基于网络表示学习的证券市场操纵行为检测研究

The securities market has become the most important capital market in the market economy. In securities market, listed companies can raise funds and traders can make money in short period, and it is an important way for the country to optimize allocation of resources. Because Chinese securities market is an emerging market, there are a lot of problems showing up, such as excessive speculation, bankers manipulation and herd behavior. As the reflection of the national economic development, the function of securities market has been jeopardized, and even the financial and social stability will be impacted. In recent years, with the development of technology, manipulation of the securities market has become more and more diversified, and manipulation approaches are more and more hidden. Traditional methods of detecting manipulation are no longer effective, and new methods of data-driven detection are urgently needed. Therefore, analysis of traders’ behavior is crucial to understand the change of stock prices and has important theoretical and practical implications to detect the potential manipulators..In this proposal, we study the behavior of participants via analysis of multi-scale structure in heterogeneous networks. We first investigate characteristics of the trading behavior of the institutional and retail account respectively. Then, we focus on the detection of abnormal trading behavior and warning of cross-market manipulation. This proposal aims to the analysis of market characteristics deeply and reveal traders behavior model by studying the trading behavior in the market. The findings are of great practical significance for timely warning of manipulation and regulation of securities market .

证券市场是上市公司实现资金融通和大众进行投资的重要渠道，是国家进行资源优化配置的重要途径。我国证券市场是一个新兴的市场，存在投机行为多、庄家操纵、散户盲从等问题，造成证券市场作为反映国家经济发展晴雨表的功能弱化，甚至会给金融和社会稳定造成冲击。近年来，随着技术的不断进步，证券市场上的操纵手段越来越多样化，操纵行为越来越隐蔽，传统的依靠规则的操纵检测方法不再适用，亟需新的数据驱动的新型检测手段。因此，研究交易者的交易行为对于理解股票价格变化、检测潜在的操纵者有重要的理论价值和现实意义。本项目拟从个体行为建模入手，采用网络分析方法，研究机构账户和散户账户交易行为特征、异常交易行为识别、跨市场操纵行为预警三个方面的基本问题。旨在通过研究市场交易者的交易行为、深入分析市场特性、揭示交易者行为模式和价格形成的内在规律，对于及时预警操纵行为、规范市场秩序具有广泛的现实意义。

习总书记在8月17日召开的中央财经委员会第十次会议中指出，要统筹做好重大金融风险防范化解工作，提升监管数字化智能化水平。近年来，“互联网+”概念结合下的金融市场呈现出一片蓬勃发展的景象，金融相关数据往往以令人叹为观止的速度产生。在海量数据的掩护下，异常行为与异常用户往往就隐藏于其中，难以察觉。这其中既有多主体间信息不透明，信贷风险、投资风险难以全面掌控的原因，导致系统性风险难以有效控制，最终给政府部门、金融机构和投资者造成了巨大损失。因此，本项目从典型的金融市场出发，研究用户间的相互关系、金融市场行为建模、金融市场异常检测。为金融市场监管提供有效工具和指导建议。我们提出了基于网络结构的异常行为检测，在几种典型金融市场如证券市场、比特币市场及银行交易场景中得到了实际的应用效果，分析了大量股票交易数据、期货交易数据、比特币交易数据、银行转账数据。除了金融监管机构，欺诈检测的工作也应用到了蚂蚁金服中的贷款管理中，提高了金融市场监管的效率。