UNITS 1-2-3: MULTIPLE LINEAR REGRESSION MODEL / INFERENCE

1. Discuss if the following statements are true or false and reason your answers:

a) Considering that the disturbances of the MLR model are Normal, the OLS
estimators for the model parameters follows a (
N 0,s 2 ( X ' X )
-1
) distribution.
b) We require the explanatory variables (X) of the MLR model to be uncorrelated,
otherwise there would be autocorrelation in the disturbances.

2. Consider the following model Yt = b 0 + b1 x1t + b 2 x2t + ut where E (ut ) = 0;

E (ut2 ) = s 2 ; E (ut ut -k ) = r ks 2 , could we confirm that the OLS estimators are
unbiased, efficient and consistent?

3. The model Y = X b + u has a matrix X with n rows and k columns. The hypothesis
Rank(X) = k is established.

a) What econometric problem appears when Rank (X) = k–1?

b) What consequences does it have for the OLS estimators of b?
Reason your answers.

4. We want to explain the number of students enrolled in the Economics Degree (At) every
year with respect to a Spanish economic index, the per capita gross domestic product
(GDPt) and a variable related to the number of offered places by other universities with
regards to the population that can access to a university degree.

a) Propose a linear model to estimate

(At) taking into consideration that we
want to obtain the elasticities of the
enrolled students with regards to all
the explanatory variables.

b) What econometric resource would

you consider using in order to study
the accuracy of the model?
c) How would you test if the elasticity
of the GDP with respect to the
number of enrolments is unitary?

5. We know that a relevant variable (X3t) has been omitted from the next model
Yt = b 0 + b1 x1t + b 2 x2t + ut In addition, the omitted variable
fulfills X 3t = 0.6 X 3t -1 and Cov ( x1t , x3t ) = Cov ( x2 t , x3t ) = 0

a) Is the expected value of ut

equal to zero? Why?

1
 !
b) Are the OLS estimators of b
unbiased? Why?
c) Are the disturbances ut
autocorrelated? Why?

6. An analyst has estimated the following equation for the demand of a product with
respect to its price, the price of another product and the per capita income (all variables
are in logarithms and the standard deviations are between parentheses).

LQAt = 3.56- 0.84 LPAt + 0.36 LYt + 0.25 LY + ut R 2 = 0.75

(2.85) (0.051) (0.022) (0.017)

a) What do the estimated

coefficients represent? How is
the product A with regards to
the product B?
b) Taking into account R2, is the
model explaining properly QAt?
Reason your answer.
c) Perform the individual and joint
significance tests for all the
variables.
d) If it is known that s u = 1 , what
2

would be the prediction for the

demand of the product A if its
price is 6 mu, the income is 6
mu and the price of the other
product is 2 mu?
e) Considering a 95% confidence
level, what would be the
confidence interval for the
section D) prediction?
Useful information: critical values for the F distribution (5%):
F(1,118)=3.92; F(2,118)=3.07; F(3,118)=2.68; F(4,118)=2.45

7. A researcher has estimated the next model:

LYt = 2.5 + 0.42 LX 2t +0.46 LZ 3t + ut R 2 = 0.86; SR = 0.14; N = 105
æ 1.21 -0.26 -0.02 ö
ç ÷
( X `X )
-1
= ç -0.26 0.92 0.14 ÷
ç -0.02 0.14 1.03 ÷
è ø

a) Perform the individual and joint significance tests using a 5% significance level.
b) Test if the sum of elasticities is unitary.
c) What is the prediction and the confidence interval for Yt if X2t is equal to 1 and Z3t
is equal to 2? Consider a 95% confidence level.

2
8. Consider the model Yt = b 0 + b1 x1t + b 2 x2t + ut and the following estimations:

Yt = 655.8- 3.359 X 2t + 0.0556 X 3t + ut R 2 = 0.969; SR = 53938.8; N = 14

(57.57) 0.784 (0.178)

a) What interpretation can you

make of R2?

b) Is the variable X2t relevant to

explain Yt? And X3t? Analise it.
c) Perform the joint significance
test for the variables of the
model.

F(2,11)=3.98; F(3,11)=3.58
d) What is the prediction and the
confidence interval for Yt if X2t
is equal to 0.5 and X3t is equal to
0.7?
Consider a 96% confidence
level.

9. You’ve been hired to study the profits of the investments made in two different assets
during the last 20 years where it is needed to consider the taxing effect on the obtained
profits.

a) Design a regression model to

explain the profits using
elasticities.
b) Make an economic interpretation
of the coefficients and the sign
they should have.
c) If you want to test if the profits
vary in the same proportion as
the investment made in the first
asset, what would you propose?
And if you suspect that the
investment made in the second
asset isn’t yielding any profit,
how would you confirm if this is
true or not?
d) If you think that the relations
between the variables could have
changed in 2007, how would
you test if it actually happened?

