lOMoARcPSD|35105109

EC2020 2016 exam questions and answers

Introduction to economics (University of London)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Examiners’ commentaries 2016

EC2020 Elements of econometrics

Important note

This commentary reflects the examination and assessment arrangements for this course in the
academic year 2015–16. The format and structure of the examination may change in future years,
and any such changes will be publicised on the virtual learning environment (VLE).

Information about the subject guide and the Essential reading

references

Unless otherwise stated, all cross-references will be to the latest version of the subject guide (2011).
You should always attempt to use the most recent edition of any Essential reading textbook, even if
the commentary and/or online reading list and/or subject guide refer to an earlier edition. If
different editions of Essential reading are listed, please check the VLE for reading supplements – if
none are available, please use the contents list and index of the new edition to find the relevant
section.

General remarks

Learning outcomes

At the end of this course, and having completed the Essential reading and activities, you should be
able to:

• describe and apply the classical regression model and its application to cross-section data
• describe and apply the:
• Gauss–Markov conditions and other assumptions required in the application of the
classical regression model
• reasons for expecting violations of these assumptions in certain circumstances
• tests for violations
• potential remedial measures, including, where appropriate, the use of instrumental
variables
• recognise and apply the advantages of logit, probit and similar models over regression
analysis when fitting binary choice models
• competently use regression, logit and probit analysis to quantify economic relationships
using standard regression programmes (Stata and EViews) in simple applications
• describe and explain the principles underlying the use of maximum likelihood estimation
• apply regression analysis to fit time-series models using stationary time series, with
awareness of some of the econometric problems specific to time series applications (for
example, autocorrelation) and remedial measures
• recognise the difficulties that arise in the application of regression analysis to nonstationary
time series, know how to test for unit roots, and know what is meant by cointegration.

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Common mistakes committed by candidates

• A large number of candidates were not able to distinguish between sample variance and
covariance, and population variance and covariance (this is happening year after year).
They treat them as the same. This results in incorrect analysis and candidates lose
significant marks.
Consider an example: Suppose data are deviations from the respective sample means and
the regression model is:
yt = xt + ut , t = 1, 2, . . . , T.

The ordinary least squares estimator of is:

P
T P
T
x t yt xt ut
b= t=1 t=1
= + .
PT PT
x2t x2t
t=1 t=1

In terms of variances and covariances (a large number of candidates prefer this

terminology), this can be written as:

b= Cov(x, u)
+ .
Var(x)

Here Cov(x, u) and Var(x) are sample[Cov(x, u)] and sample[Var(x)].

P
T P
T
Candidates should realise that ut , xt ut , Cov(x, u) and Var(x) given above are sample
t=1 t=1
P
T PT
moments and as such ut 6= 0, xt ut 6= 0 and Cov(x, u) 6= 0. However, if we take the
t=1 t=1
expectation, then:
E(ut ) = 0
by assumption. Therefore:
" T
# T
X X
E xt ut = xt [E(ut )] = 0
t=1 t=1

as the xt s are fixed so they can be taken out of the expectation, and so:
" T
#
1X
E[Cov(x, u)] = E x t ut = 0
T t=1

as previously argued. This makes E( b) = , i.e. b is an unbiased estimator of .

To prove consistency take the plim to get:
0 1
1
P
T
⇣ ⌘ xt ut
B T
t=1
C
plim b = + plim B
@
C
A
1
PT
T x2t
t=1
✓ ◆
1
P
T
plim T xt ut
t=1
= + ✓ ◆
1
PT
2
plim T xt
t=1

plim(sample Cov(x, u))

= +
plim(sample Var(x))
population Cov(x, u)
= + .
population Var(x)

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

⇣ ⌘
By assumption, population Cov(x, u) = 0 and population Var(x) > 0, hence plim b = ,
in other words b is a consistent estimator of .
Remember that in general:

plim(sample variance) = population variance

and:
plim(sample covariance) = population covariance.
This concept has been used in many questions. This simple mistake of not distinguishing
between sample variance and covariance, and population variance and covariance, results in
a significant loss of marks which might result in the loss of a degree class or even be the
difference between pass and fail.
• Candidates struggled to give competent answers to the interpretation of empirical results.
When interpreting an empirical result you should discuss the significance of the coefficients,
magnitude and sign of the coefficients. Also, you should make sure that the Gauss–Markov
conditions hold. Gauss–Markov conditions have to be explicitly specified. Only writing that
the Gauss–Markov conditions hold is not sufficient.
• Just as last year, many candidates did not appear to read the questions carefully enough
and often omitted to give answers to parts of questions which asked for details of such
things as the assumptions necessary for a particular result to be true.

Key steps to improvement

Essential reading for this course includes the subject guide and the following.

Dougherty, C. Introduction to econometrics. (Oxford: Oxford University Press, 2011) 4th

edition [ISBN 9780199567089];
http://global.oup.com/uk/orc/busecon/economics/dougherty4e/

Apart from Essential readings you should do some supplementary readings. One very good book of
the same level is:

Gujarati, D.N. and D.C. Porter Basic econometrics. (McGraw–Hill, 2009, International
edition) 5th edition [ISBN 9780071276252].

To understand the subject clearly it is important to supplement C. Dougherty, Introduction to

econometrics (fourth edition) with the subject guide EC2020 Elements of econometrics (2014),
especially Chapter 10 which covers maximum likelihood.

It is very important to carefully go through the subject guide. The subject guide contains solutions
to the questions given in the main textbook and also some additional questions and solutions.
Working through these will improve your understanding of the subject.

The chapter in the subject guide on maximum likelihood (Chapter 10) includes some additional
theory which has not been covered in the main textbook. It is important to read the additional
theory given in the subject guide to have a better understanding of the principles of maximum
likelihood and tests based on the likelihood function.

Please check the VLE course page for resources for this subject such as a downloadable copy of the
subject guide EC2020 Elements of econometrics (2014), PowerPoint slideshows that provide
graphical treatment of the topics covered in the textbook, datasets and statistical tables. Candidates
should utilise datasets using standard regression programmes (STATA or EViews). This will help in
the understanding of the subject.

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Examination revision strategy

Many candidates are disappointed to find that their examination performance is poorer than they
expected. This may be due to a number of reasons. The Examiners’ commentaries suggest ways of
addressing common problems and improving your performance. One particular failing is ‘question
spotting’, that is, confining your examination preparation to a few questions and/or topics which
have come up in past papers for the course. This can have serious consequences.

We recognise that candidates may not cover all topics in the syllabus in the same depth, but you
need to be aware that the examiners are free to set questions on any aspect of the syllabus. This
means that you need to study enough of the syllabus to enable you to answer the required number of
examination questions.

The syllabus can be found in the Course information sheet in the section of the VLE dedicated to
each course. You should read the syllabus carefully and ensure that you cover sufficient material in
preparation for the examination. Examiners will vary the topics and questions from year to year and
may well set questions that have not appeared in past papers. Examination papers may legitimately
include questions on any topic in the syllabus. So, although past papers can be helpful during your
revision, you cannot assume that topics or specific questions that have come up in past examinations
will occur again.

If you rely on a question-spotting strategy, it is likely you will find yourself in difficulties
when you sit the examination. We strongly advise you not to adopt this strategy.

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Examiners’ commentaries 2016

EC2020 Elements of econometrics

Important note

This commentary reflects the examination and assessment arrangements for this course in the
academic year 2015–16. The format and structure of the examination may change in future years,
and any such changes will be publicised on the virtual learning environment (VLE).

Information about the subject guide and the Essential reading

references

Unless otherwise stated, all cross-references will be to the latest version of the subject guide (2016).
You should always attempt to use the most recent edition of any Essential reading textbook, even if
the commentary and/or online reading list and/or subject guide refer to an earlier edition. If
different editions of Essential reading are listed, please check the VLE for reading supplements – if
none are available, please use the contents list and index of the new edition to find the relevant
section.

Comments on specific questions – Zone A

Candidates should answer EIGHT of the following TEN questions: ALL of the questions in
Section A (8 marks each) and THREE questions from Section B (20 marks each). Candidates are
strongly advised to divide their time accordingly.

Section A

Answer all questions from this section.

Question 1

A random variable Y has unknown population mean µY and population variance

2
Y . A sample of n observations {Y1 , . . . , Yn } has been generated. Consider the
following estimator:

✓ ◆
1 1 3 1 3 1 3
Ỹ = Y1 + Y2 + Y3 + Y4 + · · · + Yn−1 + Yn
n 2 2 2 2 2 2

where the number of observations n is assumed to be even for convenience. Show

that Ỹ is a consistent estimator of µY .

