STRATHMORE INSTITUTE OF MATHEMATICAL SCIENCES

Bachelor of Business Science Financial Economics

BSE 4211: Panel Data

Special Examination

Date: 5th May 2022 Time: 2 Hours

Read the Following Instructions Carefully

1. This examination consists of FIVE questions.

2. Question One carries 30 marks.

3. All other questions carry 20 marks EACH

4. Answer Question ONE (COMPULSORY) and any other TWO ques-

tions.

5. This is a CLOSED BOOK exam.

BSE 4211: Panel Data Page 2 of 12 April 2022

Question 1

1. (a) Which of the following is a disadvantage of the random effects approach to estimat-
ing a panel model? (2 marks)
a) The approach may not be valid if the composite error term is correlated with
one or more of the explanatory variables
b) The number of parameters to estimate may be large, resulting in a loss of degrees
of freedom
c) The random effects approach can only capture cross-sectional heterogeneity and
not temporal variation in the dependent variable.
d) All of (a) to (c) are potential disadvantages of the random effects approach.
(b) The ”within transform” involves (2 marks)
a) Taking the average values of the variables
b) Subtracting the mean of each entity away from each observation on that entity
c) Estimating a panel data model using least squares dummy variables
d) Using both time dummies and cross-sectional dummies in a fixed effects panel
model
(c) Treating each data point in panel data as an individual observation and not account-
ing for the fact that individuals are observed over many time periods is referred to
as estimating a ( ) model. (2 marks)
A)static time-series
B)heteroskedastic consistent
C)pooled cross-section
D)fixed-effects
(d) Fixed-effects models are more appropriate than random-effects models when (2 marks)
A) the error terms are autocorrelated.
B)the independent variables are constant over time.
C)the time invariant component of the error term is correlated with any of the
independent variables.
D)the error term is not constant.
(e) In order to determine whether to use a fixed effects or random effects model, a
researcher conducts a Hausman test. Which of the following statements is false? (2 marks)
a) For random effects models, the use of OLS would result in consistent but ineffi-
cient parameter estimation
BSE 4211: Panel Data Page 3 of 12 April 2022

b) If the Hausman test is not satisfied, the random effects model is more appropriate.
c) Random effects estimation involves the construction of ”quasi-demeaned” data
d) Random effects estimation will not be appropriate if the composite error term is
correlated with one or more of the explanatory variables in the model

(f) True or False: Give a brief discussion supporting your answer for full marks - It is
much easier to obtain instruments with panel data than with cross-sectional data(2 marks)

(g) Consider a panel regression of unemployment rates for the G7 countries (United
States, Canada, France, Germany, Italy, United Kingdom, Japan) on a set of ex-
planatory variables for the time period 1980-2000 (annual data). If you included
cross-sectional effects and time effects with an intercept, you would need to specify
the following number of binary variables: (2 marks)

a. 21
b. 6
c. 28
d. 26

(h) Which of the following is a difference between a fixed effects estimator and a first-
difference estimator? (2 marks)

a. The fixed effects estimators are always larger than the first difference estimators
in a two-period panel data analysis.
b. The fixed effects estimator is more efficient than the first-difference estimator
when the idiosyncratic errors are serially uncorrelated.
c. The first difference estimator is more sensitive to nonnormality and heteroskedas-
ticity.
d.The bias in the first difference estimator depends on the time period (T) of analysis
while the bias in the fixed effect does not depend on T .

(i) Which of the following is a property of dummy variable regression? (2 marks)

a. This method is best suited for panel data sets with many cross-sectional obser-
vations.
b. The R-squared obtained from this method is lower than that obtained from
regression on demeaned data.
c. The degrees of freedom cannot be computed directly with this method.
d. The major statistics obtained from this method are identical to that obtained
from regression on demeaned data.
BSE 4211: Panel Data Page 4 of 12 April 2022

(j) How does the random effects model capture cross-sectional heterogeneity? (6 marks)

(k) Explain how fixed effects models are equivalent to an ordinary least squares regres-
sion with dummy variables. (6 marks)

Total for Question 1: 30 marks

BSE 4211: Panel Data Page 5 of 12 April 2022

Question 2

2. (a) You use a dataset called JTRAIN to determine the effect of the job training grant
on hours of job training per employee. The basic model for the three years (1987-
1989) is:

hrsemp it =β0 + δ1 d88t + δ2 d89t + β1 grant it + β2 grant i,t−1

+ β3 log ( employ it ) + ai + uit

The dependent variable is annual hours of training per employee, at the firm level.

grant = 1 if firm received grant for training

d89 = 1 if year = 1989
d88 = 1 if year = 1988
employ # employees at plant

The equation is estimated using fixed effects.

i. How many firms are used in the FE estimation? (1 mark)

ii. How many total observations would be used if each firm had data on all variables
(in particular, hrsemp) for all three years? (1 mark)

iii. Interpret the coefficient on grant and comment on its significance. (2 marks)

iv. Interpret the coefficient on the year dummy variables. (2 marks)

v. Is it surprising that grant −1 is insignificant? Explain. (2 marks)

BSE 4211: Panel Data Page 6 of 12 April 2022

vi. Do larger firms provide their employees with more or less training, on average?
How big are the differences? (4 marks)

