ISET Econometrics II

Tbilisi State University Lecturer: Ketevani Kapanadze

MIDTERM EXAM
Version B

May 4

Instructions:
Solve all the following questions.
Part I: 15 Multiple choice questions (1pts each) and 5 True/False statements (1 pts each) –
(20 points)
Part II: Open question (5 points)
Part III: Open question (15 points)
Total exam points 40.
+Bonus (2 points)
In the open questions show your work clearly. You can round your final answers to the 4th
decimal place (or leave them unrounded).
All statistical tests are at the 5% significance level. Use critical t-value of 1.96.

Remember, only full answers will grant the full points.

Good Luck 
PART I (20 points)

Multiple Choices, choose one correct answer (a,b,c,d):

1) Which of the following may be consequences of one or more of the CLRM assumptions
being violated?

i) The coefficient estimates are not optimal

ii) The standard error estimates are not optimal

iii) The distributions assumed for the test statistics are inappropriate

iv) Conclusions regarding the strength of relationships between the dependent and
independent variables may be invalid.

a) (ii) and (iv) only

b) (i) and (iii) only

c) (i), (ii), and (iii) only

d) (i), (ii), (iii), and (iv)

2) Which of the following are feasible approaches to dealing with a model that exhibits
heteroscedasticity?

i) Take logarithms of each of the variables

ii) Use suitably modified standard errors
iii) Use a generalised least squares procedure
iv) Add lagged values of the variables to the regression equation.

a) (ii) and (iv) only

b) (i) and (iii) only

c) (i), (ii), and (iii) only

d) (i), (ii), (iii), and (iv)

 3) If OLS is used in the presence of autocorrelation, which of the following will be likely
consequences?
i) Coefficient estimates may be misleading
ii) Hypothesis tests could reach the wrong conclusions
iii) Forecasts made from the model could be biased
iv) Standard errors may inappropriate
a) (ii) and (iv) only
b) (i) and (iii) only
c) (i), (ii), and (iii) only
d) (i), (ii), (iii), and (iv)

4) An econometrician ran the following regression of average annual spending in a

household (S) (measured in thousands of dollars) on the number of people in the family (N)
for a random sample of 50 families:

^S = 16.7 + 0.0023N , R2=0.89

Suspecting heteroscedasticity in the data, he then runs another regression as such:

^
S/ N = 0.015 + 27.5(1/N) , R2=0.93

Which of the following is true about the first model?

a)The OLS estimators are biased in the first model due to heteroscedasticity.
b) The OLS estimators are efficient in the first model despite heteroscedasticity.
c)The OLS estimators are efficient and unbiased due to heteroscedasticity.
d) The OLS estimators are unbiased but not efficient due to heteroscedasticity.
5) Assumption of 'No multicollinearity' means the correlation between the regressand (Y) and
regressor (X) is

a) High

b) Low

c) Zero

d) Any of the above

6. Which of the following tests helps in the detection of heteroskedasticity?

a. The Breusch-Pagan test
b. The Breusch-Godfrey test
c. The Durbin-Watson test
d. The Chow test

7. What will you conclude about a regression model if the Breusch-Pagan test results in a
small p-value?
a. The model contains homoskedasticty.
b. The model contains heteroskedasticty.
c. The model contains dummy variables.
d. The model omits some important explanatory factors.

8. A test for heteroskedasticty can be significant if _____.

a. the Breusch-Pagan test results in a large p-value
b. the White test results in a large p-value
c. the functional form of the regression model is misspecified
d. the regression model includes too many independent variables
9. Which of the following is a difference between the White test and the Breusch-Pagan test?
a. The White test is used for detecting heteroskedasticty in a linear regression model while
the Breusch-Pagan test is used for detecting autocorrelation.
b. The White test is used for detecting autocorrelation in a linear regression model while the
Breusch-Pagan test is used for detecting heteroskedasticity. .
c. The number of regressors used in the White test is larger than the number of
regressors used in the Breusch-Pagan test.
d. The number of regressors used in the Breusch-Pagan test is larger than the number of
regressors used in the White test.

10. Which of the following is true?

a. In ordinary least squares estimation, each observation is given a different weight.
b. In weighted least squares estimation, each observation is given an identical weight.
c. In weighted least squares estimation, less weight is given to observations with a higher
error variance.
d. In ordinary least squares estimation, less weight is given to observations with a lower error
variance.