(8 marks)

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapter R.14 (Probability limits

and consistency).

Gujarati, D.N. and D.C. Porter. Basic Econometrics (fifth edition) (ISBN 9780071276252)
Chapter A.7 (Statistical Inference: Estimation).

Approaching the question

The definition of consistency should be given. To answer this question, the sufficient condition of
consistency should be used. The answer is as follows.
p
Ỹ is a consistent estimator of µY if Ỹ ! µY . Ỹ is constructed by applying a weight of 1/2 to the
n/2 ‘odd’ observations, and a weight of 3/2 to the remaining n/2 observations. We have:
✓ ◆
1 1 3 1 3 1 3
E(Ỹ ) = E(Y1 ) + E(Y2 ) + E(Y3 ) + E(Y4 ) + · · · + E(Yn−1 ) + E(Yn )
n 2 2 2 2 2 2
✓ ◆
1 1 n 3 n
= ⇥ ⇥ µY + ⇥ ⇥ µY
n 2 2 2 2
= µY

and:
✓ ◆
1 1 9 1 9
var(Ỹ ) = var(Y1 ) + var(Y2 ) + · · · + var(Yn−1 ) + var(Yn )
n2 4 4 4 4
✓ ◆
1 1 n 9 n
= ⇥ ⇥ Y2 + ⇥ ⇥ Y2
n2 4 2 4 2
2
Y
= 1.25 .
n

Therefore, Ỹ is an unbiased estimator of µY , and because var(Ỹ ) ! 0 as n ! 1, Ỹ is consistent.

Question 2

Consider a two variable linear model:

Yi = 0 + 1 Xi + ui i = 1, 2, . . . , n

where E(ui | X) = 0 for all i; E(u2i ) = 2 ; and E(ui uj ) = 0 if i 6= j. Show that the
ordinary least squares (OLS) estimators b0 and b1 are unbiased estimators of 0 and
1 , respectively.

(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) 2.3 (The random components and
unbiasedness of the OLS regression coefficients).

Dougherty, C. Subject guide Chapter 2 (Properties of the regression coefficients and hypothesis
testing).

Gujarati, D.N. and D.C. Porter. Basic Econometrics (fifth edition) (ISBN 9780071276252)
Chapter 3A.2 (Linearity and Unbiasedness Properties of Least-Squares Estimators).

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Approaching the question

The derivation of the OLS estimators is not required. To show unbiasedness, candidates should
take expectations of the estimators. The answer is as follows.

The OLS estimator of the slope 1 is:

P
b1 (Xi X̄)(Yi Ȳ )
= P
(Xi X̄)2
P
(Xi X̄)([ 0 + 1 Xi + ui ] [ 0 + 1 X̄ + ū])
= P
(Xi X̄)2
P
(Xi X̄)(ui ū)
= 1 + P .
(Xi X̄)2

Also:
X X X
(Xi X̄)(ui ū) = (Xi X̄)ui ū (Xi X̄)
X ⇣X ⌘
= (Xi X̄)ui ū Xi nX̄
X
= (Xi X̄)ui
P
since Xi = nX̄. Hence:
P X
b1 = (Xi X̄)ui
1+ P = 1 + a i ui
(Xi X̄)2
P
where ai = (Xi X̄)/ (Xi X̄)2 . Therefore:
⇣X ⌘ X
E( b1 ) = E( 1) +E a i ui = 1 + ai E(ui ) = 1

since E(ui ) = 0 by assumption.

The OLS estimator of the intercept 0 is:

b0 = Ȳ b1 X̄

where b1 is the OLS estimator of the slope 1. Taking the expectation of b0 , we have:
✓ ◆
1X
E( b0 ) = E(Ȳ b1 X̄) = E 0 + 1 X̄ + ui b1 X̄
n
X
= 0 + E( 1
b1 )X̄ + 1 E(ui | Xi )
n
= 0

where the third equality in the above equation has used the facts that b1 is unbiased so
E( 1 b1 ) = 0, and E(ui | Xi ) = 0.

Question 3

Show that the R2 in the regression of Y on X (with an intercept) is the squared

value of the sample correlation between X and Y (i.e. R2 = rXY
2
).

(8 marks)

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapter 1.6 (Goodness of fit: R2 ).

Dougherty, C. Subject guide Chapter 1 (Simple regression analysis).

Approaching the question

R2 is given by:
Explained Sum of Squares (ESS)
R2 = .
Total Sum of Squares (TSS)
Simple algebraic manipulations of ESS and TSS is required to answer this question. The answer
is as follows.

The coefficient of determination, R2 , is given by:

P b
ESS
2 ( Yi Ȳ )2
R = =P .
TSS (Yi Ȳ )2

Given that:
P
b0 = Ȳ b1 X̄ b1 = (Xi X̄)(Yi Ȳ )
and P
(Xi X̄)2
we have:
X X
ESS = (Ybi Ȳ )2 = ( b0 + b1 Xi Ȳ )2
Xh i2
= b1 (Xi X̄)
X
= b2 (Xi X̄)2
1
⇥P ⇤2
(Xi X̄)(Yi Ȳ )
= P .
(Xi X̄)2

Hence:
⇥P ⇤2
2 ESS (Xi X̄)(Yi Ȳ )
R =P = P P
(Yi Ȳ )2 (Xi X̄)2 (Yi Ȳ )2
2 P 32
1
n−1 (Xi X̄)(Yi Ȳ )
= 4q q 5
1
P 2 1
P
n−1 (X i X̄) n−1 (Yi Ȳ )2
 2
sXY
=
sX sY
2
= rXY .

Question 4

Explain how you would estimate the parameters ↵ and for the model
(Mt /Pt ) = ↵rtβ given data on money (Mt ), prices (Pt ) and the nominal interest rate
(rt ). Interpret the results.

(8 marks)

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapter 4.2 (Logarithmic

transformations).

Dougherty, C. Subject guide Chapter 4 (Transformation of variables).

Gujarati, D.N. and D.C. Porter. Basic Econometrics (fifth edition) (ISBN 9780071276252)
Chapter 6.4 (Functional Forms of Regression Models).

Approaching the question

To answer this question, a logarithmic transformation of the given model is required. The answer
is as follows.

We transform the equation into an estimable form by taking logs to give:

✓ ◆
Mt
log = log(↵) + log(rt )
Pt

that is, a log-log model. Now we can estimate the parameters by using OLS. The estimate of ↵ is
given by exp(constant), and the slope is the estimate of . In this model is the interest rate
elasticity of demand for money (the percentage change in the demand for money for a 1 per cent
change in the interest rate). Alternatively, use a non-linear estimation process.

Question 5

What do you understand by an instrumental variable (IV)? How you would estimate
in the model yt = xt + ut , where xt and ut are correlated, using zt ,
t = 1, 2, . . . , T as an IV? Examine the consistency of the IV estimator.

(8 marks)

Reading for this question

Dougherty, C. Introduction to econometrics (fourth edition): Chapter 8.5 (Instrumental

variables).

Dougherty, C. Subject guide (2011): Chapter 8.5 (Stochastic regressor and measurement errors).

Approaching the question

Candidates should define an instrumental variable (IV) and when it is used. It is required to
prove that the IV estimator is consistent. The answer is as follows.

Consider the model:

yt = xt + u t , t = 1, 2, . . . , T.
If xt is not independently distributed of ut then the OLS estimator of will be inconsistent.
Consider a variable z that is correlated with x but not correlated with u. z can be considered as
an instrumental variable. An estimator of based on z is known as an instrumental variable
(IV) estimator. It is defined as: P
bIV = P zt yt .
zt x t

It can be shown that bIV is a consistent estimator of .

P P P
bIV = P zt yt = zt ( x t + u t )
P = +P
zt u t
zt x t zt x t zt x t

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

and: P
plim ( zt ut /T )
plim( bIV ) = + P =
plim ( zt xt /T )
hence bIV is a consistent estimator of .

Note that: ⇣X ⌘ ⇣X ⌘
plim zt ut /T =0 and plim zt xt /T 6= 0.

Section B

Answer three questions from this section.

Question 6

A researcher wants to test the hypothesis that drug companies practice price
discrimination against less developed countries. To do this she estimates two
demand for pharmaceuticals equations using data from 32 countries.

