ECO 321-Econometrics II
Practice Question set 1
Course Instructor: Bertha Nguluwe
QUESTION 1
Given the following model 𝑌𝑖 = 𝛽2 𝑋𝑖 + 𝜀𝑖 and 𝜎𝑖2 = 𝜎 2 𝑋𝑖2
2 ∑ 𝑋2
a. ̂2 ) = 𝜎
Show that 𝑣𝑎𝑟(𝛽
𝑖
2
2
(∑ 𝑋𝑖 )
b. Step-by-step describe a procedure you would follow to test that 𝜎𝑖2 = 𝜎 2 𝑋𝑖2
c. What are the limitations of the outlined test?
d. If the test in ‘b’ confirms the relationship, what will be the implications on the properties
̂2
of the OLS estimator of 𝛽
e. To obtain constant variances, the researcher needs to use the generalized least squares
technique. Why is the GLS technique recommended for this type of problem?
QUESTION 2
Suppose you have two models one of which is correctly specified
𝑦𝑡 = 𝛼 + 𝛽𝑥𝑡 + 𝜀𝑡
𝑦𝑡 = 𝛼 + 𝛽𝑥𝑡 + 𝜆𝑥𝑡2 + 𝜀𝑡
a. Explain how you would detect the correctly specified model
b. Illustrate that it is better for the second model to be mis-specified than the first model to be
mis-specified.
QUESTION 3
Consider the following linear regression model,
𝑌𝑖 = 𝛽0 + 𝛽1 𝑋𝑖 + 𝜀𝑖 and 𝑣𝑎𝑟(𝜇𝑖 ) = 𝜎 2 exp(𝑋𝑖𝛼 𝜇𝑖 )
Comment on the properties of the OLS estimator of 𝛽1
a. Derive an expression for the variance expression that can be estimated using OLS
b. Explain how you could use the expression to conduct a simple test of heteroscedasticity
(Hint: heteroscedasticity is about the relationship between the error variance and one of the
regressors).
c. Explain how you will use the expression to obtain efficient estimates of the regressor.
QUESTION 4
Examine whether the following statements are true or false. Give an explanation.
a. In multiple regression, a high correlation in the sample among the regressors implies that
the least squares estimators of the coefficients are biased.
b. If the coefficient estimates in an equation have high standard errors, this is evidence of
high multicollinearity.
c. Heteroskedasticity in the errors leads to non-efficient estimates of the regression
coefficients and their standard errors.
d. Multicollinearity is a data problem and not estimation problem.
e. Any heteroskedasticity and autocorrelation test can be validly applied if the model is
correctly specified
QUESTION 5
The table below presents results of an import demand model estimated Stata for one of the
European countries.
Imports(dependent variable) Coefficient Standard error
GDP 0.0265 0.1877
Stock formation 0.4148 0.3223
Consumption 0.2515 0.2868
Intercept -19.8261 4.1543
ESS 2576.87
RSS 71.440
N 18
a. Justify the use of the three variables as regressors in the import demand model.
b. Compute 𝑅 2 and F-statistic and test the significance of the model
c. Test the null hypotheses that the slope coefficients are equal to zero.
d. From the tests in (b) and (c), which classical linear regression model is likely violated?
e. Suppose the following auxiliary regression was obtained from the same data
𝐶𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 = 0.6539𝐺𝐷𝑃 − 0.0341𝑆𝑡𝑜𝑐𝑘 + 12.2010𝑅 2 = 0.998
𝑆. 𝐸.(0.008)(0.2901)(1.8454)
i. Comment on the results of the consumption equation in the context of the identified
problem.
ii. Compute and interpret the variance inflation factor for consumption
iii. What would be the consequence of dropping either GDP or consumption as an
alternative of solving the problem in the original model?
QUESTION 6
The following data was used to estimate the model below:
a. Interpret the coefficient on variable GDP
b. Generate a scatter plot of residuals on year and comment on the pattern
c. Use the Durbin-Watson d statistic to test for first-order serial correlation
d. Explain why you should not jump to correct for autocorrelation basing on your results
in ‘b’ above
e. If the conclusion of the test statistic in ‘b’ above is to be trusted, what implications do
they have on the estimated parameters
f. Generate a series of transformed data that can be used to estimate the parameters of
the model asymptotically efficiently
g. Use the data to test for the presence of heteroskedasticity using the Park test and the
Goldfeld-Quandt Test.
Question 7
In trying to understand the determinants of economic growth, a researcher obtained the
following results. In the model, labour measures labour supply in million dollars, capital
stands for capital stock in million dollars and gdp stands for gross domestic product in
million dollars. The prefix ln- stands for natural logarithm. Data was used for a period
1974 to 2012. Study the results carefully and answer the following questions.
a. Derive the equation that depict the theoretical relationship between gdp, labour and
capital the researcher assumed to come up with model specification 1 and 2.
b. Comment on the estimates from the two model specifications
c. Determine the model that has a better fit based on AIC
d. What is the difference between the test in question “b” and Ramsey’s RESET?
e. Test the significance of the variables in the correct specification based on “c” and
interpret the results
f. Use the Durbin Watson’s d statistic to test for autocorrelation in the two models and
comment on the findings
g. Comment on what the researcher was supposed to do based on the findings in “d”
above to improve the model
h. State any three weaknesses of the test in “d” above
i. In 1990, the government implemented structural reforms to the economy and the
researcher believes that the general performance of the economy changed and that the
influence of labour and capital on gdp changed as well. Illustrate how the researcher
would test these (2 marks)
QUESTION 8
In the stata output presented below, lngdp is the log of GDP, lnlabour is the log of Labour while
lncapital is the log of capital. Laguhat is the lagged value of the residual term and d stands for the
difference operator. The estimation uses data from 1980 to 2008.
Model 1
Model 2
Model 3
a. The researcher first estimated model 1
i. Test the null hypothesis that lnlabour does not influence lngdp
ii. Test the fit of the model
iii. The Durbin Watson d statistic presented after model 1, was computed after the
data was arranged in increasing order of lncapital which is not included in the
model. Use the statistic to test if the influence of lncapital is captured in the
residential term.
b. The estimates on lnlabour in models 1 and 2 are different. Show that this result is an
expected result
c. Explain how the Ramsey RESET test after model 2 is derived and used
d. Interpret the results of the Ramsey’s RESET test
e. After assessing the time series properties of the variables, the researcher was convinced
that he should estimate model 3
i. What is the name given to model 3
ii. Under which conditions are you expected to estimate this model
iii. Interpret the coefficient on laguhat
iv. How would you interpret the coefficient on dlncapital
