基于贝叶斯分析的中国商业银行内部欺诈分析

发布时间：2018-03-02 18:03

本文选题：金融学　切入点：操作风险　出处：《运筹与管理》2012年04期 　论文类型：期刊论文

【摘要】：内部欺诈事件类型是中国商业银行最严重的操作风险类型。但由于操作风险本质特征和中国商业银行内部欺诈损失数据收集年度较短,数据匮乏,小样本数据容易导致参数结果不稳定。为了在小样本数据下进行更准确的度量,本文采用贝叶斯马尔科夫蒙特卡洛模拟方法,在损失分布法框架下,假设损失频率服从泊松-伽马分布,而损失强度服从广义帕累托-混合伽马分布,分析后验分布的形式,获得中国商业银行不同业务线的内部欺诈损失频率和损失强度的后验分布估计,并进行蒙特卡罗模拟获得不同业务线内部欺诈的风险联合分布。结果表明,拟合结果很好,与传统极值分析法相比,基于利用贝叶斯的分析获得的后验分布可以作为未来的先验分布,有利于在较小样本下获得较真实的参数估计,本方法有助于银行降低监管资本要求。
[Abstract]:The type of internal fraud is the most serious type of operational risk in Chinese commercial banks. However, due to the essential characteristics of operational risk and the short period of data collection for internal fraud losses of Chinese commercial banks, the data are scarce. In order to measure more accurately in small sample data, Bayesian Markov Monte Carlo simulation method is used in the framework of loss distribution method. Assuming that the loss frequency is from Poisson gamma distribution and the loss intensity clothing is from the generalized Pareto-mixed gamma distribution, the form of posteriori distribution is analyzed. The posterior distribution estimation of internal fraud loss frequency and loss intensity of different lines of business in Chinese commercial banks is obtained, and Monte Carlo simulation is carried out to obtain the joint risk distribution of internal fraud in different lines of business. The results show that the fitting results are very good. Compared with the traditional extreme value analysis, the posterior distribution based on Bayesian analysis can be regarded as the future prior distribution, which is helpful to obtain the real parameter estimation under the smaller sample. This method is helpful to reduce the capital requirement of the bank.
【作者单位】：山东财经大学工商管理学院;中国科学技术大学管理学院;中国科学院科技政策与管理科学研究所;
【基金】：山东省自然科学基金高校、科研单位专项资助项目(ZR2010GL011)
【分类号】：F832.2

【参考文献】