Equation A: Pi = 32.60 + 1.66GDP Ni 0.88CDi + e1i R2 = 0.70 i = 1, . . . , 32

(6.27) (0.24) (0.25)

Equation B: ln(Pi ) = 2.26 + 0.95 ln(GDP Ni ) 0.45 ln(CDi ) + e2i R2 = 0.71 i = 1 . . . , 32

(0.24) (0.13) (0.09)

where Pi is the price index for pharmaceuticals in each country, GDP Ni is the gross
domestic product (GDP) per capita for each country, and CDi is the per capita
consumption of pharmaceutical drugs. ln(Pi ), ln(GDP Ni ), ln(CDi ) are the
logarithms of these quantities. Figures in parentheses are the estimated standard
errors.

(a) For each of the estimated equations, are the estimated partial regression
coefficients individually statistically significant at the 5 percent level of
significance? Specify and justify the alternative hypothesis in each test.
(3 marks)
(b) Test the hypothesis that the slope coefficients are all simultaneously zero in
each equation. What do you conclude from your results?
(3 marks)
(c) Show that in a log-log model, such as (B), the slope parameters are elasticities.
(4 marks)
(d) The researcher claims that: (i) as the population of a country becomes richer
the cost of drugs decreases, and hence the demand becomes less elastic; and (ii)
as people consume more drugs the market becomes more developed, and hence
more competitive. To what extent, if at all, do the results in (A) and (B)
support these claims?
(6 marks)
(e) Are serial correlation and/or heteroscedasticity likely to be a problem in the
linear and log-linear regression models (A) and (B)? Explain.
(4 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 2.6 (Testing hypotheses

relating to the regression coefficients), 3.5 (Goodness of fit: R2 ) and 4.2 (Logarithmic
transformations).

10

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Dougherty, C. Subject guide Chapter 2 (Properties of the regression coefficients and hypothesis
testing).

Approaching the question

In part (a) individual coefficients are being tested, the t test is appropriate for this. Part (b)
involves a test of a joint hypothesis. To conduct the test, an F test should be used. In part (d)
the sign and significance of the coefficients should be explained. The answer is as follows.

(a) The critical t value for both equation (A) and equation (B) is 2.045 (degrees of freedom 29,
5% significance level for a 2-tailed test). The estimated t values are:

32.6 1.66 0.88

Equation (A) t= = 5.2, t= = 6.92 and t= = 3.52
6.27 0.24 0.25
2.26 0.95 0.45
Equation (B) t= = 9.42, t= = 7.31 and t= = 5.
0.24 0.13 0.09

These are all larger in absolute terms than the critical value. Hence we can say that all the
coefficients are significantly different from zero. The estimated parameters show that as
GDP N increases so does the price of pharmaceuticals, the income elasticity of the price for
pharmaceuticals is 0.95 (equation (B)). The estimated parameters also show that as CD
increases the price for pharmaceuticals decreases with elasticity 0.45.

(b) To test the hypothesis that all coefficients are simultaneously zero we need to use an F test,
which would normally be based on the residual sum of squares (RSS). Here the only
information we have is the R2 , which contains the same information. The F test is:

R2 /k 0.70/2
F = = = 33.83
(1 R2 )/(n k 1) (1 0.70)/(32 2 1)

for equation (A), and the F critical value is 3.33 with (2, 29) degrees of freedom. Hence we
reject the null hypothesis of zero coefficients. Equation (B) is similar: replace the value of
R2 with 0.71, and repeat the computation. The answer is similar.

(c) If the equation is log(Yi ) = 0 + 1 log(Xi ) + ui , and we differentiate it we get:

@Yi 1 1 @Yi Xi
= 1 or = 1 = elasticity.
@Xi Yi Xi @Xi Yi

(d) The hypothesis might suggest that as the income per capita increases the price of drugs
falls. It also says that as the income of the population increases the slope coefficient on
equation (B) falls. We have shown that the first statement is not correct – the coefficient on
GDP N is positive and significant. The second statement is not covered by the regressions,
which simply model fixed elasticities.
The second hypothesis says that as the consumption of drugs rises the market becomes more
competitive. There is evidence that this is supported by the equations; the coefficient on
CD (the consumption of drugs) is negative and significantly different from zero, which
suggests that as the consumption of drugs increases the price falls (i.e. the market becomes
more competitive).

(e) The data are cross-sectional, hence there is no possibility of serial correlation, but one might
expect heteroskedasticity since countries are going to be of very different sizes, and hence
the error term is likely to mirror some of these values.

11

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Question 7

The following estimates were calculated from a sample of 7,634 women respondents
from the General Household Survey 1995. The dependent variable takes the value 1
if the woman was in paid employment, and 0 otherwise.

OLS Logit Probit

high 0.093 0.423 0.259

(0.015) (0.071) (0.043)

noqual 0.210 0.898 0.554

(0.013) (0.056) (0.035)

age 0.038 0.173 0.107

(0.003) (0.124) (0.008)
age2 0.051 0.230 0.142
(0.003) (0.069) (0.009)

mar 0.024 0.103 0.063

(0.009) (0.057) (0.035)

Constant 0.068 2.587 1.593

(0.049) (0.225) (0.137)

Where high is one if the respondent has a higher educational qualification, zero
otherwise; noqual is one if the respondent has no qualifications, zero otherwise; age
is age in years; age2 is (age ⇥ age) 100; mar is one if married, zero otherwise.
Conventionally calculated standard errors are given in brackets for the ordinary
least squares (OLS) results and asymptotic standard errors in brackets elsewhere.

(a) Explain briefly how Probit estimates are calculated when the model has no
intercept and only one explanatory variable.
(6 marks)
(b) For all three sets of estimates, test the null hypothesis that the coefficient of
mar is zero. Which test statistics would you consider more reliable? Explain.
(6 marks)
(c) Using the OLS and Probit estimates, calculate the estimated probabilities of
being in employment for a married woman aged 40 with a higher educational
qualification. Comment on your results.
(5 marks)
(d) Test the null hypothesis that all the slope coefficients of the probit model are
jointly equal to zero, given that:

ln LR = 416.01
ln LU = 321.25

where ln LR and ln LU are the logs of the likelihood from the restricted and the
unrestricted probit models, respectively.
(3 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 10.1 (The linear

probability model), 10.3 (Probit analysis) and 10.6 (An introduction to maximum likelihood
estimation).

12

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Dougherty, C. Subject guide (2011) Chapter 10 (Binary choice and limited dependent variable
models, and maximum likelihood estimation).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 15.2 (The Linear Probability Model (LPM)), 15.5 (The Logit Model) and 15.9 (The
Probit Model).

Approaching the question

(a) The probit model uses the cumulative standardised normal distribution. The maximum
likelihood technique is used to obtain the estimates of the parameters. Estimates have the
standard maximum likelihood properties, i.e. the estimators are consistent, asymptotically
efficient and asymptotically normally distributed.
[For technical details see Dougherty (fourth edition) Section 10.3].
(b) To test the null hypothesis the t test should be used. t tables are attached with the
examination paper. Candidates should know how to look at the critical values. The null and
alternative hypotheses should be clearly stated:
H0 : coefficient of mar = 0
H1 : coefficient of mar 6= 0.
t test statistics are 2.67 (OLS), 1.81 (logit), and 1.8 (probit).
The test based on the OLS estimates gives outright rejection. However, the standard errors
for this test are wrongly calculated because of the heteroskedasticity of the error term.
Therefore, we prefer probit or logit estimates.
(c) A large number of candidates wrongly thought that to obtain the probability in the case of
probit a sophisticated calculator is needed. The probability in the case of probit can be read
directly from the supplied statistical tables. We have:
0.093 + 0.038 ⇥ 40 0.051 ⇥ 16 + 0.024 ⇥ 0.068 = 0.753 (OLS)
0.259 + 0.107 ⇥ 40 0.142 ⇥ 16 + 0.063 1.593 = 0.737 (probit).
The statistical tables give a probability of 0.77, approximately.
They are all fairly close. The OLS estimates do not fall outside the probability bounds.
(d) To test the null hypothesis, the large sample likelihood-ratio test should be used. In large
samples 2(ln LR ln LU ) is distributed as a chi-squared distribution with degrees of
freedom equal to the number of restrictions imposed by the null hypothesis. The null and
alternative hypotheses should be clearly stated as:
H0 : all the slope coefficients are = 0
H1 : at least one slope coefficient is 6= 0.
We have:
2
2(ln LR ln LU ) ⇠ 5
and:
2( 416.01 ( 321.25)) = 189.92.
2
The critical value of 5 at the 5% significance level is 11.07, hence we reject H0 .

