TOULOUSE SCHOOL OF ECONOMICS

MASTER 1 - FALL 2014

ECONOMETRICS

TD4 Tests and Restricted LS

Exercise 4.1. Consider the standard linear regression model y = X1 1 +
X2 2 + .
0
1. Show that the constrained LS estimate of = ( 01 ; 02 ) subject to
the constraint that 1 = c (for some given vector c) is the OLS
regression of y X1 c on X2 .
2. Consider the case where X1 and X2 are univariate. Rewrite the
model so that the restriction 2 1 = 1 becomes a single zero
restriction (i.e., = 0).

Exercise 4.2. Consider the standard linear regression model y = X1 1 +

X2 2 + with E[ jX] = 0 and V ar( jX) = 2 I .

1. Let b 1 and b 2 be the corresponding subvectors of b . Recall the

formula of b 1 and give its conditional variance.
2. Let b R be the restricted estimator under the constraint 2 = 0.
What are the corresponding subvectors of b 1R and b 2R ? What are
their conditional variances?
3. Show that the matrix X10 X1 X10 MX2 X1 is semi-positive de nite.
Deduce from this result an ordering of the variance of b 1R and b 1 .
4. Does the previous result imply that the estimated standard errors of
the components of b 1R will always be smaller than the ones of b 1 ?

Exercise 4.3. Show that Cov( b ; b R jX) = Var( b R jX). What does this
imply for Var b b R jX ?

Exercise 4.4. Resolve the following paradox: The Gauss-markov theorem

states that b is the minimum-variance linear unbiased estimator, whereas
the restricted LS b R is clearly more e cient linear unbiased when the
restrictions hold.

1
Exercise 4.5. You have two independent samples y1 ; X1 and y2 ; X2 such
that
2
yj = Xj j + j E( j jXj ) = 0 V ar( j jXj ) = jI j = 1; 2

where both X1 and X2 have K columns. Let b j , j = 1; 2 be the OLS es-

timators. For simplicity, assume that both samples have the same number
of observations n.

1. Find the asymptotic distribution of

p
n b2 b1 ( 2 1) :

2. Find an appropriate test statistic for H0 : 1 = 2

3. Find the asymptotic distribution of this statistic under H0 .

4. What happens when 1 6= 2?

Exercise 4.6. Are rent rates in uenced by the student population in a col-
lege town? We have a random sample of US college towns for which we
observe the following variables. Let rent be the average monthly rent paid
on rental units in a college town in the United States. Let pop denote the
total city population, avginc the average city income, and pctstu the stu-
dent population as a percentage of the total population. One model to
test for a relationship is
log(rent) = 0 + 1 log(pop) + 2 log(avginc) + 3 pctstu + ":

1. State the null hypothesis that size of the student body relative to the
population has no ceteris paribus e ect on monthly rents. State the
alternative that there is an e ect.
2. What signs do you expect for 1 and 2?

3. The equation estimated using 1990 data for 64 college towns is

\ = :043 + :066 log(pop) + :507 log(avginc) + :0056 pctstu
log(rent)
(:844) (:039) (:081) (:0017)

n = 64; R2 = :458:
What is wrong with the statement: "A 10% increase in population
is associated with about a 6.6% increase in rent"?

2
 4. Test the hypothesis stated in the rst question at the 1% level.
5. Test the hypothesis that 1 0 against the alternative that 1 >0
at the 5% level.
6. Using the results of the regression in 3), compute a con dence interval
for 2 at the 95% level.

Exercise 4.7. We study the relation between the price of a house and its
characteristics. To this end, we consider the following model:

log(price) = 1 + 2 sqrf t + 3 bdrms + 4 sqrf t bdrms + 5 bthrms +"

where price is the selling price of the house, sqrf t is the square footage
of house, bdrms is the number of bedrooms and bthrms is the number of
bathrooms.

1. Derive the marginal impact of the number of bedrooms in the house.

The model has been estimated on a sample of 546 residential houses
during between July, August and September 2003 in the area of
Windsor, Canada.

Dependent variable: log(price) Estimated param. Std.Error

constant 7:92151 0:21912
sqrf t 0:00027 0:00008
bdrms 0:06491 0:03640
sqrf t bdrms 0:00056 0:00043
bthrms 0:09039 0:05221
n = 546, R2 = 0:261
2. Compute the marginal impact of the number of bedrooms in the house
for sqrf t = 100 and for sqrf t = 200. Interpret.
3. Does the impact of the number of bedrooms in the house depend on
the square footage of the house?

4. Does the hypothesis H0 : 3 = 5 has any economic meaning? Test

this assumption against H1 : 3 6= 5 at the 5% level, knowing that
the estimated covariance between c3 and c5 is 0.00004