基于共同冲击和异质风险叠加传导的风险传染研究——来自中国上市银行网络的传染模拟

本文将银行系统遭遇外部共同冲击作为研究起点,建立了一个共同冲击和异质风险交互传导与放大的简化模型,冲击的传导包括“原始冲击”、“增量冲击”和“违约冲击”三个风险传染阶段。基于2018年我国15家上市银行的股票收益率和年报数据、2006年至2018年的银行评级数据,本文构建了贝叶斯分层图模型和银行间拆借矩阵,并利用蒙特卡洛模拟测度不同触发银行所引发的系统性风险损失、单个银行的系统性风险杠杆能力(文中定义为“传染乘数”指标)以及政府监管介入的效果。模拟结果显示:共同冲击损失远大于异质风险损失;规模和网络关联性是决定传染乘数的重要因素,且当规模因素不突出时,网络关联性对传染乘数的决定作用相对更强,极容易出现小规模、高关联性银行具有较高的传染乘数;当银行风险资产损失率在10%至25%之间时,造成系统性风险损失的杠杆能力普遍增强;政府监管介入能较好地降低系统性风险。本研究的相关结论为系统性风险的监管设计提供经验证据和参考。

A Study of Risk Contagion Based on the Interaction Between Common Shocks and Idiosyncratic Risks:Evidence From the Simulation of Listed Banks in China

“Systemic risk” is a widely used term that is difficult to define and quantify.The emergence and evolution of systemic risk resembles a “black box” with the process controlled by exogenous or endogenous shocks and where the contagion channels and orders dominated by various factors.To simplify the combination of factors involved, academics usually focus on three kinds of risk: internal contagion of idiosyncratic risk, extensive damage caused by external common shocks, and imbalance caused by the accumulation of risk over time.These three forms of risk are not exclusive, and usually occur simultaneously and interact with each other in real crises.The key to identifying systemic risk events is to estimate the potential “system state” and evaluate the ability of these events to further change the “system state.” The vulnerability of the entire system caused by external shocks determines the depth and breadth of idiosyncratic risk contagion.Thus, the results of this paper shed light on the effects of common external shocks on the banking system and establish a simplified interactive contagion model for common shocks and idiosyncratic risks.We introduce a systemic risk contagion model with three stages: original shocks, incremental shocks, and default shocks, based on their dominant risk factors.The model depicts the dynamic and interactive contagion process of common shocks and idiosyncratic risks, the effects of which spill over into the financial market and inter-bank debt chain.The current paper has three main findings.First, losses caused by common shocks are much greater than those caused by idiosyncratic risks.Second, the key factors that determine the infection multiplier are size and network relevance, and network relevance has a stronger role in determining the infection multiplier when the size factor is not prominent.For example, it is often the case that small-scale and highly related banks have a higher infection multiplier.Third, the leverage that causes systemic loss is generally enhanced when the loss rate of the risk assets of the bank is between 10% and 25%.Furthermore, regulatory intervention by the government can effectively reduce the systemic risk.This study makes the following contributions.First, we deconstruct common external shocks.We assume that there is a bank present to bear the external shock (called the “trigger bank”).This bank's assets loss risk spills over in ordered layers throughout the banking network.The microcosmic simulation of common external shocks is helpful to depict the sensitivity of the systemic loss caused by the assets loss of a single bank (called the “infection multiplier”).As such, this finding will help to inform early warning systems and interventions in cases of systemic risk.Second, we build direct and indirect networks based on risk correlations.To distinguish the transmission paths of common risks and idiosyncratic risks, we construct a relationship network for common risk exposure and idiosyncratic risk exposure.The default order of banks is decided dynamically by mutual verification between the direct debt network of the interbank market and the indirect relational networks of common shock contagion models.Compared with the risk superposition model, this model not only better captures each stage of risk contagion, it also does not rely on the historical default data of banks.Third, we measure the systemic risk caused by different trigger banks.Compared with the leave-one-out method, which evaluates losses by directly removing a single bank one at a time, the trigger bank in this study remains in the simulation network until the solvency bankruptcy occurs.This more closely reflects a real risk scenario.The conclusions of this study provide empirical evidence that will help to make regulatory decisions in the context of systemic risk.First, the results suggest that the assessment of common risk exposure in the banking system should be increased to control systemic risk from the source, and attention should be paid to the identification and evaluation of the network correlation of small and medium-sized banks.Second, in view of the inflection point, a phenomenon in which an infection multiplier triggers banks, we suggest that regulators intensify the blocking intervention mechanism against systemic risk, for example by setting an early warning rate of risk-weighted asset loss to prevent risk escalation.