Question 8

The relationship between a dependent variable Y and an explanatory variable X is

given by the linear model
Yi = ↵ + Xi + ui ; i = 1, 2, . . . , n
for an i.i.d. sample {Yi , Xi }, but where the errors ui are correlated with the
explanatory variable, E(Xi ui ) = Xu 6= 0.

13

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

✓ ◆
1 P
n
1 P
n
(a) Assuming that plim n
(Xi X̄)2 = XX , where X̄ = n
Xi , show that
i=1 i=1
the OLS estimator of is inconsistent.
(4 marks)
(b) Three possible instrumental variables z1 , z2 and z3 are available in the data set.
What properties do these instruments need to have in order to be able to use
them to estimate consistently by the Instrumental Variable, or the Two Stage
Least Squares (2SLS), estimator?
(4 marks)
(c) Using all three instruments, describe in detail the 2SLS estimator.
(4 marks)
(d) Assuming conditional homoskedasticity of the errors, E(u2i | zi ) = 2 , explain in
detail how you would test whether E(zi ui ) = 0. What are the consequences for
the 2SLS estimator if the test result implies a rejection of this hypothesis?
(4 marks)
(e) Explain in detail how would you test whether E(Xi ui ) = 0
(Durbin–Wu–Hausman test). Which estimator of would you use if this
hypothesis does not get rejected?
(4 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 8.3 (Asymptotic

properties of the OLS regression coefficients), 8.5 (Instrumental variables) and 9.3 (Instrumental
variable estimation).

Dougherty, C. Subject guide Chapters 8 (Stochastic regressors and measurement errors) and 9
(Simultaneous equations estimation).

Approaching the question

The plim should be used to verify the consistency of the OLS estimators. Properties of the
instrumental variables should be clearly explained. The answer is as follows.

(a) We have:
0 ⇣ ⇣P P ⌘⌘ 1
✓P ◆ plim 1
Xi ui ui n1 j Xj
(Xi X̄)ui n i
plim( bOLS ) = plim P = plim @ P A
(Xi X̄)2 plim n1 (Xi X̄)2

1
1 n Xu
= 6= 0.
XX

(b) Candidates need to explain that zi must be correlated with Xi but not with ui , and does
not already appear in the equation in its own right. Hence in the reduced form:
Xi = ⇡0 + ⇡1 zi1 + ⇡2 zi2 + ⇡3 zi3 + !i
⇡1 , ⇡2 and ⇡3 should all not be zero (if one is zero then it should be dropped from the list).
Furthermore, E(zi ui ) = 0.
(c) Estimate by OLS:
bi = ⇡
X b0 + ⇡
b1 zi1 + ⇡
b2 zi2 + ⇡
b3 zi3 .
Next, estimate by OLS:
b i + "i
Yi = ↵ + X
in the second stage. Xb only utilises exogenous variation due to the instruments, and hence
identifies the causal effect .

14

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

(d) nR2 test: Regress the IV residuals:

u
bIV = Yi bIV
↵ bIV Xi

on the instruments:
u
bIV = 0 + 1 zi1 + 2 zi2 + 3 zi3 + ⇣i .
2
nR in this regression follows a chi-squared distribution with 2 degrees of freedom under the
null hypothesis that E(zi ui ) = 0. If the test statistic is large, then reject the null hypothesis.
If the test rejects the null hypothesis, then the IV estimator is not consistent.
(e) Add the first-stage residuals to the model:
Y i = ↵ + Xi + !
b i + ⇠i .
Testing for exogeneity is a test of H0 : = 0. A t test is appropriate as standard errors are
correct under the null hypothesis. If the test does not reject the null hypothesis use the OLS
estimator of .

Question 9

Answer the following questions:

(a) Explain what a trend stationary series and what a difference stationary series
are. What is an important difference between the two types of stationarity?
(5 marks)
(b) Let the model be:
Yt = Yt−1 + ut ; t = 1, 2, . . . , T
where ut are independently and identically distributed as N (0, 2 ). Explain
how would you test H0 : | | = 1 against the alternative of a zero mean,
covariance stationary AR(1) process. Give the assumptions this test requires.
(5 marks)
(c) Consider an ADL(1,1) model:
Yt = ↵1 + ↵2 Yt−1 + ↵3 Xt + ↵4 Xt−1 + ut
where both Yt and Xt are I(1). Express the ADL(1,1) model in an error
correction form and interpret the coefficients of the error correction model.
Discuss the advantages of the error correction form.
(10 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 13.1 (Stationarity and

nontationarityy), 13.4 (Tests of nonstationarity) and 13.6 (Fitting models with nonstationary
time series).

Dougherty, C. Subject guide Chapter 13 (Introduction to nonstationary time series).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition) (ISBN 9780071276252):
Chapters 21.5 (Trend Stationary (TS) and Difference Stationary (DS) Stochastic Processes) and
21.9 (The Unit Root Test).

Approaching the question

In part (a) the meaning of a trend stationary series and a difference stationaary series should be
explained. Part (b) requires a discussion of the Dickey–Fuller test, and in part (c) the error
correction model (ECM) should be derived and the interpretation of the coefficients of the ECM
should be given. The answer is as follows.

15

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

(a) If after removing the trend from a nonstationary series the resulting variable becomes
stationary, then the variable is called trend stationary. Let:
Z t = Xt ↵1 t = ↵ 0 + ut
2
where E(ut ) = 0, var(ut ) = and E(ut ut−s ) = 0 for all s and t, then:
E(Zt ) = E(↵0 + ut ) = ↵0
2
var(Zt ) = Var(↵0 + ut ) =

cov(Zt , Zt−s ) = E[(Zt E(Zt ))(Zt−s E(Zt−s ))] = E(ut ut−s ) = 0.

This means that Zt has constant mean and variance for all t, and covariance is zero for all
s > 0. It implies that the series is trend stationary.
If a nonstationary process can be transformed into a stationary process by differencing then
the series is said to be difference stationary.
Let Xt be a random walk with a drift:
Xt = 0 + Xt−1 + "t (i)
2
where E("t ) = 0, var("t ) = and E("t "s ) = 0 for all s and t, s 6= t.
Subtract Xt−1 from both sides of (i) to get:
∆Xt = Xt Xt−1 = 0 + "t .
It can be easily checked that E(∆Xt ) = 0 , var(∆Xt ) = "2 and cov(∆Xt , ∆Xt−s ) = 0 for all
s and t, s 6= t. This means that ∆Xt is stationary. This implies that Xt is difference
stationary. It is important to know whether a variable is difference stationary or trend
stationary because for difference stationary variables shocks have a permanent effect
whereas for trend stationary variables shocks are transitory.
(b) The standard test for a unit root is due to Dickey and Fuller. In order to test the null
hypothesis of a random walk without drift against the alternative of a zero mean covariance
stationary AR(1) process:
yt = yt−1 + "t
subtract yt−1 from both sides:
yt yt−1 = yt yt−1 + "t

∆yt = ( 1)yt−1 + "t = ⇢yt−1 + "t .

Test H0 : ⇢ = 0 using:
⇢b 0
. ⌧b =
s.e.(b
⇢)
We cannot use the standard t test procedure in this case because the distribution of the
statistic is not a t distribution, so critical values have to be computed by Dickey and Fuller
using Monte Carlo techniques (Dickey–Fuller tables). The test is sensitive to the presence of
serial correlation in the error term so we need to take steps to remove the effects of this
serial correlation – this is done by including lagged values of ∆yt in the regression. The
statistic has the same asymptotic distribution as ⌧b.
(c) Consider a simple ADL(1,1) model (this is also known as ARDL(1, 1)):
Yt = ↵1 + ↵2 Yt−1 + ↵3 Xt + ↵4 Xt−1 + ut (i)
Rewrite (i) as:
Yt Yt−1 = ↵1 + ↵2 Yt−1 Yt−1 + ↵3 Xt ↵3 Xt−1 + ↵3 Xt−1 + ↵4 Xt−1 + ut

∆Yt = ↵1 (1 ↵2 )Yt−1 + ↵3 ∆Xt + (↵3 + ↵4 )Xt−1 + ut


↵1 ↵3 + ↵4
∆Yt = ↵3 ∆Xt (1 ↵2 ) Yt−1 Xt−1 + ut
1 ↵2 1 ↵2

∆Yt = ↵3 ∆Xt (1 ↵2 )[Yt−1 1 2 Xt−1 ] + ut

16

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

or:
∆Yt = ↵3 ∆Xt ⇡[Yt−1 1 2 Xt−1 ] + ut (ii)
where:
↵1 ↵3 + ↵4
⇡=1 ↵2 , 1 = and .
1 ↵2 1 ↵2
Equation (ii) is the ECM.
When the two variables Y and X are cointegrated, the ECM incorporates not only the
short-run effects but also the long-run effects. The long-run equilibrium Yt 1 2 Xt−1 is
included in the model together with the short-run effect captured by the differenced term.
All the terms in the ECM, given by (ii), are stationary. As Y and X are I(1), then ∆X and
∆Y are I(0). As Y and X are cointegrated their linear combination:

ut = Yt 1 2 Xt−1

is I(0).
The coefficient ⇡ provides us with the information about the speed of adjustment in cases of
disequilibrium.
• If ⇡ = 1 then 100% of the adjustment takes place within the period. In other words
adjustment is instantaneous and full.
• If ⇡ = 0.5 then 50% adjustment takes place each period.
• If ⇡ = 0 then there is no adjustment.

