ECON 710 Midterm: ECON 710 UW Madison Midterm 2013a
![](https://new-preview-html.oneclass.com/rKZ3w4gpqWELjZVElweBj8nX2G5YyBMv/bg1.png)
Econometrics 710
Midterm Exam
March 12, 2013
Sample Answers
This exam concerns the model
yi=m(xi) + ei(1)
m(x) = 0+1x+2x2+ +pxp(2)
E(ziei) = 0 (3)
zi= (1; xi; :::; xp
i)0(4)
g(x) = d
dxm(x)(5)
with iid observations (yi; xi); i = 1; :::; n: The order of the polynomial pis known.
1. How should we interpret the function m(x)given the projection assumption (3)? How should we
interpret g(x)? (Brie‡y)
The model does not specify that m(x)is the conditional mean. Rather, equation (3) speci…es that
it is a projection model. Thus m(x)is the best linear predictor of yigiven linear functions of zi:
Equivalently, it is the best predictor in the class of pth order polynomials in xi:It is also the best
mean-square approximation to the conditional mean, in the class of pth order polynomials in xi:The
function g(x)is the derivative of the best linear predictor, and equals
g(x) = 1+ 22x+ 33x2+ +ppxp1
=h(x)0
where = (0; :::; p)0and h(x) = (0;1;2x; 3x2; :::; pxp1)0:
2. Describe an estimator ^g(x)of g(x):
Since g(x) = h(x)0is linear in ; the plug-in approach suggests replacing with the e¢cient estimator
for : Under the projection assumption (3) OLS is the asymptotically e¢cient estimator. It equals
^
= (Z0Z)1(Z0Y)where Yand Zare the stacked observations on yiand zi:Then the estimator for
g(x)is
^g(x) = h(x)0^
=^
1+ 2^
2x+ 3^
3x2+ +p^
pxp1
3. Find the asymptotic distribution of pn(^g(x)g(x)) as n! 1:
Under the projection assumption (3) plus regularity conditions, we know that as n! 1;pn^
!d
N(0; V)where Vb=Q1Q1with Q=E(ziz0
i)and = Eziz0
ie2
i. Then as n! 1
pn(^g(x)g(x)) = pnh(x)0^
h(x)0
=h(x)0pn^
!dh(x)0N(0; V) = N(0; h(x)0Vh(x)):
Document Summary
This exam concerns the model yi = m(xi) + ei m(x) = (cid:12)0 + (cid:12)1x + (cid:12)2x2 + (cid:1)(cid:1)(cid:1) + (cid:12)pxp. The model does not specify that m(x) is the conditional mean. Rather, equation (3) speci(cid:133)es that it is a projection model. Thus m(x) is the best linear predictor of yi given linear functions of zi: = h(x)0(cid:12) where (cid:12) = ((cid:12)0; :::; (cid:12)p)0 and h(x) = (0; 1; 2x; 3x2; :::; pxp(cid:0)1)0: describe an estimator ^g(x) of g(x): Since g(x) = h(x)0(cid:12) is linear in (cid:12); the plug-in approach suggests replacing (cid:12) with the e cient estimator for (cid:12): under the projection assumption (3) ols is the asymptotically e cient estimator. ^(cid:12) = (z 0z)(cid:0)1(z 0y ) where y and z are the stacked observations on yi and zi: then the estimator for g(x) is. = ^(cid:12)1 + 2^(cid:12)2x + 3^(cid:12)3x2 + (cid:1)(cid:1)(cid:1) + p^(cid:12)pxp(cid:0)1: find the asymptotic distribution of pn (^g(x) (cid:0) g(x)) as n !