基于智能多Agent的金融破产传染分析和预警

It is widely recognized that the global financial and economic system exhibits a tendency to become more and more remarkably intertwined with the continual increase in economic globalization, and this provided a diffusion path for the United States (U.S.) subprime mortgage crisis in 2008. In a network economy, when a firm goes bankruptcy or becomes financially distressed, probabilities of bankruptcy in connected firms increase. In other words, a single bankruptcy may have systemic repercussions through an avalanche of bankruptcies. It has been widely conceded among academic researchers and business practitioners that it is an important challenge to better understand these dynamics and complexities of financial contagion as the analytical underpinning of an early warning system with respect to financial instability. But there is a larger problem. The mathematical machinery is cumbersome and requires drastic assumptions and simplifications to get tractable results. At the same time, software agents enjoy the following characteristics: autonomy, social ability, reactivity and pro-activeness. Intelligent multi-agent-based systems potentially present a way to model the financial economy as a complex model, while taking human adaptation and learning into account. Such systems allow for the creation of a kind of virtual universe, in which many players can act in complex and realistic ways. Therefore, this research is to develop some intelligent multi-agent systems to further our understanding and communication about financial contagion in companies and financial institutions. In the aim to aid and instruct decision makers such as corporate managers and investors, the other main objective of our work is to propose the methodological process for analysis, verification and prediction for investigating some measures for controlling bankruptcy diffusion. This research is useful for risk management, financial intelligence, and etc.

2008年全球性金融危机形成的关键原因是美国次债危机在全球金融市场的蔓延和传染。特别是在"破产雪崩"时，单一的破产事件将会极大增加市场中相关企业或机构的破产风险。如何度量金融传染效应，尤其是破产情况下的传染，并及时做出预警是近几年来金融领域非常关注的研究问题。然而，经典研究方法所需的前提假设和简化条件在高度复杂的市场中，大多被违背，研究效果并不理想。软件Agent由于具备优良的特性，如自治性、异构性、动态性等，能够构建出的智能多Aagent系统被广泛认为更符合真实市场特性。因此，本项目通过构建智能多Agent系统研究如何合理解释公司和金融机构间的传染现象，对特定环境下可能导致的扩散途径和传染现象进行有效预警，并最终对降低特定关系中的破产传染风险提供一些有效的建议。本项目的开展，在风险管理、投资决策、金融智能等领域都会有较好的应用价值。

本项目针对公司和金融机构间的传染现象，研究了构建智能多Agent系统来满足复杂网络环境和交互结构约束，主要研究成果包括三部分：1）偶发性事件借助企业间的密切关系而引发的传染现象研究；2）复杂破产传染现象的确定性和非确定模型及其动态分析；3）多重网络环境中混杂传染的模型。通过本项目的研究，有效地解决了复杂网络环境中特定环境下扩散途径和传染现象的建模。.本项目已完成了原申请书中所要求的研究内容，并达到了预期研究目标。取得了一系列创新成果，发表（含录用）学术论文4篇（包括IEEE TRANSACTIONS ON SYSTEMS, 和AAMAS等顶级学术会议），其中SCI 检索期刊2篇，EI 检索会议2篇；申请发明专利1项、软件著作权2项。参加学术会议5次，在研究过程中与新加坡管理大学建立了良好的学术合作关系。