ECM9/ECM10/ECM11/MGM3
MPhil in Economics
MPhil in Economic Research
MPhil in Finance & Economics
MPhil Finance
Friday 10th May 2024 10:00am - 12:00pm
R301A
TIME SERIES
Candidates are required to answer …ve questions from Section A and
one question from Section B.
Section A is weighted as 60% and Section B as 40%.
Write your candidate number (not your name) on the cover of each booklet.
Write legibly.
If you identify an error in this paper, please alert the Invigilator, who will notify
the Examiner. A general announcement will be made if the error is validated.
STATIONERY REQUIREMENTS
20 Page booklet x 1
Rough work pads
SPECIAL REQUIREMENTS TO BE SUPPLIED FOR THIS
EXAMINATION
Calculator - students are permitted to bring an approved calculator
You may not start to read the questions printed
on the subsequent pages of this question paper
until instructed that you may do so by the
Invigilator.
1 of 3
SECTION A
1. Consider the ARMA(1,1) model
yt = yt 1 + "t + " t 1 ; t = ::: 1; 0; 1; : : : ; T;
where "t N ID(0; 2 ): (i) What conditions are needed to ensure stationarity
and invertibility? (ii) Show that the autocorrelations decay exponentially
from lag one onwards. (iii) How you would compute the coe¢ cients, j ;
j = 0; 1; 2; :::; in the (in…nite) MA representation of the model?
2. Consider a model made up of a random walk trend and a nonstationary sea-
sonal component for a series observed twice a year, that is t = t 1 + !t;
2
where ! t is white noise with variance ! . How can the observations be made
stationary? Find the autocorrelation function and spectral density of the
stationary form of the model.
3. What method does Hamilton (2018) propose for extracting a cycle, ct ; from
a time series? When yt is a random walk, Hamilton’s …lter amounts to con-
structing ct by taking h-th di¤erences of yt in large samples. Find the auto-
correlation function of ct : Is ct invertible?
4. A rational distributed lag model is written as
B(L) 2
yt = xt + "t ; "t N ID(0; ); t = 1; :::; T;
A(L)
where A(L) and B (L) are polynomials in the lag operator and xt is an ob-
served exogenous variable. Show that A(L) = 1 L and B (L) = gives
the geometric distributed lag. What assumption are you making about the
parameter ? How would you estimate and ? What problems might arise
in generalizing the model by letting B (L) = + 1 L?
5. Suppose that
y2t = y1;t 2 + "t ; t = 1; :::; T:
where "t is white noise and y1t is generated by an MA(1) model with distur-
bance term, t ; that is uncorrelated with "t in all time periods. Derive the
cross-correlation function. What are the bene…ts of pre-whitening y1t ?
6. Write down an IGARCH model and outline its properties.
7. Compare and contrast the GARCH-t and Beta-t-EGARCH models.
2 of 3
SECTION B
B.1. (i) Obtain the exact log-likelihood function for the stationary Gaussian AR(1)
model
2
yt = yt 1 + "t ; "t N ID(0; ); t = :::: 1; 0; 1; : : : ; T
Show that an approximate maximum likelihood (ML) estimator of ; b; can
be obtained by regressing yt on yt 1 : What is the estimate, avar(b); of the
large sample variance of b given by this regression? Find the information
quantity for and hence write down the asymptotic distribution for b:
(ii) What can be said about the asymptotic distribution of b when yt is a
random walk? How would you test the null hypothesis that yt is a random walk
against the alternative that it is a stationary AR(1)?
(iii) How would you test the null hypothesis that yt is white noise against
the alternative that it is a random walk?
(iv) Obtain a recursion for zt = @"t =@ in the MA(1) model
2
y t = "t + " t 1 ; "t N ID(0; ); t = 1; : : : ; T;
with "0 = 0 and explain how you would use it to estimate :
(v) It can be shown that Avar(e); the asymptotic variance of the ML estimator
of ; is given by 2 =E(z 2 ): Find an expression for Avar(e) as a function of :
t
B.2. (a) Contrast the general formulation of a nonlinear parameter driven model
with the general formulation of an observation driven model. Why are
observation driven models usually easier to estimate? How would you
generate a set of observations, yt ; t = 1; :::; T; from an observation driven
model ?
(b) The log-logistic distribution has PDF
1
(y=')
p(y) = 2; 0 < y < 1; '; > 0:
' [(y=') + 1]
(i) Formulate a score-driven …rst-order dynamic model for the scale, ';
using an exponential link function, that is = ln ': What restrictions are needed
for stationarity? Comment on the shape of the score function.
(ii) Show that the score may be written as 2 bt ; where bt ; 0 bt 1;
has a standard uniform distribution. Hence show that the score has mean zero and
that the information quantity for is 2 =3:
(iii) Write down a condition that is su¢ cient to ensure the invertibility of
a strictly stationary nonlinear model. When the log-logistic score is standardized
by dividing by the information quantity, …nd a restriction on the parameters that
guarantees invertibility.
END OF PAPER
3 of 3
