scientific article; zbMATH DE number 3971987

From MaRDI portal
Publication:3738404





[For the entire collection see Zbl 0564.00015.] Let \(X_ 1,X_ 2,...,X_ n\) be nonnegative random variables with joint density f(x), where \(x=(x_ 1,x_ 2,...,x_ n)\). Define the survival function by \(G(x)=P(X_ 1\geq x_ 1,X_ 2\geq x_ 2,...,X_ n\geq x_ n)\), the multivariate hazard rate by \(\lambda (x)=f(x)/G(x)\), and the multivariate vector-valued hazard rate, by \(h(x)=(h_ 1(x),...,h_ n(x))\), where, for each i, we have \(h_ i(x)=-\partial \log G(x)/\partial x_ i.\) The author presents a number of characterizations of bivariate distributions, particularly their degree of closeness to bivariate exponential distributions, in terms of properties of the functions G, \(\lambda\), and h. These include monotonicity properties of \(\lambda\) and h, and the new better than used in expectation property of G, which states that \[ E\{(X-t)(Y-s)| X>t,Y>s\}\leq \int^{\infty}_{0}\int^{\infty}_{0}xydG(x,y) \] for G-almost all nonnegative t, s such that \(G(t,s)>0\). The results are for the most part highly technical in nature. A typical example runs as follows: Let \(\epsilon >0\), let \(n=2\), assume that \(h_ 1\) and \(h_ 2\) are nondecreasing, and assume also that the marginals of G are exponential with parameters \(\lambda_ 1\) and \(\lambda_ 2\). Under these conditions, if \(h_ 1(x,y)+h_ 2(x,y)-\lambda_ 1-\lambda_ 2\leq \epsilon\), then we have G(x,y)-\(\exp (-\lambda_ 1x-\lambda_ 2y)\leq \epsilon \min (\lambda_ 1^{-1},\lambda_ 2^{-1})\).











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3738404)