Advanced Econometrics Practice Exam

Master Econometrics and Operations Research

School of Business and Economics

Exam: Advanced Econometrics (4.1)

Code: E EORM AECTR

Examinator: -
Co-reader: -

Date: -
Time: -
Duration: 2 hours and 45 minutes

Calculator: Not Allowed

Graphical calculator: Not Allowed
Scrap paper: Allowed

Number of questions: 3

Type of questions: Multiple Choice and Open

Answer in: English

Credit score: 100 credits counts for a 10

Grades: -
Inspection: -
Number of pages: 7, including front page

• Give justifications for your answers unless stated otherwise.

• Be complete and explicit, but also clear and concise in your statements.

• If you think that further information is needed to answer a question, or that the
question is ill-posed, then explain your reasoning

• The questions should be handed back at the end of the exam. Do not take it home.

Good luck!

1
Question 1 [40 points] - Multiple Choice

For each of the following multiple choice questions, please indicate which statement is
correct by selecting an option (a), (b), (c) or (d). Only one option is correct.

Note: You get 4 points for each correct answer and -1 point for every incorrect answer.

Note: Please write your answers on your answer sheet (so not on this sheet). Clearly
indicate your choice. No justifications are needed.

1. Consider the following Asymmetric GARCH model

xt = σ t εt for every t ∈ Z where {εt }t∈Z ∼ NID(0, 1),

where σt2 = ω + αx2t−1 + δxt−1 + βσt−1

2
for every t ∈ Z.

Which of the following statements is correct?

(a) A positive parameter δ > 0 can be used to account for the ‘Leverage Effect’. If
δ > 0, then past negative returns xt−1 < 0 tend to have a smaller effect on the
conditional volatility σt2 than positive returns xt−1 > 0 of equal magnitude.
(b) A negative parameter δ < 0 can be used to account for the ‘Leverage Effect’.
If δ < 0, then past negative returns xt−1 < 0 tend to have a smaller effect on
the conditional volatility σt2 than positive returns xt−1 > 0 of equal magnitude.
(c) A positive parameter δ > 0 can be used to account for the ‘Leverage Effect’. If
δ > 0, then past negative returns xt−1 < 0 tend to have a larger effect on the
conditional volatility σt2 than positive returns xt−1 > 0 of equal magnitude.
(d) A negative parameter δ < 0 can be used to account for the ‘Leverage Effect’.
If δ < 0, then past negative returns xt−1 < 0 tend to have a larger effect on the
conditional volatility σt2 than positive returns xt−1 > 0 of equal magnitude.

2. Let {xt }t∈Z be a strictly stationary and ergodic sequence with two
PT bounded moments
2 2 1
E|xt | < ∞. Let yt = 5xt and consider the sample average T t=1 yt .
Which of the following statements is correct?

(a) The sample average converges to the expectation E(xt ) by application of a law
of large numbers.
(b) The sample average may or may not converge to the expectation E(xt ).
(c) The sample average converges to the expectation E(5x2t ) by application of a
law of large numbers.
(d) The sample average does not converge to the expectation E(5x2t ).

2
3. Let {xt }t∈Z be a strictly stationary and ergodic sequence of stock returns with four
bounded moments E|xt |4 < ∞. Consider the following GARCH filtering equation
for the conditional volatility with θ := (ω, α, β) = (0.1, 0.15, 0.95),

σt2 = 0.1 + 0.15x2t−1 + 0.95σt−1

2
for every t ∈ Z.

Which of the following statements is correct?

(a) The filtered volatility {σ̂t2 (θ, σ̂12 )}t∈N is invertible at θ and asymptotically sta-
tionary.
(b) The filtered volatility {σ̂t2 (θ, σ̂12 )}t∈N is not invertible at θ and not asymptoti-
cally stationary.
(c) The filtered volatility {σ̂t2 (θ, σ̂12 )}t∈N is invertible at θ but not asymptotically
stationary.
(d) The filtered volatility {σ̂t2 (θ, σ̂12 )}t∈N is not invertible at θ but it is asymptoti-
cally stationary.

