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 2. Estimating weak structural V<strong>ARMA</strong> models 73<br />
Proof of Lemma 3 : L<strong>et</strong> ϑ = (ϑ1,...,ϑk0) ′ . In view of (2.15), (2.16) and (2.17),<br />
for all i ∈ {1,...,k0}, we have<br />
<br />
∂lt(ϑ)<br />
= Tr Σ<br />
∂ϑi<br />
−1∂Σe<br />
e −Σ<br />
∂ϑi<br />
−1<br />
e <strong>et</strong>(ϑ)e ′ t (ϑ)Σ−1<br />
<br />
∂Σe<br />
e +2<br />
∂ϑi<br />
∂e′ t(ϑ)<br />
Σ<br />
∂ϑi<br />
−1<br />
e <strong>et</strong>(ϑ). (2.19)<br />
Using the previous relations and (2.18), for all i,j ∈ {1,...,k0}, we have<br />
∂ 2 lt(ϑ)<br />
∂ϑi∂ϑj<br />
= Tr<br />
<br />
Σ −1<br />
e<br />
∂ 2 Σe<br />
∂ϑi∂ϑj<br />
−Σ −1<br />
e<br />
∂Σe<br />
∂ϑi<br />
Σ −1∂Σe<br />
e −Σ<br />
∂ϑj<br />
−1<br />
+Σ −1∂Σe<br />
e Σ<br />
∂ϑi<br />
−1<br />
e <strong>et</strong>(ϑ)e ′ t (ϑ)Σ−1<br />
∂Σe<br />
e +Σ<br />
∂ϑj<br />
−1<br />
−Σ −1∂Σe<br />
e Σ<br />
∂ϑi<br />
−1∂<strong>et</strong>(ϑ)e<br />
e<br />
′ <br />
t(ϑ)<br />
+2<br />
∂ϑj<br />
∂2e ′ t(ϑ)<br />
∂ϑi∂ϑj<br />
+2 ∂e′ t(ϑ)<br />
Σ<br />
∂ϑi<br />
−1<br />
<br />
∂<strong>et</strong>(ϑ)<br />
e −2Tr Σ<br />
∂ϑj<br />
−1<br />
e <strong>et</strong>(ϑ) ∂e′ t(ϑ)<br />
∂ϑi<br />
e <strong>et</strong>(ϑ)e ′ t (ϑ)Σ−1 e<br />
e <strong>et</strong>(ϑ)e ′ t (ϑ)Σ−1 e<br />
Σ −1<br />
e <strong>et</strong>(ϑ)<br />
Σ −1<br />
e<br />
∂Σe<br />
∂ϑj<br />
∂Σe<br />
∂ϑi<br />
∂ 2 Σe<br />
∂ϑi∂ϑj<br />
Σ −1∂Σe<br />
e<br />
∂ϑj<br />
<br />
. (2.20)<br />
Using E<strong>et</strong>e ′ t = Σe, E<strong>et</strong> = 0, the uncorrelatedness b<strong>et</strong>ween <strong>et</strong> and the linear past Ht−1,<br />
∂<strong>et</strong>(ϑ0)/∂ϑi ∈ Ht−1, and ∂ 2 <strong>et</strong>(ϑ0)/∂ϑi∂ϑj ∈ Ht−1, we have<br />
E ∂2 lt(ϑ0)<br />
∂ϑi∂ϑj<br />
<br />
= Tr Σ −1∂Σe(ϑ0)<br />
e0 Σ<br />
∂ϑi<br />
−1<br />
<br />
∂Σe(ϑ0)<br />
e0 +2E<br />
∂ϑj<br />
∂e′ t (ϑ0)<br />
Σ<br />
∂ϑi<br />
−1∂<strong>et</strong>(ϑ0)<br />
e0<br />
∂ϑj<br />
= J(i,j). (2.21)<br />
The ergodic theorem and the next lemma conclude. ✷<br />
Lemma 4. Under the assumptions of Theorem 2.2, the matrix<br />
is invertible.<br />
and<br />
with<br />
J = E ∂2 lt(ϑ0)<br />
∂ϑ∂ϑ ′<br />
Proof of Lemma 4 : In view of (2.21), we have J = J1 +J2, where<br />
<br />
J1(i,j) = Tr<br />
Σ −1/2<br />
e0<br />
J2 = 2E ∂e′ t (ϑ0)<br />
∂ϑ Σ−1 e0<br />
∂Σe(ϑ0)<br />
Σ<br />
∂ϑi<br />
−1/2<br />
hi = (Σ −1/2<br />
e0 ⊗Σ −1/2<br />
e0 Σ −1/2<br />
e0<br />
∂<strong>et</strong>(ϑ0)<br />
∂ϑ ′<br />
∂Σe(ϑ0)<br />
∂ϑj<br />
e0 )di, di = vec ∂Σe(ϑ0)<br />
Σ −1/2<br />
<br />
e0 = h ′ ihj, In the previous derivations, we used the well-known relations Tr(A ′ B) = (vecA) ′ vecB<br />
and vec(ABC) = (C ′ ⊗A)vecB. Note that the matrices J, J1 and J2 are semi-definite<br />
∂ϑi<br />
.