部分藤copula:基于简化假设的依赖度量与近似
统计方法学
2021-01-11 v2
摘要
简化藤copula(SVCs),或称成对copula构造,已成为高维依赖建模的重要工具。迄今为止,SVCs的设定与估计均在简化假设下进行,即藤的所有双变量条件copula均被假定为双变量无条件copula。我们引入部分藤copula(PVC),它提供了一种新的多变量依赖度量,并在利用SVCs逼近多变量分布中发挥主要作用。PVC是一种特殊的SVC,其中为任意边分配一个j阶偏copula,并构成双变量偏copula的多变量类比。我们研究了PVC在多大程度上描述了底层copula的依赖结构。我们表明PVC并不最小化与真实copula的Kullback-Leibler散度,且满足简化假设的最佳近似由藤伪copula给出。然而,在正则条件下,成对copula构造的逐步估计量收敛于PVC,无论简化假设成立与否。此外,我们阐明了为何PVC是实践中最佳可行的SVC近似。
引用
@article{arxiv.1510.06971,
title = {The partial vine copula: A dependence measure and approximation based on the simplifying assumption},
author = {Fabian Spanhel and Malte S. Kurz},
journal= {arXiv preprint arXiv:1510.06971},
year = {2021}
}
备注
A previous version of this paper was circulated on arXiv under the title "Simplified vine copula models: Approximations based on the simplifying assumption"