On the correlation order (Q2497800)

Let \((X_1,X_2,\ldots,X_n)\) and \((Y_1,Y_2,\ldots,Y_n)\) be two random vectors with the same marginals. The authors study the correlation order according to which \((X_1,X_2,\ldots,X_n)\) is smaller than \((Y_1,Y_2,\ldots,Y_n)\) if \[ \text{Cov}(f(X_i,i\in A_1),g(X_i,i\in A_2))\leq\text{Cov}(f(Y_i,i\in A_1),g(Y_i,i\in A_2)) \] for any pair of disjoint subsets \(A_1\) and \(A_2\) of \(\{1,2,\dots,n\}\), and for any nondecreasing functions \(f\) and \(g\) for which the covariances exist. It is shown that the correlation order implies the supermodular order. This extends the main result of \textit{T. C. Christofides} and \textit{E. Vaggelatou} in: ``A connection between supermodular ordering and positive/negative association. J. Multivariate Anal. 88, 138--151 (2004; Zbl 1034.60016). The authors apply their main result to obtain a majorization comparison of vectors of functions of order statistics that correspond to random vectors \((X_1,X_2,\dots,X_n)\) and \((Y_1,Y_2,\dots,Y_n)\) that are comparable according to the correlation order. The latter result extends a result of \textit{T. Hu} and \textit{J. Hu} [Comparison of order statistics between dependent and independent random variables, Stat. Probab. Lett. 37, 1--6 (1998; Zbl 0888.62048)].