4. Let θ0 be the unique maximizer of the limit deterministic criterion Q∞ : Θ → R.

Which of the following statements is correct?

(a) θ0 is identifiably unique.

(b) θ0 may or may not be identifiably unique.
(c) θ0 is identifiably unique if the parameter space Θ is compact.
(d) θ0 is identifiably unique if the criterion converges uniformly.

5. Consider the following GARCH-in-mean model

xt = µ + λσt + σt εt for every t ∈ Z where {εt }t∈Z ∼ NID(0, 1),

where σt2 = ω + α(xt−1 − µ − λσt−1 )2 + βσt−1

2
for every t ∈ Z,

where ω ≥ a > 0, α > 0 and β > 0. Say you want to estimate the parameter vector
θ = (µ, λ, ω, α, β) using Maximum Likelihood (ML).
What is the correct expression of the log likelihood function that you will use for
your estimation?
PT 1 1 2 2 (xt −µ)2
(a) t=2 − 2 log(2π) − 2 log(σ̂t (θ, σ̂1 )) − 2σ̂t2 (θ,σ̂12 ) .
PT 1 1 2 2 (xt −µ−λσ̂t2 (θ,σ̂12 ))2
(b) t=2 − 2 log(2π) − 2 log(σ̂t (θ, σ̂1 )) − 2σ̂t2 (θ,σ̂12 )
.
PT 1 1 2 2 (xt −µ−λσ̂t (θ,σ̂12 ))2
(c) t=2 − 2 log(2π) − 2 log(σ̂t (θ, σ̂1 )) − t 2σ̂ 2 (θ,σ̂ 2 )
1
.
(d) None of the other answers are correct.

3
6. Consider the regression yt = f (xt ) + εt .
Select the correct statement:

a. xt is endogenous if xt and εt are independent random variables.

b. yt is endogenous if yt is independent of xt .
c. The problem of omitted variable bias can result in xt being endogenous, in which
case xt is dependent on εt .
d. The problem of simultaneity can lead to xt being exogenous, in which case xt is
dependent on εt .

7. Let {xt }t∈Z be a sequence of independent random variables uniformly distributed

on the interval [−1, 1], i.e. {xt }t∈Z ∼ UID([−1, 1]).
Which of the following statements is correct?

(a) {xt }t∈Z is iid, white noise, weakly stationary, and strictly stationary.
(b) {xt }t∈Z is iid, weakly stationary, and strictly stationary but not white noise.
(c) {xt }t∈Z is strictly stationary but not white noise.
(d) {xt }t∈Z is iid but not strictly stationary.

8. Let {xt }t∈Z be a strictly stationary and ergodic sequence with four bounded moments
E|xt |4 < ∞. Define θ := (ω, α, β) and consider the following filtering equation for
the conditional mean of {xt }t∈Z ,

µ̂t+1 = ω + α(xt − µ̂t ) + β µ̂t for every t ∈ N.

Which of the following statements is correct?

(a) The filtered conditional mean {µ̂t (θ, µ̂1 )}t∈N is invertible at θ ∈ Θ, if |α| < 1.
(b) The filtered conditional mean {µ̂t (θ, µ̂1 )}t∈N is invertible at θ ∈ Θ, if |β| < 1,
but not asymptotically stationary.
(c) The filtered conditional mean {µ̂t (θ, µ̂1 )}t∈N is invertible at θ ∈ Θ and asymp-
totically stationary if |β| < 1.
(d) The filtered conditional mean {µ̂t (θ, µ̂1 )}t∈N is invertible at θ ∈ Θ and asymp-
totically stationary if |β − α| < 1.

4
 9. Consider two competing models, A and B. Suppose you have at your disposal an
estimation sample and an independent test sample.

(a) Under appropriate regularity conditions, the Hausman test statistic can be
used to test if model A predicts better than model B.
(b) Under appropriate regularity conditions, the Diebold-Mariano test statistic can
be used to test if model A nests model B.
(c) Under appropriate regularity conditions, the Hausman test statistic can be
used to test if model A is correctly specified using two different estimators for
the parameters of model A.
(d) Under appropriate regularity conditions, the Diebold-Mariano test statistic can
be used to test if model A is correctly specified using two different estimators
for the parameters of model A.