Question 10

(a) Consider a model:

Yt = Xt + ut ; t = 1, 2, . . . , T
where E(ut ) = 0; E(u2t ) = 2 and E(us ut ) = 0 if s 6= t for all s, t = 1, 2, . . . , T .
Xt is fixed in repeated samples. The density function of ut is given by:
✓ ◆
2 −1/2 u2t
f (ut ) = (2⇡ ) exp .
2 2
i. Derive the likelihood function.
(6 marks)
2
ii. Obtain the maximum likelihood estimators of and .
(8 marks)
(b) Consider a model:

Y t = ↵ + X t + ut ; t = 1, 2, . . . , T

where E(ut ) = 0; E(u2t ) = 2 and E(us ut ) = 0 if s 6= t for all s, t = 1, 2, . . . , T . u

is normally distributed. The parameters ↵ and have been estimated by
maximum likelihood.
Explain how the hypothesis that the coefficients are jointly equal to zero can be
tested.
(6 marks)

Reading for this question

Dougherty, C. Introduction to econometrics (fourth edition) Chapter 10.6 (An introduction to

maximum likelihood estimation).

Dougherty, C. Subject guide Chapter 10 (Binary choice and limited dependent variable models,
and maximum likelihood estimation).

17

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Approaching the question

Part (a) involves derivation of the likelihood function. To obtain the maximum likelihood
estimators the likelihood function should be differentiated with respect to the parameters and
equated to zero. Part (b) involves the likelihood-ratio test. The solution is as follows.

(a) i. The likelihood function is:

✓ P ◆
2 −T /2 (Yt Xt ) 2
L = (2⇡ ) exp 2
.
2

ii. The log-likelihood function is:

P
T T 2 (Yt Xt ) 2
log L = ln(2⇡) ln( ) 2
.
2 2 2
Differentiating, we have:
P P
@ log L 2 (Yt Xt )( Xt ) bM LE = PXt Yt
= =0 )
@ 2 2 Xt2

and:
P P bM LE Xt )2
@ log L T (Yt Xt ) 2 2 (Yt
= + =0 ) bM LE = .
@ 2 2 2 2 4 T

(b) This can be tested using the likelihood-ratio statistic 2(log L log L0 ). This is
asymptotically distributed as a chi-squared distribution with degrees of freedom equal to the
number of restrictions imposed by the null hypothesis, 2 in the present case.

18

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Examiners’ commentaries 2016

EC2020 Elements of econometrics

Important note

This commentary reflects the examination and assessment arrangements for this course in the
academic year 2015–16. The format and structure of the examination may change in future years,
and any such changes will be publicised on the virtual learning environment (VLE).

Information about the subject guide and the Essential reading

references

Unless otherwise stated, all cross-references will be to the latest version of the subject guide (2016).
You should always attempt to use the most recent edition of any Essential reading textbook, even if
the commentary and/or online reading list and/or subject guide refer to an earlier edition. If
different editions of Essential reading are listed, please check the VLE for reading supplements – if
none are available, please use the contents list and index of the new edition to find the relevant
section.

Comments on specific questions – Zone B

Candidates should answer EIGHT of the following TEN questions: ALL of the questions in
Section A (8 marks each) and THREE questions from Section B (20 marks each). Candidates are
strongly advised to divide their time accordingly.

Section A

Answer all questions from this section.

Question 1

What do you understand by an instrumental variable (IV)? How you would estimate
a model yt = xt + ut using an IV zt ; t = 1, 2, . . . , T ? Examine the consistency of
the IV estimator.

(8 marks)

Reading for this question

Dougherty, C. Introduction to econometrics (fourth edition): Chapter 8.5 (Instrumental

variables).

Dougherty, C. Subject guide: Chapter 8.5 (Stochastic regressor and measurement errors).

Approaching the question

Consider the model:

yt = xt + u t , t = 1, 2, . . . , T.

19

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

If xt is not independently distributed of ut then the OLS estimator of will be inconsistent. In

this situation an IV should be used. An IV should be:

• correlated with x
• should not be correlated with the disturbance term
• should not be an explanatory variable in its own right.

Let z be an IV. An estimator of based on z is known as an IV estimator. It is defined as:

P
bIV = P zt yt .
zt x t

It can be shown that bIV is a consistent estimator of .

P P P
bIV = P zt yt = zt ( x t + u t )
P = +P
zt u t
zt x t zt x t zt x t

and: P
plim ( zt ut /T )
plim( bIV ) = + P =
plim ( zt xt /T )

hence bIV is a consistent estimator of .

Note that: ⇣X ⌘ ⇣X ⌘
plim zt ut /T =0 and plim zt xt /T 6= 0.

Question 2

Derive the order of integration of xt in the following models. Assume in each case
that ut is stationary, x0 is fixed and E(us ut ) = 0 if s 6= t.

(a) xt = ↵0 + ut + ut−1 ; t = 1, 2, . . . , T .
(3 marks)
(b) xt = ↵0 + ↵1 xt−1 + ↵2 t + ut ; |↵| < 1; t = 1, 2, . . . , T .
(5 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition): Chapter 13.1 (Stationarity and

nonstationarity).

Dougherty, C. Subject guide: Chapter 13 (Introduction to nonstationary time series).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 21.5 (Trend Stationary (TS) and Difference Stationary (DS) Stochastic Processes) and
21.6 (Integrated Stochastic Process).

Approaching the question

In both (a) and (b), it should be shown that the mean and the variance of the variable is
constant and the covariances are independent of time but may depend on the length of the lag.
The solution is as follows.

(a) E(xt ) = ↵0 and var(xt ) = var(ut ut−1 ) = var(ut ) + var(ut−1 ) = 2 2 as E(us ut ) = 0 if s 6= t

and E(xt xt−s ) = 2 if s = 1 and 0 if s > 1. Therefore, xt is stationary or I(0).

20

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

(b) Lag this equation by one period to get:

xt−1 = ↵0 + ↵1 xt−2 + ↵2 (t 1) + ut−1 .

Subtracting this from the first equation gives:

∆xt = ↵1 ∆xt−1 + ↵2 + ut ut−1 .

Since |↵1 | < 1, ∆xt follows a stationary AR(1) process.

As E(xt ) = ↵0 + ↵1 E(xt−1 ) + ↵2 t, E(xt ) is clearly a function of time and E(xt ) 6= E(xt−1 ).
Therefore, xt is non-stationary. Hence xt is I(1) and it is trend stationary.

Question 3

Consider a model:

Yi = ↵ + Xi + ui ; i = 1, 2, . . . , 6

where E(ui ) = 0; E(u2i ) = 2

and E(ui uj ) = 0 if i 6= j.

The observations on Xi are

X1 X2 X3 X4 X5 X6
1 2 3 4 5 6

σ2
The OLS estimator of is b and V( b) = 17.5
.

An alternative estimator of is ˜ = 1
8
[Y6 + Y5 Y2 Y1 ].

Compare the sampling variance of ˜ with that of b .

(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (third edition) Chapters R.5, R.6, 1.4, 2.5, 12.3 and
12.4.

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 3.4 (Properties of Least Squares Estimators: The Gauss–Markov Theorem) and 3A.3
(Variances and Standard Errors of Least Squares Estimators).

