Вы находитесь на странице: 1из 6

Final Exam

E603: Advanced Econometrics


Prof. Dr. Markus Frlich

December 11, 2014

This exam consists of six (6) pages. Please check whether you have received all pages.
This exam consists of ve (5) questions. Please read all questions carefully and solve
all of them.
Please write your name on each sheet of paper and discuss each of the ve questions on

a separate page.
Argue in a clear and logically consistent manner. Do not skip over important assump-

tions and steps in your argumentation.


Manage your time wisely and try to be brief. Short and precise answers will receive higher

credit than lengthy essay-type explanations. Additional scrap paper will be provided if
needed.
Number your answers clearly (part a, b.i, b.ii, ...)

Clear handwriting will be greatly appreciated.

This exam is closed book.


Calculators are not allowed in this exam.
You have 120 minutes to answer the questions.
For solving the questions in this exam, assume regressors to be stochastic.

Good Luck!

1
Question 1 [22 points]

Consider the following IV regression model:

Yi = Xi + Ui ,

E(Zi Ui ) = c,

where is the unknown scalar parameter, Yi the dependent variable, Xi the single endogenous
regressor, and Zi the single instrumental variable. c is some constant (possibly dierent from
zero). Assume that c is known.
a Show that is identied as
E(Zi Yi ) c
= .
E(Zi Xi )

What do you have to assume about the relationship between the endogenous regressor
and the instrument for this identication result to be valid? [4 points]
b Consider the following estimator of :
Pn
i=1 Zi Yi nc
n,c = P n .
i=1 Zi Xi

Assuming that data are i.i.d. and the assumption imposed in part (a) holds, show
whether n,c is consistent. [8 points]
c Establish the asymptotic normality of n,c and state the (additional) assumptions about
Zi and Ui that you have to make to do so.

Derive the asymptotic variance of n,c . [10 points]

2
Question 2 [20 points]

Suppose that there are M regression equations ymi = x0mi m + mi , m = 1, . . . , M. Here m


represents the equation number and i = 1, . . . , N is the observation index. Each equation m
has a single response variable ymi and a km -dimensional vector of regressors xmi . If we stack
observations corresponding to the m-th equation into N -dimensional vectors and matrices,
then the model can be written in vector form as ym = Xm m + m , m = 1, . . . , M , where ym
and m are N 1 vectors, Xm is a N km matrix, and m is a km 1 vector. Finally, if we
stack these M vector equations on top of each other, the SUR system will take the form:


y X 0 ... 0 I 12 IN ... 1M IN
1 1 1 1 11 N

y2 0 X2 . . . 0 2 2
12 IN 22 IN . . . 2M IN
. = . .. . . .. . + . , var(|X) = = . .. .. .

. . . . .. .. . ...
. . . . . .



yM 0 0 . . . XM M M 1M IN 2M IN . . . M M IN

a True or False? Show analytically whether it is correct that if the explanatory variables
in the m = 1, . . . , M equations are identical, then the least squares residuals from the m
equations are conditionally uncorrelated with each other. For simplicity consider only
the case M = 2. [10 points]
b True or False? Show analytically whether it is correct that if the explanatory variables
in the m = 1, . . . , M equations are orthogonal to each other, then the least squares
coecient estimates for the m equations are conditionally uncorrelated with each other.
For simplicity consider only the case M = 2. [8 points]
c Generally, under which conditions is GLS estimation more ecient than OLS estimation?
[2 points]

3
Question 3 [20 points]

The following tables present results for an annual data set for 162 farmers over the years 1993
to 1998. The variables are MILK (milk output in liters per year), COWS (number of cows),
LAND (land area, constant for each farm), and FEED (feed input).
Table 1 presents the output of a pooled regression of log(MILK) on an intercept (C), log(COWS),
log(LAND) and log(FEED). Table 2 presents the output from a xed eects regression where
the estimated xed eects are not reported, and table 3 presents results from a random eects
estimation.
Table 1: Pooled OLS estimation
Dep. Variable: LOG(MILK)

variable coecient std. error t-stat prob.


C 6.976457 0.040584 171.9009 0.0000
LOG(COWS) 0.600228 0.023564 25.49150 0.0000
LOG(LAND) 0.020668 0.014120 1.463763 0.1436
LOG(FEED) 0.455605 0.013712 33.22704 0.0000

Table 2: Fixed Eects estimation


Dep. Variable: LOG(MILK)

variable coecient std. error t-stat prob.


LOG(COWS) 0.674705 0.032031 21.06433 0.0000
LOG(FEED) 0.396393 0.014944 26.52504 0.0000
xed eects not reported

(Question 3 continues on page 5)

4
Table 3: Random Eects estimation
Dep. Variable: LOG(MILK)

variable coecient std. error t-stat prob.


C 7.086916 0.057206 123.8852 0.0000
LOG(COWS) 0.657466 0.027075 24.28350 0.0000
LOG(LAND) 0.020818 0.023718 0.877749 0.3830
LOG(FEED) 0.410013 0.013630 30.08170 0.0000

a Interpret all estimated regression coecients reported in Table 1. [4 points]


b Why is the variable log(LAND) included in Tables 1 and 3, but not in Table 2? [4
points]
c Test the hypothesis that log(LAND) has no signicant eect on milk production. De-
scribe which model(s) you use, which test statistic you use and its null hypothesis.
Explain how you use the displayed output to come to your conclusion. [4 points]
d i Compare the models in Table 1 and 2. What is the dierence between them? Which
model is more restrictive? Explain how you would test for these restrictions. [4
points]
ii Compare the models in Table 2 and 3. What is the dierence between them? Which
model is more restrictive? Explain how you would test for these restrictions. [4
points]

Question 4 [8 points]

Consider two linear regressions:


Y = X + U,

Y = X + Z + U.

In nite samples, under what conditions are the OLS estimates of from the two regressions
identical? Carefully justify your answer.

5
Question 5 [10 points]

Observations in sample A are assumed to have been drawn from a distribution DA with mean
equal to . Observations in sample B are assumed to have been drawn from a distribution DB
with mean equal to 4. All observations are uncorrelated. We are interested in estimating .
The sample means are yA = 2 and yB = 6. The variances of these sample means are estimated
as Vd
ar(yA ) = 4 and Vd
ar(yB ) = 2.

a State the quadratic objective function Qn that is a function of and that is minimized
at the asymptotically ecient GMM estimate of .
Is the estimation problem under-identied, exactly identied or over-identied? What
is the degree of over-identication or under-identication, if applicable? [6 points]
b Use Qn from part a to compute the numerical value of the GMM estimate of . [3
points]
c Generally, under which condition is a GMM estimate consistent? [1 point]

Вам также может понравиться