Econometric Theory and Methods

Answers to Starred Exercises

32

Solution to Exercise 8.20

, can be written as 8.20 The IV variant of the HRGNR (6.90), evaluated at = = PUP
WX

1PW Xb + residuals, U

(8.90)

is an n n where is an n --vector of which every component equals 1, and U th th diagonal matrix with t diagonal element equal to the t element of the . vector y X Verify that this articial regression possesses all the requisite properties for hypothesis testing, namely, that = IV ; The regressand in (8.90) is orthogonal to the regressors when = IV is The estimated OLS covariance matrix from (8.90) evaluated at equal to n/(n k) times the HCCME Var h ( IV ) given by (8.65); The HRGNR (8.90) allows one-step estimation: The OLS parameter esti from (8.90) are such that IV = +b . mates b

= IV , the inner product of the The rst result is easily shown. When regressand of (8.90) with the matrix of regressors is 1PW X PUP WX U PW X (X PW U U PW X )1X PW U U 1PW X = U U PW X )1X PW X. PW X (X PW U =u is not explicitly transposed, because it is a diagonal matrix. In The matrix U IV . Since the moment conditions (8.28) = y X the last line of (S8.25), u PW X = 0, this last line is just a zero vector. Therefore, as imply that u required, we have shown that the regressand in (8.90) is orthogonal to the = IV . regressors when The second result is also easily shown. The OLS covariance matrix estimate = IV , is from (8.90), evaluated at n 1P 1PW X X PW U UPW X U nk
1

(S8.25)

The rst factor here is /(n k ), which is the OLS estimate of 2 from regression (8.90). Because the regressand is orthogonal to the regressors, the SSR is precisely . The second factor can be rewritten as 1 U PW X (X PW U U PW X )1X PW U U 1PW X X PW U U PW X )1X PW X = X PW X (X PW U IV ) given by (8.65). which is the HCCME Var h ( Copyright c 2003, Russell Davidson and James G. MacKinnon
1 1

W X (X PW X )1, = (X PW X )1X PW P

Econometric Theory and Methods

Answers to Starred Exercises

33

IV = +b . By the standard For the third result, we need to show that formula for the OLS estimator, = X PW U 1P 1PW X b UPW X U
1

1P X PW U UPW X .

We have just seen that the rst factor here can be rewritten as W X (X PW X )1. (X PW X )1X PW P The second factor is 1 U PW X (X PW U U PW X )1X PW U X PW U W X )1X PW u . = X PW X (X PW P (S8.27) (S8.26)

Postmultiplying (S8.26) by (S8.27), we observe that the last factor in (S8.26) is the inverse of the rst factor in (S8.27), so that = (X PW X )1X PW P W X (X PW P W X )1X PW u b . = (X PW X )1X PW u It is then easy to see that = (X PW X )1X PW (y X ) b = (X PW X )1X PW y (X PW X )1X PW X IV . = This establishes the third result

Copyright c 2003, Russell Davidson and James G. MacKinnon