11. For a given significance level, if the calculated value of the Durbin Watson statistic lies
between the lower critical value and the upper critical value, _____.
a. the hypothesis of no serial correlation is accepted
b. the hypothesis of no serial correlation is rejected
c. the test is inconclusive
d. the hypothesis of heteroskedasticity is accepted

12. In the presence of heteroskedasticity, the usual OLS estimates of:

a. standard errors are valid, whereas the t statistics and F statistics are invalid.
b. t statistics are valid, but the standard errors and F statistics are invalid.
c. F statistics are valid, but the standard errors and t statistics are invalid.
d. standard errors, t statistics, and F statistics are invalid.
13. The Breusch-Godfrey test statistic follows a:
a. χ2distribution.
b. t distribution.
c. normal distribution.
d. F distribution.

14. Which of the following statement(s) is correct?

a) The least squares estimator will have a smaller variance compared to the variance obtained
when serially correlated errors are correctly modelled. This will lead to greater power of
hypothesis tests and lower probability of committing a type II error.
b) The least squares estimator will have a smaller variance compared to the variance obtained
when serially correlated errors are correctly modelled. This will lead to greater power of
hypothesis tests and lower probability of committing a type I error.
c) The least squares estimator will have a larger variance compared to the variance
obtained when serially correlated errors are correctly modelled. This will lead to less
power of hypothesis tests and higher probability of committing a type II error.
d) None of the above is correct.

15. Which of the following statements is true about the White test?
a) The White test is used to detect the presence of multicollinearity in a linear regression
model.
b) The White test cannot detect heteroskedasticity in the presence of lagged values of the
dependent variable.
c) The White test can detect the presence of heteroskedasticity in a linear regression model
only if multicollinearity is not present.
d) The White test assumes that under the null hypothesis, the square of the error term
in a regression model is uncorrelated with all the independent variables, their squares
and possibly their cross-products
 TRUE AND FALSE

Indicate whether the following sentence is True or False. If it is false, explain why it is false.

1. The interpretation of goodness-of-fit measures changes in the presence of

heteroskedasticity. (1pts) FALSE

Feedback: The interpretation of goodness-of-fit measures is unaffected by the presence

of heteroskedasticty.

2. In a situation where heteroscedasticity is suspected, but there is not enough

information to identify its nature. To overcome the problem of biased standard error,
do the following: [state if the following solutions are true or false]

2.1. Use weighted least squares, i.e. transform the model by using
appropriate weights for the regression model so that the error variance
becomes homoscedastic. (1pts) FALSE
Feedback: we do not have enough information to identify its nature, so WLS not
applicable.

2.2.Run a standard OLS regression, ignoring heteroscedasticity, because in large

samples, this problem is not of any consequence and the t-tests and F-tests are
asymptotically valid. (1pts) FALSE
Feedback: t-tests and F-tests are always asymptotically invalid in the presence of
heteroscedasticity

2.3. Use White's heteroscedasticity-consistent variances and standard error

at least in large samples, in which the t-tests and F-tests are asymptotically
valid. (1pts) TRUE

3. In presence of positive serial correlation, the OLS variance formula underestimates the
true variance of the OLS estimator. (1pts) TRUE
Part II. Open Question (5 points)

Below you will find STATA regression output table of the following regression:
colgpa𝑖= 𝛽0+ 𝛽1Black𝑖+u𝑖:

colgpa is college GPA for students.

Black is dummy variable and takes value =1 if student is black

Summary Statistics for variable Black:

a) (1 pts) What is the predicted GPA for black students?

2.67-0.43=2.24
What is the predicted GPA for white students?
2.67
b) (1 pts) What is the sample mean of GPA? (Hint: use summary statistics)
2.67-0.43*0.05=2.64
 c) (2pts) Suppose you find evidence for heteroskedasticity in the error terms. Knowing
this information, is the effect of being a black student statistically significant at the
5% level? Explain.

we cannot say from this output, we need robust SE

Now assume you run new regression with new regressor “class size”.
colgpa𝑖= 𝛽0+ 𝛽1Black𝑖+ 𝛽1Size𝑖 +u𝑖:

colgpa is college GPA for students.

Black is dummy variable and takes value =1 if student is black
hsize > class size

d) (0.5 pts) What is the unit of analysis? – students

e) (0.5 pts) Interpret coefficient on “hsize” at 1% and 5% significance level.

At 1% insignificant – no interpretation***.

At 5% significant – interpret. If class size increases by one student, then GPA

decreases by 0.014 points, ceteris paribus.
Part III (15 points)

Question 1. (10 points)

We run the regression:

Population/Density𝑖= 𝛽0 + 𝛽1Income𝑖+ 𝛽2CrimeRate𝑖+u𝑖,

in a sample of n=100 observations. [Density in a county is explained by income and the

crime rate].

