Variables concomitantes et information qualitative (Q760983)

(X\({}_ i,Y_ i)\) is an n-sample of r.v. (X,Y). If \(X_{(1)}<...<X_{(i)}<...<X_{(n)}\), then \((X_{(i)},Y_{[i]}| i=1...n)\) are the concomitants with respect to X (the \(Y_{[i]}\) not being necessarily ordered). Given the distribution function of (X,Y), the order of \(X_ i\) brings some more information onto \(Y_ i\). If the question is to choose a ''project'' i, its utility being \(u(\psi_ i(Y_ i))\) (where \(\psi_ i\) is a scale function), what is the value of this extra-information? For any permutation \(\tau\) of \(\{\) 1,...,n\(\}\) define: \(\Omega_{\tau}=\{\omega \in \Omega;\quad X_{\tau (1)}(\omega)<...<X_{\tau (n)}(\omega)\}.\) Due to symmetry, \(P(\Omega_{\tau})=1/n!\). The expected value of the best choice is then \(\hat p=\max_{i}E(u(\psi_ i),\) whereas the information on X provides an expected utility: \(p^*=(1/n!)\sum_{\tau}p^*_{\tau}\) where \(p^*_{\tau}=\max_{i}E u(\psi_{\tau (i)}(Y_{[i]})).\) Two examples are given \((Y/X=x\) uniform and Gaussian (X,Y)).