THÈSE Estimation, validation et identification des modèles ARMA ...
THÈSE Estimation, validation et identification des modèles ARMA ...
THÈSE Estimation, validation et identification des modèles ARMA ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Chapitre 3. Estimating the asymptotic variance of LSE of weak V<strong>ARMA</strong> models 100<br />
where Eij = ∂Bℓ ′/∂bij,ℓ ′ is the d×d matrix with 1 at position (i,j) and 0 elsewhere.<br />
We then have<br />
∞ ∂<strong>et</strong>(θ)<br />
= B ∗ ij,h<strong>et</strong>−ℓ ′ −h(θ). (3.10)<br />
∂bij,ℓ ′<br />
Similarly to Lemma 3.4, we have<br />
∂<strong>et</strong>(θ)<br />
∂b ′ ℓ ′<br />
h=0<br />
= [N11(L)<strong>et</strong>−ℓ ′(θ) : N21(L)<strong>et</strong>−ℓ ′(θ) : ... : Ndd(L)<strong>et</strong>−ℓ ′(θ)]<br />
<br />
d×d2 = N(L)(Id 2 ⊗<strong>et</strong>−ℓ ′(θ)) = N(L)<strong>et</strong>−ℓ ′(θ) = Ut−ℓ ′(θ),<br />
where N(L) = [N11(L) : N21(L) : ... : Ndd(L)] and Ut−ℓ ′ are respectively the d×d3 and<br />
d×d 2 matrices. Then, we have<br />
∂<strong>et</strong>(θ)<br />
∂b ′ ℓ ′<br />
=<br />
∞<br />
B ∗ h<strong>et</strong>−ℓ ′ −h(θ) =<br />
h=0<br />
∞<br />
B ∗ k−ℓ ′<strong>et</strong>−k(θ) = Ut−ℓ ′(θ),<br />
where B ∗ k−ℓ ′ = 0 when k < ℓ′ . With these notations, we obtain<br />
∂<strong>et</strong>(θ)<br />
∂b ′<br />
=<br />
=<br />
k=0<br />
∞ <br />
∗<br />
Bk−1<strong>et</strong>−k(θ) : B ∗ k−2<strong>et</strong>−k(θ) : ... : B ∗ <br />
k−q<strong>et</strong>−k(θ) <br />
d×d2q ∞<br />
B ∗ θ,k (Id2q ⊗<strong>et</strong>−k(θ)),<br />
k=0<br />
k=0<br />
whereB ∗ θ,k = B ∗ k−1 : B∗ k−2 : ... : B∗ k−q<br />
L<strong>et</strong><br />
is thed×d 3 q matrix. The conclusion follows. ✷<br />
Proof of Proposition 2 : L<strong>et</strong><br />
∂<br />
∂θ ′<strong>et</strong>(θ0) <br />
∂<br />
= <strong>et</strong>(θ0),...,<br />
∂θ1<br />
∂<br />
<br />
<strong>et</strong>(θ0)<br />
∂θk0<br />
˜ℓn(θ) = − 2<br />
n log˜ Ln(θ)<br />
= 1<br />
n <br />
dlog(2π)+logd<strong>et</strong>Σe + ˜e<br />
n<br />
′ t(θ)Σ −1<br />
e ˜<strong>et</strong>(θ) .<br />
t=1<br />
In Boubacar Mainassara and Francq (2009), it is shown that ℓn(θ) = ˜ ℓn(θ)+o(1) a.s.,<br />
where<br />
ℓn(θ) = 1<br />
n <br />
dlog(2π)+logd<strong>et</strong>Σe +e<br />
n<br />
′ t (θ)Σ−1 e <strong>et</strong>(θ) .<br />
t=1<br />
.