After estimating the above equation, we get the residuals, and their squared values (uhatt2).
Then we run the following regressions and receive the regression output Table A and Table B
(where cr2 is crime rate squared; income2 is income squared; and crinc=Income*crime rate).

a) (2 pts) Compute the Breusch-Pagan LM test statistic of Heteroskedasticity. Does

heteroskedasticity exist? Explain in detail (explicitly write down hypothesis testing,
calculated statistic, and decision rule).

1. H0: There is homoscedasticity.

Ha: There is heteroskedasticity

2. 0.0207*100=2.07
3. If calculated statistic is greater than CV reject null, if not fail to reject

2.07<5.991 (chi(2))(CV at 5% significance level) – fail to reject null, we have

homoskedasticity in errors
 b) (2 pts) Compute the White LM test statistic of Heteroskedasticity. Does
heteroskedasticity exist? Explain in detail (explicitly write down hypothesis testing,
calculated statistic, and decision rule).1

1. H0: There is homoscedasticity.

Ha: There is heteroskedasticity

2. 0.24*100=24
3. If calculated statistic is greater than CV reject null, if not fail to reject

24>11.007(chi(5))( (CV at 5% significance level) – reject null, we have

heteroskedasticity in errors

Table A

1
The Breusch-Pagan test only checks for the linear form of heteroskedasticity. The White test on the other
hand is more generic, it is able to detect more general form of heteroskedasticity than the Breusch-Pagan test.
The White can lose its power very quickly if the model has many regressors. That could be a reason why BP
contradicts White. Such situation is very rare. [ if student highlights this point , please be sure that you give
credits +0.5 points to him/her]
Table B

c) (2 pts) Suppose you find that heteroskedasticity exists in your data, but there is not
enough information to identify its nature. What would be your immediate suggestion
 to overcome the problem of biased standard error? Considering your solution, will the
estimated coefficients change? Suggest a solution and explain.

Use Robust SE, coefficients do not change.

d) (2 pts) Now assume you know a form of heteroskedasticity. Your main objective is to
solve the problem. Describe a procedure for how you will obtain efficient estimators.
And what if you misspecified the functional form of heteroskedasticity?

Using WLS (procedure should be explained at least shortly); More trouble,

misspecification bias in the functional form may cause standard errors to be
even more biased.

e) (2pts) Now assume the form is unknown to you, and you want to remedy a situation.
How will you express heteroskedasticity form, in which functional form? Explain
your decision and why you do so.

Using FGLS (procedure should be explained at least shortly); using exponential

functional form, to have positive variance.
Question 2. (5 points)

Given is the STATA regression output table of:

wkearnst= 𝛽0+ 𝛽1Wkearnst+ 𝛽2Wkhourst+ut

wkearns> is real weakly earnings

wkhours> is weekly hours

outphr> is output per labor hour

Table A
Table B
Table C

You should clearly explain what the auxiliary equation for the tests are, how you would
compute the test statistic, what is the null hypothesis, and what does it mean if the null
hypothesis is true.

a) (2pts) What is the (approximate) Durbin-Watson statistic for serial correlation?

Compute.

1) Ho: undefined

H1: There is serial correlation, rho>0

2) 2(1-0.84)=0.32

3) If d-statistic > CV of du fail reject null

If d-statistic < CV of dl reject null

If dl (=<) d-statistic (=<) du inconclusive

b) (2pts) What is the non-simple t-test statistic for serial correlation? Compute.

1) Ho: There is no serial correlation.

 H1: There is serial correlation.

2) 0.8701/0.102=8.53

3) If t-statistic > CV of reject null

If t-statistic < CV of fail to reject null

c) (1pts) What is the Breusch-Godfrey LM statistic for serial correlation? Compute.

1) Ho: There is no serial correlation.

H1: There is serial correlation.

2) 40*0.6=24

3) If LM-statistic > CV (chi1) reject null

If LM-statistic < CV (chi1) fail to reject null

Bonus Points  (2 points)

1. The equation u2t = α0 + α1u2t – 1 + vt is an autoregressive model in _____.

a. ut
b. u2t
c. vt
d. ut – 1

2. Which of the following statements is true?

a. When explanatory variables are not strictly exogenous, the t test for serial correlation is
valid.
b. When explanatory variables are not strictly exogenous, the Durbin Watson test for serial
correlation is valid.
c. Breusch-Godfrey test can be used to check for fifth order serial correlation.
d. White test can be used to check for second order serial correlation.