基于变位置参数贝叶斯预测银行内部欺诈研究

发布时间：2018-08-30 13:40

【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,为了在小样本数据下进行更准确的度量,本文采用贝叶斯后验预测分布方法,其中,假设损失频率服从泊松一伽马分布,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。由于在广义帕累托分布的参数估计中,位置参数的确定对估计结果的影响很大,因此,本文采用变位置参数线性趋势的贝叶斯分析以增强参数预测稳定性,降低位置参数选择对结果产生的影响,获得中国商业银行内部欺诈损失频率和损失强度的后验预测分布和边际分布,进而采用蒙特卡罗模拟,联合损失频率分布和损失强度的预测分布获得内部欺诈的风险联合分布。与传统Poisson-GPD极值分析法相比,在险值和预期超额损失明显降低,有利于银行降低内部欺诈操作风险资本。利用贝叶斯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。
[Abstract]:Internal fraud is the most serious type of operational risk in China's commercial banks. However, due to the intrinsic characteristics of operational risk and the shortage of data, this paper adopts Bayesian posterior distribution method to measure the internal fraud loss of Chinese commercial banks more accurately with small sample data. The loss frequency obeys the Poisson-Gamma distribution and the loss intensity obeys the generalized Pareto-Mixed Gamma distribution. The form of a posterior distribution is analyzed. The posterior and marginal distributions of fraud loss frequency and loss intensity in Chinese commercial banks are obtained by reducing the influence of location parameter selection on the results. Then the Monte Carlo simulation is used to obtain the joint distribution of fraud risk with the joint distribution of loss frequency and loss intensity. Compared with the traditional Poisson-GPD extreme value analysis, the risk value and the expected excess loss are obviously reduced, which is beneficial for banks to reduce the risk capital of internal fraud operation.
【作者单位】：山东财政学院工商管理学院;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) 国家自然科学基金资助项目(70701033)
【分类号】：F832.2;F224

【参考文献】