10. Consider the following random coefficient autoregressive model,

xt+1 = βt xt + εt

where {βt }t∈Z and {εt }t∈Z are iid random variables with two bounded moments.
Which of the following statements is correct?

(a) If E(βt ) < 1 ∀ t then {xt }t∈Z is strictly stationary and ergodic.
(b) E|βt | < 1 is a necessary condition for {xt }t∈Z to be stationary and ergodic.
(c) E log |βt | < 1 is a sufficient condition for {xt }t∈Z to be stationary and ergodic.
(d) None of the other statements is correct.

5
Question 2 [30 points]
In economics and finance, time-series may sometimes exhibit time-varying conditional
mean and volatility.

Let {xt }t∈Z be generated according to

x t = µ t + σt ε t for every t ∈ Z ,

where {εt }t∈Z is a sequence of Gaussian iid random variables {εt }t∈Z ∼ NID(0, 1). Suppose
that the time-varying conditional mean {µt }t∈Z satisfies


µt = 0.2 xt−1 − µt−1 + 0.7µt−1 for every t ∈ Z.

Furthermore, let the time-varying volatility {σt }t∈Z be determined by an exogenous se-
quence {zt }t∈Z , according to


σt = 1 + tanh(zt ) for every t ∈ Z.

Finally, let {zt }t∈Z be generated by the following random coefficient autoregressive model

zt+1 = ρt zt + vt for every t ∈ Z ,

where {ρt }t∈Z is a sequence of iid random variables with uniform distribution {ρt }t∈Z ∼
UID(0 , 1.5) taking values in the interval [0 , 1.5], and {vt }t∈Z is a sequence of Student-t
iid random variables with two degrees of freedom {vt }t∈Z ∼ T ID(2).

Note: the acronym iid stands for independent identically distributed.

Note: the function 1 + tanh(·) is uniformly bounded between 0 and 2.
Note: the random variable vt satisfies E|vt |n < ∞ for 0 < n < 2.

(a) (13pts) Can you show that {σt }t∈Z is strictly stationary and ergodic?

(b) (17pts) Can you show that E|xt |2 < ∞? Is {xt }t∈Z weakly stationary?

6
Question 3 [30 points]

Some econometricians claim that the temporal dependence in the growth rate of the Gross
Domestic Product (GDP) is stronger during economic recession periods and weaker during
expansions.

Let the sample of GDP growth rates {xt }Tt=1 at your disposal be a subset of the realized
path of a strictly stationary and ergodic time-series {xt }t∈Z with bounded moments of
fourth order E|xt |4 < ∞. Consider the Gaussian logistic Self Excited Smooth Transition
Autoregressive (SESTAR) model

xt = α + g(xt−1 ; θ)xt−1 + εt for every t ∈ Z where {εt }t∈Z ∼ NID(0, σε2 )

γ
and g(xt−1 ; θ) := for every t ∈ Z.
1 + exp(βxt−1 )
Suppose that the parameters θ = (α, γ, β, σε2 ) of the model are estimated by maximum
likelihood (ML) on a compact parameter space Θ with σε2 > 0. Note also that g(x; θ) is
uniformly bounded since |g(x; θ)| ≤ |γ| for every (x, θ). Suppose that the ML estimator
θ̂ T is consistent for a parameter θ 0 in the interior of Θ.

(a) (13pts) Can you derive the limit distribution

√ of the derivative of the log likelihood
function evaluated at θ 0 (multiplied by T )?

(b) (10pts) Explain how you would show asymptotic normality of θ̂ T . Be explicit about
the conditions and concepts you would use. No derivations are needed.
Note: you can assume that certain functions are well behaved and continuously differentiable.

(c) (7pts) Explain how you can use the approximate distribution of θ̂ T to test the
claim that the temporal dependence in the growth rate of Dutch GDP is stronger
during economic recession periods and weaker during expansions.