中文摘要：

对低频率高损失的操作风险进行量化非常困难，其特点为较少发生，而一旦发生通常都会造成巨额损失，因此数据匮乏是一个严重的挑战，内部欺诈风险是此类风险的典型代表，也是中国商业银行有代表性的重大操作风险类型。在损失分布法的框架下，通过极值理论对内部欺诈风险进行度量；考虑到使用损失数据对频率直接拟合带来的问题，借助POT模型的重要性质对内部欺诈风险强度和频率进行估计；在此基础上，使用基于吉布斯抽样的贝叶斯MCMC方法估计POT模型的参数，以解决样本数据不足时极大似然估计中误差增大的问题。在这一方法框架下，以中国商业银行的内部欺诈损失数据为样本，对中国商业银行的内部欺诈风险进行度量，并估计了相应的经济资本。

英文摘要：

Internal fraud is the significantly representative loss type in China banking industry. For this low frequency/high severity loss, however, risk measurement and the corresponding economic capital estimation is very difficult, due to the lack of adequate loss data. This paper makes use of the POT Model, which proved to be effective in extreme loss measurement, to deal with the loss severity and frequency under the Loss Distribution Method framework. A Bayesian MCMC model with Gibbs sampling is established to estimate the parameters required in POT. Furthermore, based on some internal fraud loss data from Chinese banks, the paper gives the estimation to the internal fraud risk of an assumed moderate - size bank and the corresponding risk capital for the first time.