1 I<’n-1>rj (∑. =:; 1 {⅛∕n-,Y,∕n∣<M" U/n (σ2(ɪ.) - σ2(Xi/n)) d«
∑n-1
j = 1 1 {∣ Xj∙∕n~X≈∕n∣<ξn}
~^∕""~
Dn,r
Now, given Lemma 1, Dn,r = oa.s. (1), provided that n1 /2+εξn → 0, as n → ∞. It is immediate
to see that Hn,r is of a smaller order of probability than Gn,r.
Let < Gn >r denote the discretized quadratic variation process of Fn,r. By a similar argument
as in Bandi and Phillips (2003, pp.271-272),
we have that
plim n→∞
< Gn >r
- 2 ∞ σ4 ( a )
-∞
L X ( r, a )2
Lχ (1, a)
Thus, by the same argument as in the proof of Theorem 3 in Bandi and Phillips (2003), the
statement in Step 1 follows.
Proof of Step 2: It follows from Theorem 1 in Barndorff-Nielsen and Shephard (2004a).
Proof of Step 3: The discretized covariation process < An,Bn >r,
< An, Bn >
L(n-1’rj [(n-1’rj
i =1 j =1
'1 { ∣ .χiin-X∕,∣ <<n} J/N /n ( ' - χ∕n) σ(χ. )dW 2 '
∑n=1 1 { ∣ χ3∕n-χi∕n ∣ <ξn}
L(n-1’rj L(n-1’rj
= 2 n ∑ ∑
i =1 j =1
1 { ∣ χj∕n-χi∕n ∣ <ξn}σ 4( Xj/n + 0a-s-(1))
Vn-1 m, , ,
2=1=1 1 {∣ Xj∕n-Xi∕n ∣ <ξn}
(19)
= 2
1 {∖Xu-Xa∖<ξn}σ 4( XU )d U
Jθ1 1 {∖Xu-Xa∖<ξn} d U
d a + 0a.s. (1)
[∞ J ∣,∞ 1 u a <ξ. σ4l∏lχ(∣: l∏<∖u
J-oo -J-OO J-∞ 1 {∖u-a∖<ξn}LX(1 ,u)du
Lx(r, a)da + oa.s. (1),
where the 2 (instead of 4) on right hand side of (19) comes from Lemma 5.3 in Jacod and
Protter (1998). Along the lines of Bandi and Phillips (2001, 2003), by the change of variable
u-a
ξn
= z,
< An, Bn >
21