Approaching the question

Replace the values of the Xi s in the equation given by ˜. The variance of ˜ has to be calculated
and compared with the variance of b (the variance of b is given). While deriving the variance, be
careful to take into account that E(ui uj ) = 0 if i 6= j. The answer is as follows.

Replace the value of the Xi s:

˜ = 1
[Y6 + Y5 Y2 Y1 ]
8
1
= [(↵ + 6 + u6 ) + (↵ + 5 + u5 ) (↵ + 2 + u2 ) (↵ + + u1 )]
8
1
= [8 + u6 + u5 u2 u1 ] (i).
8

21

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

From (i), it is easy to see that:


˜ 1
var( ) = var (u6 + u5 u2 u1 ) (since var(8 ) = 0)
8
1
= [var(u6 ) + var(u5 ) + var(u2 ) + var(u1 )] (since E(ui uj ) = 0 if i 6= j)
82
4 2
=
64
2
2
= (since var(ui ) = for all i).
16
Hence var( ˜) > var( b) and so b is more efficient in comparison to ˜.

Question 4

Consider a model:
yt = ↵yt−1 + ut
where
ut = ✓"t−1 + "t ; t = 1, 2, . . . , T
and E("t ) = 0, E("2t ) = 2
and E("s "t ) = 0 if s 6= t for all s, t = 1, 2, . . . , T .

Show that the OLS estimator of ↵ is inconsistent if ✓ 6= 0.

(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition): Chapter 12 (Definition and

consequences of autocorrelation).

Dougherty, C. Subject guide: Chapter 12 (Properties of regression models with time series data).

Approaching the question

The model contains a lagged dependent variable as an explanatory variable and also has
autocorrelation. The ordinary least squares estimator will be inconsistent. To prove
inconsistency, plim should be used. The answer is as follows.

The OLS estimator of ↵ is:

P P P P
yt yt−1 ut yt−1 "t yt−1 ✓ "t−1 yt−1
b= P 2
↵ =↵+ P 2 =↵+ P 2 + P 2
yt−1 yt−1 yt−1 yt−1
where yt has been substituted as yt = ↵yt−1 + "t + ✓"t−1 .

Since yt−1 = ↵yt−2 + "t−1 + ✓"t−2 hence:

P
"t yt−1
plim = 0.
T
It is easy to check that: P
"t−1 yt−1
plim 6= 0.
T
It is also assumed: P 2
yt−1
plim
T
is greater than zero and finite.

Therefore, plim(b
↵) 6= ↵, and so ↵
b is inconsistent.

22

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Question 5

Consider an ADL(1, 1) model:

Yt = ↵1 + ↵2 Yt−1 + ↵3 Xt + ↵4 Xt−1 + ut ; t = 1, 2, . . . , T.

where both Yt and Xt are I(1). ut is the disturbance term where E(ut ) = 0;
E(u2t ) = 2 and E(us ut ) = 0 for all s 6= t. Express the ADL(1, 1) model in an error
correction form and interpret the coefficients of the error correction model.

(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition): Chapter 13.6 (Fitting models with
nonstationary time series).

Gujarati, D.N. and D.C. Porter, Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapter 21.11 (Cointegration: Regression of a Unit Root Time Series on Another Unit Root
Time Series).

Approaching the question

The ADL(1,1) model should be expressed in error correction form. It involves some simple
algebraic manipulations. A detailed interpretation of the coefficients of the error correction
model should be given. The answer is as follows.

Consider a simple ADL(1,1) model (this is also known as ARDL(1, 1)):

Yt = ↵1 + ↵2 Yt−1 + ↵3 Xt + ↵4 Xt−1 + ut (i)

Rewrite (i) as:

Yt Yt−1 = ↵1 + ↵2 Yt−1 Yt−1 + ↵3 Xt ↵3 Xt−1 + ↵3 Xt−1 + ↵4 Xt−1 + ut

∆Yt = ↵1 (1 ↵2 )Yt−1 + ↵3 ∆Xt + (↵3 + ↵4 )Xt−1 + ut


↵1 ↵3 + ↵4
∆Yt = ↵3 ∆Xt (1 ↵2 ) Yt−1 Xt−1 + ut
1 ↵2 1 ↵2

∆Yt = ↵3 ∆Xt (1 ↵2 )[Yt−1 1 2 Xt−1 ] + ut

or:
∆Yt = ↵3 ∆Xt ⇡[Yt−1 1 2 Xt−1 ] + ut (ii)
where:
↵1 ↵3 + ↵4
⇡=1 ↵2 , 1 = and .
1 ↵2 1 ↵2
Equation (ii) is the ECM.

When the two variables Y and X are cointegrated, the ECM incorporates not only the short-run
effects but also the long-run effects. The long-run equilibrium Yt 1 2 Xt−1 is included in the
model together with the short-run effect captured by the differenced term.

All the terms in the ECM, given by (ii), are stationary. As Y and X are I(1), then ∆X and ∆Y
are I(0). As Y and X are cointegrated their linear combination:

ut = Y t 1 2 Xt−1

is I(0).

The coefficient ⇡ provides us with the information about the speed of adjustment in cases of
disequilibrium.

23

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

• If ⇡ = 1 then 100% of the adjustment takes place within the period. In other words
adjustment is instantaneous and full.
• If ⇡ = 0.5 then 50% adjustment takes place each period.
• If ⇡ = 0 then there is no adjustment.

Section B

Answer three questions from this section.

Question 6

(a) Describe an adaptive expectations model.

(6 marks)
(b) Koyck investigated the relationship between investment in railcars and the
volume of freight carried on the U.S. railroads using annual data for the period
1884–1939. Assuming that the desired stock of railcars in year t depended on
the volume of the freight in year t 1 and t 2 and a time trend, and assuming
that investment in railcars was subject to a partial adjustment process, he fitted
the following regression using ordinary least squares:

Ibt = 0.077Ft−1 + 0.017Ft−2 0.003t 0.110Kt−1 ; R2 = 0.85; t = 1, 2, . . . , T

where It = Kt Kt−1 is investment in railcars in year t (thousands), Kt is the

stock of railcars at the end of year t (thousands), and Ft is the volume of freight
handled in year t (ton-miles).
i. Explain how Koyck’s model can be derived from an adjustment equation
Kt Kt−1 = (Kt∗ Kt−1 ) and a behavioural equation with dependent
variable Kt∗ , using variables F (lagged) and t (time).
(6 marks)
ii. Using Koyck’s estimated equation estimate the parameters of your
behavioural equation. What are the implications for the behaviour in the
long run?
(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapter 11.4 (Models with a lagged
dependent variable).

Dougherty, C. Subject guide: Chapter 11 (Models using time series data).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapter 17.5 (Rationalization of the Koyck Model: The Adaptive Expectations Model).

Approaching the question

(a) An adaptive expectations model involves a learning process in which, in each time period,
the actual value of the variable is compared with the value that has been expected. If the
actual value is greater, the expected value is adjusted upwards for the next period. If it is
lower, the expected value is adjusted downwards. The size of the adjustment is hypothesised
to be proportional to the discrepancy between the actual and expected value.
If X is the variable in question, and Xte is the value expected in time period t given the
information available at time period t 1, then:
e
Xt+1 Xte = (Xt Xte ) 0 1

24

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

or:
e
Xt+1 = Xt + (1 )Xte .

The model is derived to capture the changing nature of the formation of expectations, often
in variables that are also changing with time. It is an attempt at a ‘simple learning’ solution
to model building in order to forecast macroeconomic variables. Such variables include
investment, savings and the demand for assets.

The model is estimated by repeated substitution for the expected variable, by its lagged
variant which has known components of the previous period and the unobserved expectation
lagged, until the term on the unobserved expectation (1 )s is so small as to be ignored
resulting in a model where all the variables are observed, where s is the period lagged and
is the speed of adjustment of expected and actual and is between 0 and 1. Technical
details should be given.

(b) i. Given the information in the question, the model may be written as:

Kt∗ = 1 Ft−1 + 2 Ft−2 + 3t + ut .

It is given that:
Kt Kt−1 = It = (Kt−1
∗
Kt−1 ).

