基于可解释机器学习的银行系统性风险分析  

Systemic risk analysis of the banking system based on interpretable machine learning

作　　者：肖朋林 夏梦 王直杰[1] XIAO Penglin;XIA Meng;WANG Zhijie(College of Information Science and Technology,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学信息科学与技术学院,上海201620

出　　处：《计算机应用》2022年第S02期302-309,共8页journal of Computer Applications

基　　金：上海市哲学社会科学规划课题(2019BGL004)。

摘　　要：针对银行系统性风险的成因分析,由于银行系统参数多,且参数对系统的影响是耦合的、非线性的,传统机理建模的研究方法和计算机仿真分析方法都难以准确分析参数与系统性风险间的映射关系,而最近兴起的可解释机器学习为这种分析提供了可能。针对可解释算法在系统性风险分析中的应用进行探索研究,提出了基于决策树可解释性的分析方法,从宏观层面分析银行系统性风险的问题;通过与计算机仿真结果进行对比分析,验证了可解释性算法在系统性风险研究方面的有效性。实验结果表明,与银行间拆借利率、投资波动率、银行间连接度、资本储蓄比、银行资产连接度相比,储蓄波动率、投资平均收益率、用户储蓄利率与储备金率对银行系统性风险的影响程度更大;同时发现以储蓄波动率、平均投资收益、用户储蓄利息和准备金率这4个参数为主导的银行系统中,储蓄波动率越小的银行系统越稳定;但如果此时降低平均收益率并提高用户储蓄利率到一定水平,则会使得该银行系统从稳定状态变成不稳定状态。Aiming at the cause analysis of banking systemic risk,as there are several parameters in the banking system and the influence of parameters is coupled and nonlinear,it is difficult to effectively examine the mapping relationship between parameters and systemic risk by using traditional mechanism modeling research methods and computer simulation analysis methods.However,this kind of study is now possible because of the emergence of interpretable machine learning.Based on an interpretable decision tree,a banking systemic risk analysis method was proposed,and as a preliminary exploration of the risk analysis method,the existing interpretable machine learning algorithms were applied.At the macro level,the causes of systemic risks in the banking system were analyzed.By comparing the results of computer simulation analysis,the effectiveness of the interpretable algorithm in systemic risk research was verified.According to the results of the experiments,savings volatility,average return on investment,user savings rate and reserve rate have a greater impact on banking systemic risk than interbank loan rate,investment volatility,interbank connection,capital-savings ratio and bank asset connectivity.Additionally,in a banking system dominated by these four parameters,such as savings volatility,average return on investment,user savings interest and reserve rate,the smaller the savings volatility,the more stable the banking system.But when the average return on investment decreases and the user savings rate increases to a certain level,the banking system will become unstable.

关 键 词：可解释性 机器学习 决策树 部分依赖图 系统性风险 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]