3
 10. A Spanish company that only exports to the USA wants to analyse the sensitivity of its
exports (Xt) through time with regards to the price of the product sold (Pt) , the USA per
capita income (Yt) and the €/$ exchange rate (tc). After using a regression model, the
following results have been obtained:

Method of estimation: Ordinary least R^2: 0.97

squares (OLS=MCO) DW: 1.97
no. of observations 132 Ljung-Box Q-statistic (20): 182 [0.0012]
Dependent variable: LXt Hetero test: F(5,128) = 0.65 [0.99551]
Current simple: en00 to dic10
LX t = 0.1- 0.5 LPt + 0.34 LYt + 1.28L tct + ut
(0.03) (0.026) (0.031)
a) If the exchange rate shifts 1%, the
exports would be increased in the
same proportion? Reason your
answer using the corresponding
test.
b) Could you confirm that the
exports are sensible to the per
capita income variations? Support
your answer using an appropriate
test.

Perform a joint significance test

and make and economic
interpretation.
c) Are all the MLR model
hypotheses fulfilled. In case
hypotheses are not met, explain
the consequences it may have
over the model.

11. Reason if the following pair of assessments are coherent:

!
a) if it is different form the identity matrix, it
b MCO
OLS son eficientes
is not homocedastic model and there is
Y = XB + U autocorrelation between the residuals are not efficient
con E (UU ') = s 2 S siendo S ¹ I TRUE, it is not homocedastic
this is the residual variance

b) !
b MCO no son óptimos
Y = XB + U are not BLUE
E (U ) ¹ 0 TRUE, because it implies that B^OLS arent
unbiassed. So if it is not unbiassed, it cannot
be BLUE (best linear unbiased estimators)

c)
Yt = b1 + b 2 x2t + b3 x3t + ut The multicollinearity hypothesis
siendo x2t = 5 "t is not fulfilled
TRUE, the hypothesis 7 says there is no
multiollinearity and as here says it is not fullfilled.
to say there is multicollinarity, tiene que haber que por ejemplo, x2t= 5 + x2. Tiene que haber la x para que sea
multicollineraty. en este caso, como es 5 y es solo una constante, no depende de la otra por lo tanto, no hay multicollinearity.

Multicollinarity es cuando una variable depende de la otra. En este caso, es constante (5) por lo que no hay.
4
 12. Consider the model LYi = b1 + b 2 Lx2i + b3 x3i + ui , and the following information:

æ 1.191 -0.172 0.052 ö æ19 132 20 ö æ 16 ö

ç ÷ ç ÷ ç ÷
(X 'X ) = ç -0.172 0.028 -0.021÷ ; ( X ' X ) = ç 961 150 ÷ ; ( X ' Y ) = ç114 ÷ ; åY
-1
i
2
= 16
ç 0.052 -0.021 0.082 ÷ ç 36 ÷ø ç 22 ÷
è ø è è ø

RSS=0.192; TSS=15.29

a) Estimate the parameters of the model, the residual variance and the variance/covariance
matrix for the estimated parameters.

æ ö
æ ö ç ÷
! ç ÷; !
b MCO =ç ÷
!
s u!2 = (
; MVC b MCO ) =ç
ç
÷
÷
ç ÷ ç ÷
è ø
è ø

b) Make an interpretation on b 2 and b3 .

c) Are the variables of the model jointly influential? Reason your

answer using a test.

13. Design a model to explain the sensitivity of the Spanish exports (It) between 1980-2013
with regards to the GDP, the exchange rate (€/rest of the currencies) and relative prices
(foreign prices/national prices).

a) Detail your estimated model.

b) How would you test if a decrease in exports varies in the same proportion as the relative
prices do?

c) What p-value should have the tests d) In 1986 Spain joined the EU and in 2002
you already know to make sure that Spain entered into the euro currency system.
the OLS estimators are optimal? How would you test if those facts had a
WHAT HAPPENS WHEN P-VALUE IS structural impact in the variables?
SMALLER THAN THE SIGNIFICANCE
LEVEL?

5
 14. In order to explain the price of the apartments (Pi) in thousands of euros, a model has
been designed where the number of bathrooms (Bi) and the number of bedrooms (Di)
have been included as variables:
!
LPi = 7.1+ 0.215 Bi + 0.078 Di + ut
(0.023) (0.015)

If an apartment has one more bedroom, would its price increase in 7,800 € ceteris
paribus?

