A diagnostic test for specification of copulas under censorship
提出一种适用于审查数据的连接函数设定检验，可诊断被拒绝的零假设并提供替代连接函数建议，通过蒙特卡洛模拟和保险数据验证了有效性。
We propose a copula specification test for copula-based multivariate survival models under censorship. This flexible test is applicable to both Archimedean and non-Archimedean copulas and provides useful pointers for alternative copula construction when a null distribution is rejected. It is shown to be consistent and asymptotically distribution free. We demonstrate its good finite sample performance via Monte Carlo simulations. We apply this test to an insurance dataset on losses and expenses. Our test rejects the hypothesis of Gaussian copula. Furthermore, its diagnostic information suggests an alternative copula specification that captures the extreme-value dependence exhibited in the data.