Advanced Econometrics Takeshi Amemiya
Estimation of p
Because 2 defined in (5.2.9) depends on a2 only through a scalar multiplication, fio defined in (6.1.3) does not depend on a2. Therefore, in obtaining FGLS (6.2.1), we need to estimate only p. The most natural estimator of p is
2 Й/-ІЙ/
р=Ц—. (6.3.3)
Й?_,
t-г
where й, = у, — xlfi. The consistency of pis straightforward. We shall prove its asymptotic normality.
Using Uf = put- + e„ we have
Tt%™+6'
1 t ’
•* г-2
where
4, - 4= У О» - + (6-3.5)
and
*2 = J. i2 U - Д>'*г-.]2 + (fi - Д>'*-іЧг-. • (6-3.6)
If we assume that НтпГ Г_1Х'Х is a finite nonsingular matrix, it is easy to show that both Д, and Д2 converge to 0 in probability. For this, we need only the consistency of the LS estimator fi and — fi) = 0(1) but not the
asymptotic normality. Therefore, by repeated applications of Theorem 3.2.7 (Slutsky), we conclude that 4Т(р — p) has the same limit distribution as
|
T
|
Ik*-'
Hence the asymptotic normality of p follows from the result of Section 5.4.