Hence:
It = 1 Ft−1 + 2 Ft−2 + 3t Kt−1 + ut .

ii. Estimates of the coefficients should be obtained and interpreted. The working and
interpretations are given below.
From the fitted equation:

b = 0.110, b1 = 0.077 = 0.70, b2 = 0.017 = 0.15 and b3 = 0.0033

= 0.030.
0.110 0.110 0.110

Hence the short-run effect of an increase of 1 million ton-miles of freight is to increase

investment in rail-cars by 77 one year later and 17 two years later. It does not make
much sense to talk of the short-run effect of a time trend.
In the long-run equilibrium, neglecting the effects of the disturbance term, Kt and Kt∗
are both equal to the equilibrium value K̄, and Ft−1 and Ft−2 are both equal to their
equilibrium value F̄ . Hence:
K̄ = ( 1 + 2 )F̄ + 3 t.

Therefore, an increase of one million ton-miles of freight will increase the stock of
rail-cars by 850 and the time trend will be responsible for a secular decline of 33 rail-cars
per year (obsolescence/damage).

Question 7

The following estimates were calculated from a sample of 7,634 women respondents
from the General Household Survey 1995. The dependent variable takes the value 1
if the woman was in paid employment, and 0 otherwise.

25

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

OLS Logit Probit

high 0.093 0.423 0.259

(0.015) (0.071) (0.043)

noqual 0.210 0.898 0.554

(0.013) (0.056) (0.035)

age 0.038 0.173 0.107

(0.003) (0.124) (0.008)
age2 0.051 0.230 0.142
(0.003) (0.069) (0.009)

mar 0.024 0.103 0.063

(0.009) (0.057) (0.035)

Constant 0.068 2.587 1.593

(0.049) (0.225) (0.137)

Where high is one if the respondent has a higher educational qualification, zero
otherwise; noqual is one if the respondent has no qualifications, zero otherwise; age
is age in years; age2 is (age ⇥ age) 100; mar is one if married, zero otherwise.
Conventionally calculated standard errors are given in brackets for the ordinary
least squares (OLS) results and asymptotic standard errors in brackets elsewhere.

(a) Explain briefly how Probit estimates are calculated when the model has no
intercept and only one explanatory variable.
(6 marks)
(b) For all three sets of estimates, test the null hypothesis that the coefficient of
mar is zero. Which test statistics would you consider more reliable? Explain.
(6 marks)
(c) Using the OLS and Probit estimates, calculate the estimated probabilities of
being in employment for a married woman aged 40 with a higher educational
qualification. Comment on your results.
(5 marks)
(d) Test the null hypothesis that all the slope coefficients of the probit model are
jointly equal to zero, given that:

ln LR = 416.01
ln LU = 321.25

where ln LR and ln LU are the logs of the likelihood from the restricted and the
unrestricted probit models, respectively.
(3 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 10.1 (The linear

probability model), 10.3 (Probit analysis) and 10.6 (An introduction to maximum likelihood
estimation).

Dougherty, C. Subject guide (2011) Chapter 10 (Binary choice and limited dependent variable
models, and maximum likelihood estimation).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 15.2 (The Linear Probability Model (LPM)), 15.5 (The Logit Model) and 15.9 (The
Probit Model).

26

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Approaching the question

(a) The probit model uses the cumulative standardised normal distribution. The maximum
likelihood technique is used to obtain the estimates of the parameters. Estimates have the
standard maximum likelihood properties, i.e. the estimators are consistent, asymptotically
efficient and asymptotically normally distributed.
[For technical details see Dougherty (fourth edition) Section 10.3].
(b) To test the null hypothesis the t test should be used. t tables are attached with the
examination paper. Candidates should know how to look at the critical values. The null and
alternative hypotheses should be clearly stated:

H0 : coefficient of mar = 0
H1 : coefficient of mar 6= 0.

t test statistics are 2.67 (OLS), 1.81 (logit), and 1.8 (probit).
The test based on the OLS estimates gives outright rejection. However, the standard errors
for this test are wrongly calculated because of the heteroskedasticity of the error term.
Therefore, we prefer probit or logit estimates.
(c) A large number of candidates wrongly thought that to obtain the probability in the case of
probit a sophisticated calculator is needed. The probability in the case of probit can be read
directly from the supplied statistical tables. We have:

0.093 + 0.038 ⇥ 40 0.051 ⇥ 16 + 0.024 ⇥ 0.068 = 0.753 (OLS)

0.259 + 0.107 ⇥ 40 0.142 ⇥ 16 + 0.063 1.593 = 0.737 (probit).

The statistical tables give a probability of 0.77, approximately.
They are all fairly close. The OLS estimates do not fall outside the probability bounds.
(d) To test the null hypothesis, the large sample likelihood-ratio test should be used. In large
samples 2(ln LR ln LU ) is distributed as a chi-squared distribution with degrees of
freedom equal to the number of restrictions imposed by the null hypothesis. The null and
alternative hypotheses should be clearly stated as:

H0 : all the slope coefficients are = 0

H1 : at least one slope coefficient is 6= 0.

We have:
2
2(ln LR ln LU ) ⇠ 5

and:
2( 416.01 ( 321.25)) = 189.92.
2
The critical value of 5 at the 5% significance level is 11.07, hence we reject H0 .

Question 8

Economists have tried to examine the catch-up hypothesis where it is predicted that
poorer countries will grow faster than richer countries to ‘catch-up’ to the richer
countries. To assess this hypothesis an economist regresses the growth rate of GDP
(gross domestic product), gri , on log of GDP per capita, ln(gdpi ), as well as a
regression including other variables which might influence growth rates. The
following regression results come from regressions using data from 96 countries:

gri = 0.058 0.009 ln(gdpi ) + e1i (A)

(0.023) (0.002)

n = 96, R2 = 0.122 and s = 0.025

27

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

where estimated standard errors are given in brackets, gri is the growth rate of the
ith country over the period 1985–1990, ln(gdpi ) is the logarithm of GDP for country
i, e1i is the residual and s is the estimated standard error of the residuals.

A second regression yields

gri = 0.019 + 0.001 ln(gdpi ) + 0.12invi + e2i (B)

(0.023) (0.002)

n = 96, R2 = 0.196 and s = 0.024

where e2i is the estimated residual and invi is the ratio of investment to GDP in
1985.

(a) Test, for each regression, the hypothesis that the coefficient on ln(gdpi ) is zero.
What does the result of your tests say about the ‘catch-up’ hypothesis?
(7 marks)
2
(b) The R measure is higher for equation (B) than for (A). Is this what you would
expect? Explain.
(5 marks)
(c) The regression results may be affected by heteroskedasticity. Explain what you
understand about heteroskedasticity, why the estimation results might be
affected by heteroskedasticity and what the effects are if heteroskedasticity is
present.
(8 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 3.3 (Properties of multiple

regression coefficients), 3.5 (Goodness of fit: R2 ) and 7.1 (Heteroscedasticity and its
implications).

Dougherty, C. Subject guide: Chapters 3 (Multiple regression analysis) and 7

(heteroscedasticity).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 3.5 (The Coefficient of Determination R2 : A Measure of ‘Goodness of Fit’), 11.1 (The
Nature of Heteroscedasticity), 11.2 (OLS Estimation in the Presence of Heteroscedasticity) and
13.9 (Model Selection Criteria).

Approaching the question

(a) It is required to conduct tests of significance and relate the results of the tests to the
‘catch-up’ hypothesis.
The t-values are (A) 4.5 and (B) 0.043. The degrees of freedom are (A) 94 and (B) 93 giving
95% two-tailed critical values of ±1.99. We reject the null hypothesis for (A), but do not
reject for (B). A better answer would be to use a one-tailed test since the alternative
hypothesis is that as GDP increases, the growth rate will fall, hence the coefficient will be
negative, but the answer will be the same although the critical value is now 1.66. The
results show that (A) supports the ‘catch-up’ hypothesis, but (B) does not. If (B) is the true
model then (A) will suffer from omitted variable bias so that the ‘catch-up’ hypothesis is in
doubt.

(b) It should be explained that R2 is a non-decreasing function of the number of explanatory

variables in the model. Candidates should show that as the number of explanatory variables
is increased, R2 will never decrease.

28

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

R2 is defined as:
explained sum of squares
R2 = .
total sum of squares
As extra explanatory variables are added to an equation the explained sum of squares
increases or does not change. Total sum of squares does not change, hence R2 will never
decrease. The result is as expected.
(c) The definition of heteroskedasticity is required. Candidates should explain the consequences
of heteroskedasticity on the properties of ordinary least squares estimators.
The data are cross-sectional and deal with countries which are probably at very different
levels of GDP, hence heteroskedasticity is to be expected. Heteroskedasticity is the
condition where the error term variance is not constant and ordinary least squares produces
unbiased and consistent, but inefficient, parameter estimates. Also standard errors are not
correct hence making t and F tests invalid.

