算法交易对金融市场的影响有助于金融从业者和监管者理解市场微观结构，控制产生的系统性风险。本项目基于网络博弈理论开展算法交易对连续双向拍卖市场的影响的研究。首先，应用部分可观测马尔科夫决策过程，通过投资者交易行为和市场信息来识别算法交易者和非算法交易者，高频交易者和低频交易者。利用网络博弈理论和投资者属性，研究合作或非合同下投资者之间的网络构成理论，构造投资者之间的信息传导过程。通过深度学习的方法，描述算法交易者和非算法交易者的决策过程，进而构造算法交易对连续双向拍卖市场影响的微观机理。其次，基于建立的市场微观机理，通过大数据和人工智能技术，采集和发掘市场和投资者的行为数据，识别和评估算法交易产生的潜在系统性风险，建立系统性风险监控与预警体系，对历史性系统性风险事件进行复盘。最后，从算法交易策略、算法交易者种类和市场机制的角度，展开算法交易对市场整体表现和个人投资者行为的影响的实证研究。

随着人工智能、大数据技术等计算机技术的发展，算法交易已经成为金融市场活动中的重要组成部分。面向当下数字金融的发展机遇，研究算法交易对市场的影响有助于更好的理解和监管金融市场。首先，本项目研究了金融市场不同的算法交易策略，包括基于Hawkes-Contact模型的投资者情绪资产定价策略、基于投资者情绪和宏观因素的异质性主体价格形成、基于深度强化学习的期权对冲策略、基于狼群算法和支持向量机的选股策略。然后，从实证的角度，基于机器学习研究了股票市场的市场操纵行为识别问题。其次，从订单薄的角度，研究了2008年与2015年两次股灾中国股票市场下行期间的流动性特征及系统性风险，分析和对比了股票市场下跌前后的流动性和系统性风险各个指标的变化。接着，研究了基于网络博弈理论的市场微观机制。基于人工智能的多智能体系统的理论，构建了包含市场操纵者、技术交易者、基本面交易者和噪声交易者的双向拍卖市场机制。探讨了不同类型的交易者的表现，以及对金融市场的影响。同时，研究和分析了包括高频交易、低频交易和噪音交易者的双向拍卖市场机制。