15. A Spanish telephone wire manufacturer has used the following model to forecast its
sales during 1998-2013. The considered variables are.
VENT_CABLE: Yearly sales (million)
PIB: Gross domestics product in US dollars (thousand million)
CONEX: Home connections made (thousand)
T_DESEM: Unemployment rate (%)
EQ(1) Modelling VENT_CABLE by OLS.
The dataset is: C:CABLE.XLS. The estimation sample is:1998 to 2013

Coefficient Std.Error t-value t-prob Part.R^2

Constant 195.604 2308. 0.0847 0.934 0.0006
PIB 6.2083 1.906 3.26 0.007 0.4693
CONEX 1.4954 0.6254 2.39 0.034 0.3227
T_DESEM -469.718 198.6 -2.37 0.036 0.3180

sigma 859.306 RSS 8860877.2

R^2 0.601254 F(3,12) = 6.031 [0.010]**
log-likelihood -128.5 DW 1.58
no. of observations 16 no. of parameters 4

when the log-likelihood constant is not included:

AIC 13.7246 SC 13.9177
HQ 13.7345 FPE 923008

when the log-likelihood constant is included:

AIC 16.5624 SC 16.7556
HQ 16.5723 FPE 1.38887e+007

AR 1-2 test: F(2,10) = 1.4476 [0.2804]

ARCH 1-1 test: F(1,10) = 0.7343 [0.4088]
Normality test: Chi^2(2) = 1.3303 [0.5142]
Hetero test: F(6,5) = 0.5446 [0.7598]
Hetero-X test: not enough observations
RESET test: F(1,11) =0.046170 [0.8338]

Additional results:

Test for linear restrictions (RB=r):

R matrix
Constant PIB CONEX T_DESEM
0.00000 1.0000 0.00000 0.000000
r vector
0.00000

LinRes F(1,12) = 10.612[0.0069]** 0,0069 < 0,05 —> reject H0

Test for linear restrictions (RB=r):

6
R matrix
Constant PIB CONEX T_DESEM
0.00000 1.0000 0.00000 0.000000
.00000 0.0000 1.00000 0.000000
r vector
0.00000 0.00000

LinRes F(2,12) = 8.0026[0.0062]** 0,0062 < 0,05 —> reject H0

a) Write the model that the manufacturer has specified:

Vent_Cable^= 195,604+6,2083PIB+1,4954CONEX-469,718T_DESEM

b) If the number of connections increases in 1,000 units (ceteris paribus), what would be
the expected increase in sales?

c) Are the individual significance “t” tests valid? Why?

d) Detail the linear restrictions test using matrices? What are the conclusions of those
tests?

e) Knowing that the omitted test uses the following hypotheses:

Vent _ cable = b1 + b 2 PIBt + b3CONEX t + b 4T _ Desemt + b 5GANANC _ LINEAt + ut

H 0 : b5 = 0

The following result was obtained:

Omitted variable test: F(1,11)=13.7294[0.0035]**

Added variables:
[0]: GANANC_LINEA

Being GANANC_LINEA: profit pr line for the client in %

Taking into consideration the information that this test has provided, what properties
would the OLS estimator have in the estimated model? Would you consider that it is
necessary to make any change in the model?

7
 TEST

1. To correctly estimate a MLR model with k–1 variables, we need that:

a. The rank of the X matrix is equal to k

b. The rank of the X matrix is equal to k–1
c. The rank of the X matrix is equal or greater than k–1
d. The rank of the X matrix is equal or smaller than k–1

2. Consider the model Ci = b1 + b 2Yi + b 3TI i + ui where Ci is the individual

consumption, Yi is the gross income and TIi are indirect taxes. If the hypothesis
H 0 : b 2 = - b3 is accepted, it means that:
a. There is exact multicollinearity.
b. The impact of the gross income on the individual consumption is the same that
the impact of the available income (Yai = Yi – TIi) on the individual
consumption.
c. The correct model is: Ci = b1 + b 2Yi + ui
d. Both variables (Y and TI) are jointly significative.

3. The relation between the explained variable and the explanatory variables in the MLR
is stochastic because:

a. The explanatory variables are stochastic because one of the hypotheses of the
MLR states it.
b. The explanatory variables are measured without any error.
c. There is a disturbance in the model.
d. It is sure that in the model there aren’t other factors that determine the
behaviour of the explained variable.

4. Given the model Yt = b1 + b 2 x2t + b3 x3t + b 4 x4t + ut , an analyst decides to

redesign it as:
Yt = b1 + b 2 x2t + b 3 x3t + ut
a. The OLS estimators are always unbiased and consistent.
b. The OLS estimators are unbiased and efficient.
c. There is an evidence of multicollinearity.
d. The OLS estimators could be biased.

5. If the OLS estimators are unbiased:

a. The disturbance (Ut) must be heteroscedastic and uncorrelated.

b. The explanatory variables (X) must be uncorrelated.
c. The mean of U must be equal to zero "t
d. Ut follows a normal distribution "t