Question 9

(a) Explain the concept of the likelihood function and the maximum likelihood
estimator.
(6 marks)
(b) Suppose that an event occurs with probability p. In a simple random sample of
size n the event occurs m times.
m
i. Show that the maximum likelihood estimator (MLE) of p is n
. Verify that
the MLE maximises the likelihood function.
(8 marks)
ii. Derive the likelihood ratio statistic for the null hypothesis p = p0 . If m = 40
and n = 100, test the null hypothesis p = 0.5.
(6 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapter 10.6 (An introduction to

maximum likelihood estimation).

Dougherty, C. Subject guide: Chapter 10 (Binary choice and limited dependent variable models,
and maximum likelihood estimation).

Gujarati, D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 4.4 (The Method of Maximum Likelihood), 4A.1 (Maximum Likelihood Estimation of
Two-Variable Regression Model) and 8A2 (Likelihood Ratio (LR) Test).

Approaching the question

(a) It is required to explain the concept of the likelihood function and the maximum likelihood
estimator (MLE).
Definition: Let the probability density function (pdf) of X be f (x; ✓) where ✓ is a
parameter, then the likelihood function of a simple random sample {X1 , X2 , . . . , Xn } from
X is the product of the individual densities of the Xi s taken as a function of ✓.
The joint pdf of {X1 , X2 , . . . , Xn } is:

f (x1 , x2 , . . . , xn ; ✓) = f (x1 ; ✓) f (x2 , ✓) · · · f (xn ; ✓).

The likelihood function L(✓; x1 , x2 , . . . , xn ) has the same formulation as the joint pdf, but
now it is a function of ✓.

29

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

Maximum Likelihood Estimator (MLE)

Let X ⇠ f (x; ✓). The MLE ✓b of the parameter ✓ is an estimator that maximises the
likelihood function.
To obtain the MLE set:
@L
=0
@✓
b
and solve for ✓.
It is easier to take logs and then maximise, as in both the situations we will arrive at the
same maximum, because:
@ ln L 1 @L @L
= =0 ) = 0.
@✓ L @✓ @✓
To verify the maximum, the second derivative evaluated at the value of the MLE should be
checked. It should be negative.
(b) To test the null hypothesis a large sample likelihood ratio test should be used.
i. p is the probability of the event occurring, hence (1 p) is the probability that the event
does not occur. The event occurs m times and does not occur (n m) times (n is the
total number of observations).
The joint probability of the event occurring and not occurring is pm (1 p)n−m . The
log-likelihood function is:

log L(p) = m log(p) + (n m) log(1 p).

Differentiating with respect to p we get:

d log p m n m
= .
dp p 1 p
Equating it to zero we obtain the maximum likelihood estimator pb of p as:
m
pb = .
n
We should check that the second differential is negative to verify the maximum:
d2 log p m n m
= .
dp2 p2 (1 p)2
Evaluating it at p = m/n we obtain:

d2 log p n2 n m
= < 0.
dp2 m (1 m/n)2
So we have chosen a value of p which maximises the likelihood function.
ii. The log-likelihood function is:

log L(p) = m log(p) + (n m) log(1 p).

The likelihood-ratio statistic is:

h⇣ m ⇣ m ⌘⌘ i
LR = 2 m log + (n m) log 1 (m log p0 + (n m) log(1 p0 ))
n n
 ✓ ◆ ✓ ◆
m/n 1 m/n
= 2 m log + (n m) log .
p0 1 p0
If m = 40 and n = 100 the LR statistic for H0 : p = 0.5 is:
 ✓ ◆ ✓ ◆
0.4 1 0.4
LR = 2 40 log + 60 log = 4.03.
0.5 1 0.5
The critical value of 21 at the 5% significance level is 3.84, and at the 1% significance
level is 6.64. We would reject H0 at the 5% but not at the 1% significance level.

30

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

Examiners’ commentaries 2016

Question 10

(a) There are six missing values (denoted by x’s) in the given Stata output
i. Model SS
ii. Model df
iii. R-squared
iv. avetemp t value
v. harvrain 95% Conf. Interval.
Give the formulae for them and obtain their values based on the given output.
Source | SS df MS Number of obs = 27
-------------+------------------------------ F( 3, 23) = 21.90
Model | xxxxxxxxxx xx 2.58519261 Prob > F = 0.0000
Residual | 2.71468685 23 .118029863 R-squared = xxxxxx
-------------+------------------------------ Adj R-squared = 0.7069
Total | 10.4702647 26 .402702488 Root MSE = .34355
------------------------------------------------------------------------------
lnprice | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
wintrain | .001282 .0005765 2.22 0.036 .0000894 .0024747
avetemp | .7123178 .1087674 xxxx 0.000 .4873154 .9373202
harvrain | -.0036242 .0009646 -3.76 0.001 xxxxxxxxx xxxxxxxxx
_cons | -13.44433 1.969396 -6.83 0.000 -17.51834 -9.370326
------------------------------------------------------------------------------
where the explanatory variables are
wintrain: the level of winter rain (October – March) in millimetres
avetemp: average temperature in the growing season (April – September)
harvrain: level of harvest rain (August – September) in millimetres
and the dependent variable (lnprice) is the price of mature red wine from the
Bordeaux region of France at harvest time. The data relate to wines of different
vintages (ages) in 1987.
(10 marks)
(b) Interpret the regression results in (a).
(3 marks)
(c) If the age of the wine is included in the equation the estimates become:
Source | SS df MS Number of obs = 27
-------------+------------------------------ F( 4, 22) = 26.39
Model | 8.66443586 4 2.16610897 Prob > F = 0.0000
Residual | 1.80582883 22 .082083129 R-squared = 0.8275
-------------+------------------------------ Adj R-squared = 0.7962
Total | 10.4702647 26 .402702488 Root MSE = .2865
------------------------------------------------------------------------------
lnprice | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
wintrain | .0011668 .000482 2.42 0.024 .0001671 .0021665
avetemp | .6163926 .0951755 6.48 0.000 .4190107 .8137745
harvrain | -.0038606 .0008075 -4.78 0.000 -.0055353 -.0021858
age | .0238474 .0071667 3.33 0.003 0.0089846 .0387103
_cons | -12.31227 1.677212 -7.34 0.000 -15.79059 -8.833945
------------------------------------------------------------------------------
i. What is the interpretation of the coefficient of the variable age?
(3 marks)

31

Downloaded by A A ([email protected])
 lOMoARcPSD|35105109

EC2020 Elements of econometrics

ii. If the true model is given by the specification in (c) do you think that the
model in (a) should show evidence of specification error? Does it? Carefully
explain your observations.
(4 marks)

Reading for this question

Dougherty, C. Introduction to Econometrics (fourth edition) Chapters 4.2 (Logarithmic

Transformations) and 6.2 (The effect of omitting a variable that ought to be included).

Dougherty, C. Subject guide: Chapters 4 (Transformation of variables) and 6 (Specification of

regression variables).

Gujarati D.N. and D.C. Porter Basic Econometrics (fifth edition): (ISBN 9780071276252),
Chapters 6.6 (Semilog Models: Log-Lin and Lin-Log Models) and 13.3 (Consequences of Model
Specification Errors).

Approaching the question

(a) It is required to calculate the missing values. Formulae for each must be given.
i. Model SS = 7.7556.
ii. R-squared = 0.741.
iii. Model df = 3.
iv. avetemp t value = 6.55.
v. harvrain 95% CI = [ 0.0056, 0.0016].
(b) The dependent variable is in log form. The answer must explain the implication of this.
The critical t value is t27−4 = 2.069 hence all coefficients are significantly different from 0.
wintrain increases the log price by 0.001 per millimetre, avetemp increases log price by
0.712 per degree and harvrain reduces log price by 0.0036 per millimetre.
(c) In answering part (ii), it should be mentioned that age appears to be part of the true model.
Hence we have an omitted variable. The sign and size of omitted variables bias should be
discussed.
The model in (a) has a specification error in that age appears to be part of the true model.
The omitted variable bias will be related to the true coefficient and the correlation between
the omitted variable and the remaining variables – the true coefficient is presumably
non-zero and positive, but age is unlikely to be correlated to any of the included variables,
hence one would not expect any measurable bias – and this is what you observe. The
difference between the two models is very small.

32

Downloaded by A A ([email protected])