关键词: 图生成网络, 广义方差分解, 金融风险传染, 风险预警, 网络预测

Abstract: Preventing and resolving systemic financial risks is the eternal theme of financial regulation. With the comprehensive deepening of financial reform in China, various types of financial institutions have increasingly overlapped business scopes and customer bases, leading to more frequent fund flows, strengthening the interconnectedness of financial institutions, and forming a complex financial network. As a result, the risk impact faced by a single financial institution may spread through the financial network to different financial institutions, departments, and the entire financial system, amplifying the intensity and scope of risk impact, and even causing systemic risks. Against the backdrop of China's economic development entering a phase of industrial structure adjustment and slowing growth, internal accumulated risks are gradually being exposed. Issues such as the breakdown of funding chains in real estate enterprises, excessively high non-performing asset ratios in some local commercial banks, and frequent corporate bond defaults are emerging. Externally, the environment faces impacts from international geopolitical events such as Sino-U.S. trade frictions, Russia-Ukraine conflicts, and energy crisis. Therefore, effectively addressing the transmission of financial risks caused by internal and external shocks to promote steady economic recovery remains a major challenge for the financial system. The report of the 20th National Congress of the Communist Party of China further points out that “preventing financial risks still needs to address many major issues” and emphasizes the need to “strengthen the financial stability guarantee system, bring all types of financial activities under regulation in accordance with the law, and maintain the bottom line of preventing systemic risks.” Therefore, establishing a scientific and accurate early warning model regarding the transmission paths and evolutionary laws of financial risks holds significant theoretical value and practical significance for enhancing the supervision of systemically important financial institutions, preventing and resolving systemic financial risks, improving China's macro-prudential management system, and promoting high-quality economic development.
Therefore, this paper proposes a novel financial risk contagion early warning model based on a graph generating network model to predict the financial risk contagion network. Firstly, the risk contagion network structure, contagion paths, and risk intensity of 81 financial institutions in China from 2013 to 2022 are characterized based on LASSO-VAR-GFEVD. Secondly, using the risk contagion intensity of each financial institution and the risk contagion network as training dataset, the graph generating network models CondGEN and SCGG are employed to learn the information within the risk contagion network, thus constructing the financial risk contagion early warning model. The model's performance is comprehensively evaluated using metrics such as the Jaccard coefficient and event study method. Finally, extreme risk shocks in the capital market are simulated, and based on social network analysis methods, the early warning of future risk contagion paths, intensity, and systemically important institutions for the next year are identified.
Our empirical results show: (1)The structure of China's financial risk contagion network is predictable, and models based on graph generating network can accurately identify financial risk contagion network in China, particularly during crisis periods, with an average prediction accuracy of 85%. (2)Compared to the SCGG model, the CondGEN model exhibits superior overall performance, and systemically important institutions it predicts are closer to the sample network. (3)Under simulated shocks, the model's early warning of the financial risk contagion network reveals significant “outflow” and “inflow” community structures, with substantial differences in risk contagion intensity within and between different communities. (4)Emerging financial institutions such as real estate, diversified finance, and cross-industry holdings rank high in terms of contagion intensity within their respective communities and across the entire early warning network, indicating their roles as sources of risk contagion in the future financial system. The research results of our work provide new and feasible tools and references for financial regulatory authorities in preventing and resolving systemic financial risks.

Key words: graph generative network, LASSO-VAR-GFEVD, risk contagion, risk warning, network prediction