(b) The Im-Pesaran-Shin Test (IPS test) is an example of an alternative to the Levin
Lin Chu test above. Briefly describe the test, making sure to highlight how it
compares to the LLC test (8 marks)

Total for Question 2: 20 marks

BSE 4211: Panel Data Page 7 of 12 April 2022

Question 3

3. (a) The file MATHPNL contains panel data on school districts in Michigan for the years
1992 through 1998. The response variable of interest in this question is math4, the
percentage of fourth graders in a district receiving a passing score on a standardized
math test.

mathit =δ1 y94t + . . . + δ5 y98t + γ1 log (rexppit ) + γ2 log rexppi,t−1

+ ψ1 log ( enrol it ) + ψ2 lunch it + ai + uit ,

The key explanatory variable is rexpp, which is real expenditures per pupil in the
district. The amounts are in 1997 dollars. This spending variable will appear in
logarithmic form.

The lunch variable is the percent of students in the district eligible for free or
reduced-price lunches, which is determined by poverty status. enrol is the school
enrolment rate in the district

Additionally,
y94 = 1 if year == 1994
y95 = 1 if year == 1995
y96 = 1 if year == 1996
and so on ...

You first estimate the model by pooled OLS and report the usual standard errors
in parentheses. The results are reported below:

i. Discuss the estimated size, sign and statistical significance of the effects of the
spending variables (3 marks)

ii. Is the sign of the lunchit coefficient what you expected? Interpret the magni-
tude and significance of the coefficient. Would you say that the district poverty
rate has a big effect on test pass rates? (4 marks)
BSE 4211: Panel Data Page 8 of 12 April 2022

iii. Now, you estimate the equation by fixed effects. Is the lagged spending variable
still the same size and is it significant? What has changed in this model? (3 marks)

(b) The Arrelano-Bond (1991) and Arrelano-Bover (1995)/Blundell-Bond (1998) dy-

namic panel estimators are increasingly popular. Both are general estimators de-
signed for situations with “small T, large N” panels, dynamic relationships, fixed
individual effects and heteroskedasticity and autocorrelation within individuals.

Provide a detailed discussion of The Arrelano & Bond Estimator and the Arrelano
& Bover/Blundell & Bond Estimator making sure to highlight the key differences
between the two estimators? (10 marks)

Total for Question 3: 20 marks

BSE 4211: Panel Data Page 9 of 12 April 2022

Question 4

4. (a) Explain what cross-sectional dependence is (2 marks), 3 possible causes (3 marks),

and how it is tested for in Panel Data (2 marks) (7 marks)

(b) In the traffic dataset, you have 1982–1988 state-level data for 48 U.S. states on
traffic fatality rates (deaths per 100,000). You seek to model the highway fatality
rates (fatal ) as a function of 4 independent variables: beertax, the tax on a case of
beer, spircons, a measure of spirits consumption and two economic factors: the state
unemployment rate (unrate) and state per capita personal income, $000 (perinc)

Answer the following questions after carrying out Panel Data Estimation in STATA.

i. You estimate the above model using a Fixed Effects Estimation. Is the model
jointly significant? How did you infer this? (2 marks)

ii. Given the above results how do we interpret the coefficients on the state un-
employment rate (unrate) and state per capita personal income, $000 (perinc).
Focus on the size, sign and statistical signficance of the two coefficients (4 marks)

iii. Are there significant unobserved effects from the model? How did you infer the
result? (2 marks)

iv. You generate an interaction term between: spircons: a measure of spirits con-
sumption and the state unemployment rate (unrate). You then run the same
BSE 4211: Panel Data Page 10 of 12 April 2022

model as above, NOW including the interaction variable (a total of 5 indepen-

dent variables now). Results are below
How would you NOW interpret the size and sign effect of unrate, the state
unemployment rate on the fatality rates? (5 marks)

Total for Question 4: 20 marks

BSE 4211: Panel Data Page 11 of 12 April 2022

Question 5

5. (a) Panel unit root testing emerged from time series unit root testing. The major
difference to time series testing of unit roots is that we have to consider asymptotic
behavior of the time-series dimension T and the cross-sectional dimension N.

With the help of relevant formulae and hypothesis, provide a detailed discussion of
the Fisher Type Test (attributed to Maddala- Wu (1991) and Choi (2001)) (6 marks)

(b) Diaz-Fernandez, et al (2017) authored a paper title “HRM practices and Innovation
Performance: A Panel-Data Approach”. The purpose of this paper is to study the
relationship between human resource management (HRM) practices and innovation
performance in Spanish manufacturing firms. The paper focuses on the number of
existing patents, analyzing the extent to which this variable is favored by HRM
practices.

The variables and results from 2 model estimations are presented below. Answer
the questions that follow:

i. Using the results from Model 2, interpret the effect of Employment Security on
Innovation performance (2 marks)

ii. Using the results from Model 2, how is Investment in training in new technolo-
gies in the firm related to innovation performance? (2 marks)

iii. Using results from Model 1, interpret the results on “OECD2” and “OECD3”(2 marks)
BSE 4211: Panel Data Page 12 of 12 April 2022

iv. What do the model fit statistics reported in the table tell us? Interpret accord-
ingly (2 marks)

v. Are the instruments in Model 2 valid? (2 marks)

vi. Explain the results from the AR(1) and AR(2) tests? Why are they carried
out? Do they support the model? (4 marks)

Total for Question 5: 20 marks