MPRA
n?u=RePEc:pra:mprapa:75582&r=ets
n?u=RePEc:pra:mprapa:75582&r=ets
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
To ensure the positivity of the volatilities, we assume that ω kk > 0. In view of (3), the<br />
process (η ∗ t ) can be called the vector of equation by equation (EbE) innovations of (ε t ).<br />
Let a 0 k = (a0 k1 , ..., a0 km )′ and c 0 k = (c0 k1 , . . . , c0 kr )′ be, respectively, the k-th row vectors of<br />
the matrices A 0 and C 0 . Then the vector of unknown parameters involved in the k-th<br />
equation (4) can be denoted by θ (k)<br />
0 = (ω 0 kk , a0′ k , (b0 k )2 , c 0′<br />
k )′ ∈ R d , d = m + r + 2. It is clear<br />
that an identiability condition must be required such that σkt 2 is invariant to a change of<br />
sign of the vectors a 0 k , c0 k and b0 k . Without lost of generality, we can assume that a0 k1 > 0,<br />
b 0 k > 0 and c0 k1 > 0, for k = 1, . . . , m.<br />
Let θ (k) = (ω kk , a ′ k , b2 k , c′ k )′ = (ω kk , a k1 , ..., a km , b k , c k1 , . . . , c kr ) ′ be a generic parameter<br />
vector of the parameter space Θ (k) which is an any compact subset of<br />
(0, +∞) 2 × R m−1 × [0, 1) × (0, +∞) × R r−1 .<br />
2.2 Equation-by-equation estimation of parameters<br />
Let ε 1 , . . . , ε n be observations of a process satifying the semi-diagonal BEKK-X representation<br />
(2) and x 1 , . . . , x n be observations of a process of the explanatory variables.<br />
For all θ (k) ∈ Θ (k) , we recursively dene ˜σ kt 2 (θ(k) ) for t = 1, . . . , n by<br />
( m<br />
) 2 ) 2<br />
∑<br />
˜σ kt(θ 2 (k) ) = ω kk + a kl ε l,t−1 + b 2 k˜σ k,t−1(θ 2 (k) ) + c ks x s,t−1 (5)<br />
l=1<br />
( r∑<br />
s=1<br />
with the arbitrary initial values ˜ε 0 , ˜σ 0 and x 0 . Let<br />
˜Q (k)<br />
n (θ (k) ) = 1 n∑<br />
˜l kt (θ (k) ), ˜lkt (θ (k) ) = log ˜σ 2<br />
n<br />
kt(θ (k) ) +<br />
t=1<br />
The EbE estimator, denoted by<br />
measurable solution of the following equation<br />
Let θ 0 =<br />
̂θ (k)<br />
n<br />
ε 2 kt<br />
˜σ 2 kt (θ(k) ) .<br />
(k) ˆθ n , of the true parameter vector θ (k)<br />
0 is dened as a<br />
= arg min<br />
θ (k) ∈Θ (k)<br />
˜Q(k) n (θ (k) ). (6)<br />
) ′. (θ (1)′<br />
0 , . . . , θ (m)′<br />
0 Note that θ0 includes the diagonal elements of Ω 0 and<br />
all components of the matrices A 0 , B 0 and C 0 . This parameter vector belongs to the<br />
parameter space Θ m = Θ (1) × · · · × Θ (m) , whose generic element is denoted by θ =<br />
(<br />
θ (1)′ , . . . , θ (m)′) ′<br />
. The estimator of θ0 is given by ̂θ<br />
) (̂θ(1)<br />
n =<br />
′<br />
′<br />
, . . . , which is the<br />
collection of the equation by equation estimators.<br />
6<br />
n<br />
̂θ<br />
(m)′<br />
n