Tail asymptotics for the sum of two heavy-tailed dependent risks (Q2463693)

For a heavy-tailed bivariate r.v. \((X_1,X_2)\) the tail behaviour of \(X_1+X_2\) is investigated in a general copula framework. Representations and inequalities for \(P(X_1+X_2>x)\) are given. E.g. it is shown that if \(P(X_1>x)\) is regularly varying with an index \(-\alpha\), then \( \lim\sup_{x\to\infty} P(X_1+X_2>x)/P(X_1>x) \) is less then \((\hat\lambda^{1/(\alpha+1)}+(1+c-2\hat\lambda))^{\alpha+1}\) if \(\hat\lambda\leq (1+c)/3\), where \(\hat\lambda=\lim_{x\to\infty}P(X_2>x| X_1>x)\), \(c=\lim_{x\to\infty}P(X_2>x)/P(X_1>x)<1\). Lognormal marginal distribution, Archimedian, Farlie-Gumbel-Morgenstern and linear Spearman copulas are considered as examples.