文献类型：期刊文献

英文题名：Double correlation model for operational risk: Evidence from Chinese commercial banks

作者：Xu, Chi[1];Zheng, Chunling[2];Wang, Donghua[1,3];Ji, Jingru[1];Wang, Nuan[1]

机构：[1]East China Univ Sci & Technol, Sch Business, 130 Meilong Rd,POB 114, Shanghai 200237, Peoples R China;[2]Jingdezhen Univ, Dept Informat Technol, Jingdezhen 333000, Peoples R China;[3]East China Univ Sci & Technol, Inst Financial Engn, Shanghai 200237, Peoples R China

年份：2019

卷号：516

起止页码：327

外文期刊名：PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS

收录：;EI(收录号：20184406012182);WOS:【SSCI(收录号:WOS:000452941700031)，SCI-EXPANDED(收录号:WOS:000452941700031)】；

基金：We are grateful to Wei-Xing Zhou for helpful discussions. This study is supported by National Science Foundation of China (Grant No. 71171083; Grant No. 71771087).

语种：英文

外文关键词：Operational risk; LDA; Double correlation; t-copula

摘要：The Basel Accord requires commercial banks to meet a capital requirement for operational risk. Therefore, a reliable operational risk measurement is of great significance for financial institutions. In this paper, based on the framework of LDA, we propose a more appropriate model (Double Correlation model) in correlation structure construction, which considers both frequency correlation and mean severity correlation simultaneously. On this basis, annual loss scenarios for each risk cell are restored and VaR is also estimated by Monte Carlo simulation. Empirical study adopts the data of Chinese commercial banks disclosed to the public and contains the comparison of double correlation model and the other four correlation models. The result indicates that the VaR estimated by the double correlation model is significantly lower than that upon the assumption of comonotonicity, proving the existence of risk diversification. Meanwhile, the VaR estimated by double correlation model is significantly lower than that of the other two correlation models when confidence level is lower than 99% but the relationships between them will reverse if the confidence level rises to more than 99%. (C) 2018 Elsevier B.V. All rights reserved.