Consistency of \(M\)-estimators and one-sided bracketing (Q4387136)

This paper is devoted to consistency of \(M\)-estimators. Sections 2 and 7 are close to the paper of \textit{M. Huber}, Proc. fifth Berkeley Symp. Math. Stat. Probab., Univ. Calif. 1965/66, 1, 221-233 (1967; Zbl 0212.21504), and show how his work is extended -- the main extension being that Huber's topological assumptions on \(\Theta\) are replaced by Glivenko-Cantelli properties. Section 3 defines one-sided Glivenko-Cantelli and bracketing properties and states some theorems on them to give consistency. Section 4 treats the ``log-likelihood'' case. Section 5 treats Glivenko-Cantelli classes of functions and recalls Talagrand's characterization of these classes using it to show that the Glivenko-Cantelli property is preserved under compositions \(f\to g\circ f\) with any fixed Lipschitz, monotone function \(g\). Section 6 gives as an example a way of constructing some specific classes of functions on Euclidean spaces for which limit theorems hold uniformly in \(P\). For the entire collection see [Zbl 0883.00024].