Médianes vectorielles. (Vectorial medians) (Q919697)

It is well-known that for a real-valued integrable random variable X the median can be defined as a number for which the expression E \(| X- a|\) attains its minimum. The author uses this fact to define the notions of a median and a conditional median for random variables taking values in a Banach lattice B. The author shows that if \({\mathcal B}\) is a sub-\(\sigma\)-algebra of \({\mathcal A}\) and if B is order continuous, then for a random variable \(X\in L_ 1(\Omega,{\mathcal A},P;B)\) there exists a conditional median, i.e. a random variable \(Y_ 0\in L_ 1(\Omega,{\mathcal B},P;B)\) such that \[ E | X-Y_ 0| \leq E | X-Y|, \] where Y is any random variable from \(L_ 1(\Omega,{\mathcal B},P;B)\). In particular, if \({\mathcal B}\) is the trivial \(\sigma\)-algebra, then \(Y_ 0\) is an element of B and this element is said to be a median of X. It should be noted that \(| \cdot |\) denotes the absolute value (but not the norm) of the Banach lattice B, and that the problem of minimization of E \(\| X-a\|\) is a completely different one. The author studies the question of convergence of conditional medians in the case when \({\mathcal B}_ n\) is a filtration, as well as the question of consistency of their estimate.