HDS

Exercise 4.1: Continuity of Functionals

chapter 4

(a)

Let γ be a functional that is continuous in at F. By the Glivenko–Cantelli theorem, F^nF0 almost surely, so |γ(Fn)γ(F)|0 almost surely, hence also in probability.

(b)

(i)

The mean functional is not continuous. Consider F=1[0,) and (1)Fn=(11n)1[0,)+1n1[n,). Then F has mean 0, Fn has mean 1, and FFn0.

(ii)

The Cramér–von Mises functional is continuous. For this, see that (2)(FF0)2(GF0)2=(F+G)(FG)2F0(FG)4FG0.

(iii)

The quantile function is also continuous, but only if we assume that F(x)>α for all x>Qα(F). Note that (3)x<Qα(F)F(x)<α,xQα(F)F(x)α. Let ε>0. Then (4)|xQα(F)|εxε<Qα(F)x+εF(xε)<αF(x+ε). Thus, we desire to show that (5)F(Qα(Fn)ε)<αF(Qα(Fn)+ε). Alternatively, we could show that (6)Fn(Qα(F)ε)<αFn(Qα(F)+ε). This turns out to be easier.

First, we will show that Fn(Qα(F)+ε)α. Note that (7)Fn(Qα(F)+ε)F(Qα(F)+ε)FFn. Since, by assumption, F(Qα(F)+ε)>α (the strict inequality is important here), we can make n large enough to find Fn(Qα(F)+ε)>α. The other inequality Fn(Qα(F)ε)<α follows similarly.

Published on 2 March 2021.