= -J2(Qθ -CT)
1 ʃ. ʃ. 1 (z -z A
=--1— ∑ ∑ ɪ κ ∖z1-^ I
T(T -1) ⅛ ⅛ hm L h )
× {1( yt ≤ Qθ ( xt ) - Ct ) - Fyz ( Qθ ( xt ) - Cτ∣zt )}{Fyz ( Qθ ( xs ) - C ∖z, ) - θ} (A.22)
Denote S(g) ≡ ∂F[g] / ∂g . By taking a Taylor expansion of Fy∣z (Qθ (xs ) - Cτ ∣ zs) around
Qθ(xs), we have
J2(Qθ)-J2(Qθ-Cτ)
1
τ(τ- 1)
ττ1(z
∑ ∑ _Lк∖zt
m
t=1 s≠th L
-zI
τs- I {1(yt ≤ Qθ(xt) - Ct) - Fyz(Qθ(xt) - Ct | zt)}
h)
× (-Ct )S(Qθ (Xs ))
1τ
= Ct -∑ {1(yt ≤ Qθ(xt) - Ct ) -Fyz(Qθ(xt) - Ct)}S(Qθ(Xs)) f (Zt)
τ t=1
1τ
≡ Ct V∑ UtS(Qθ(x,))f,(zà. (A.23)
τ t=1
where Qθ is between Qθ and Qθ - Cτ . Thus we have
EJ2(Qθ ) - J2(Qθ - Ct )∣
1τ
≤Λ Ct - ∑ E∖ufz (z■ )
τ t=1
1τ
≤ΛCt -∑ E {ut f z2(Zt)}
τ t=1
= O(Cτ(τhm)-1). (A.24)
where the first inequality holds due to Assumption (1)(v) and the last equality is derived by
using Lemma C.3(iii) of Li (1999) that is proved in the proof of Lemma A.4(i) of Fan and Li
(1996c).
Thus. we have
16