Evolution of the Chinese guarantee network under financial crisis and stimulus program

Financial networks, characterized by interconnected entities and mutual business relationships, are susceptible to systemic risk and cascading failures. Guarantee relationships, where one firm assumes another's debt obligations in case of default, are crucial components of these networks. Understanding the evolution of such networks during downturns is vital for managing systemic risk. China's credit market, being the world's largest, offers a unique case study. The 2007-2008 global financial crisis prompted a massive RMB ¥4 trillion stimulus program in China, aimed at mitigating the economic impact. While the stimulus succeeded in sustaining economic growth, it also led to a surge in debt and significant structural changes in the credit system. Many state-owned enterprises (SOEs), previously ineligible for loans, gained access through mutual guarantees, raising concerns about potential future financial disruptions. Subsequent monetary policy adjustments, including increased reserve requirements and interest rate hikes, attempted to counter these risks. Despite qualitative analyses of the stimulus program, quantitative, data-driven insights into the changes in guarantee networks and their relation to cascading failure risks remain limited. This study aims to fill this gap by using a large-scale dataset of Chinese guarantee network to provide a quantitative understanding of how the guarantee network evolved during the financial crisis, stimulus program, and subsequent monetary policy changes. This analysis will leverage network science methodologies to understand the network's topological structure and its relationship to economic policies and contagion risks, providing data-driven insights to inform economic policymaking and risk management.

Existing research on guarantee networks often relies on small-scale datasets, limiting insights into the stability of the entire credit market. While small-scale analyses can reveal risk connections among individual clients, they lack the breadth to understand the systemic properties and evolution of nationwide credit networks. The application of network science methodologies to analyze financial systems is growing, with studies focusing on global banking systems, international financial networks, and corporate board interlocks. However, large-scale empirical studies of nationwide guarantee networks are needed to understand their global topological properties and dynamic behavior under economic shocks and policy interventions.

The researchers utilized a comprehensive dataset provided by a major Chinese regulatory body. This dataset covers the period from January 2007 to March 2012 and encompasses nearly 80% of total loans in China, including data from 19 major banks. The data includes monthly information on loan-level guarantee relationships (borrower, guarantor, loan amount, and relationship duration) and firm-level fundamentals (assets, liabilities, credit lines, etc.). The data is divided into three phases: Phase 1 (April 2007 – November 2008) representing the global financial crisis; Phase 2 (December 2008 – December 2010) encompassing the stimulus program; and Phase 3 (January 2011 – March 2012) representing the post-stimulus adjustment period. The guarantee network was modeled as a directed network, with nodes representing firms and directed edges representing guarantee relationships. The researchers analyzed various topological properties of the network, including network size, density, degree distribution, clustering coefficient, reciprocity, and the size of strongly and weakly connected components. They also examined financial characteristics of firms within the network, such as average assets, liabilities, and loans. Exponential Random Graph Models (ERGMs) were employed to analyze the significance of mutual guarantee relationships in network formation. A simulation model, using a Fermi distribution to model default probabilities, was used to assess the resilience of the guarantee network to cascading failures under different conditions. The simulation incorporated real-world data on firm assets, liabilities, and guarantee relationships to capture the dynamic evolution of the network. The model was employed to assess how different network structures and policies affect systemic risk and cascading failures.

The analysis revealed several key findings. First, the Chinese guarantee network exhibited scale-free properties throughout the study period, with power-law degree distributions. Second, isolated mutual guarantee relationships were significantly more frequent than in a random network. These relationships primarily involved firms with low assets, high default rates, and low credit lines. Third, the network demonstrated a decentralized structure, with a relatively small largest strongly connected component. Dynamic analysis revealed distinct patterns across the three phases. During Phase 1 (financial crisis), the network shrank, becoming less connected but more stable. In Phase 2 (stimulus program), the network grew rapidly, with a significant increase in mutual guarantee relationships, particularly among firms with low assets and high risk. This surge in mutual guarantees resulted from the stimulus program’s expansion of credit access to previously ineligible firms. A brief dip in mutual guarantee relationships occurred in late 2009, likely due to People’s Bank of China’s fine-tuning of the stimulus. However, the increase resumed until the end of 2010. Phase 3 (post-stimulus) showed a decline in mutual guarantees and an increase in network resilience following monetary policy adjustments. The ERGM analysis confirmed the importance of mutual guarantee relationships in shaping the network's structure. The simulation model demonstrated that a higher prevalence of mutual guarantees led to greater fragility and a higher risk of cascading failures. Counterintuitively, the financial crisis increased network stability by eliminating weaker firms and their connections, while the stimulus program, by propping up these weaker firms, increased systemic risk. This is explained through survivorship bias. The study highlights that government bailouts can have unintended consequences, potentially increasing future systemic risk despite mitigating immediate crisis effects. The key events (Bankruptcy of NCFC, Bankruptcy of LB, Start of CESP, and End of CESP) align well with the shifts observed in network structure and risk.

The findings challenge conventional wisdom regarding systemic crises. The results show that the financial crisis, although devastating, resulted in a more resilient network by eliminating weak firms, while the government bailout, intended to prevent immediate collapse, ultimately increased systemic risk. This counterintuitive result emphasizes the importance of understanding the complex interplay between economic shocks, government interventions, and the structural properties of financial networks. Survivorship bias plays a key role in explaining this phenomenon. The study contributes to the literature by providing large-scale empirical evidence of network evolution in a major economy during a period of significant financial turmoil. The findings highlight the importance of considering network structure when designing economic policies and managing systemic risk. The study's methodology, combining empirical network analysis and agent-based simulation, offers a novel approach to evaluating the resilience of financial systems.

This study presents the first quantitative characterization of the evolution of China's entire guarantee network. The findings demonstrate that while mutual guarantees can help low-quality firms temporarily, they increase systemic risk. The study provides valuable insights for policymakers on managing credit systems and responding to financial crises. Future research could explore other types of financial networks and investigate the long-term consequences of government bailouts.

The study relies on data from a single country (China) and a specific time period. The generalizability of the findings to other countries or contexts may be limited. The data used is confidential and was provided by a specific regulatory body. While covering a large portion of the Chinese credit system, the data may not encompass the complete picture. The simulation model relies on assumptions about default probabilities and risk contagion. Modifications of these assumptions could alter the simulation results.