A Supplement to "QML Estimation of Dynamic Panel Data Models with Spatial Errors"

Liangjun Su,Zhenlin Yang

2002-01-01

Abstract:Proof of Lemma B.2. (1) Recall that B = Bn = In − λWn, A = (B′B)−1(W ′B + B′W )(B′B)−1 , and Ȧ = 2(B′B)−1[(W ′B+B′W )A −W ′W ]. Noting that both W and B−1 are all uniformly bounded in both row and column sums under our assumptions, by Lemma B.1(1) B′B, (B′B)−1, A, and Ȧ are uniformly bounded in both row and column sums. Observing that JT ⊗ In has T ones and (n− 1)T zeros on each row and each column and thus uniformly bounded in both row and column sums when T is fixed and that IT ⊗ (B′B)−1 is a block diagonal matrix uniformly bounded in both row and column sums, by (3.2) Ω = φμ(JT ⊗ In) + IT ⊗ (B′B)−1 is uniformly bounded in both row and column sums. Analogously, we can show that Ω∗ and Ω† are uniformly bounded in both row and column sums. (2) As we argue in the proof of Theorem 4.1, Ω = Ω (δ) has minimum eigenvalue bounded away from zero and maximum eigenvalue bounded away from infinity uniformly in δ ∈ Δ. With this, we can readily show that the eigenvalues of Ω−1,Ω−1(IT ⊗ A)Ω−1,and Ω−1(JT ⊗ In)Ω−1 are also bounded away from zero and infinity and all these matrices have trace of order O (n). By the Cauchy-Schwarz inequality, 1 n tr(D1ΩD2) ≤ { 1 n tr(D1ΩD′ 1)}{ 1 n tr(D′ 2ΩD2)}. For each D = Ω−1,Ω−1(IT ⊗A)Ω−1, or Ω−1(JT ⊗ In)Ω−1, we have 1 ntr(DΩD) ≤ λmax(Ω) 1 n tr(DD) ≤ λmax(Ω)λmax (D) 1 n tr(D) = O(1) as D is symmetric here. In case D = Ω−1(IT ⊗ Ȧ), we have