Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
Statistics - Time Series Level H
This paper consists of 5 pages and contains 3 question(s).
Candidates should attempt all questions.
Question 1 20 marks
Question 2 20 marks
Question 3 20 marks
Total 60 marks
The following material is made available to you:
Statistical tables∗
Statistical Tables
Formula sheet
“An electronic calculator may be used provided that it is allowed under the School of
Mathematics and Statistics Calculator Policy. A copy of this policy has been distributed
to the class prior to the exam and is also available via the invigilator.”
NOTE: Candidates should attempt all 3 questions.
1
CONTINUED OVERLEAF/
1. (a) Given a real data set, how could you determine if the correlation structure would
be well modelled by an MA(q) process? [2 MARKS]
(b) Autoregressive processes of orders 1, 2, and 3 were fitted to a time series with 120
observations, and the model output from R is shown below.
arima(x, order = c(1, 0, 0))
Coefficients:
ar1 intercept
0.9363 287.9945
s.e. 0.0160 38.0214
arima(x, order = c(2, 0, 0))
Coefficients:
ar1 ar2 intercept
0.7597 0.1901 289.9881
s.e. 0.0439 0.0441 46.6512
arima(x, order = c(3, 0, 0))
Coefficients:
ar1 ar2 ar3 intercept
0.7415 0.1184 0.0953 290.6858
s.e. 0.0445 0.0553 0.0447 51.6608
What type of AR(p) process is appropriate here? That is, what is the value of p?
Explain your answer. [3 MARKS]
(c) Consider the following AR(1) and MA(1) processes A and B.
A : Xt = 0.3Xt−1 + Zt,
B : Xt = 0.6Zt−1 + Zt.
i. Describe how correlation is induced between Xt and Xt−1 for processes A and
B. [2 MARKS]
ii. For each process, what is the value of ρ2, the lag two autocorrelation coeffi-
cient?
[2 MARKS]
(d) Consider the time series process
Xt = 0.7Zt−1 − 0.25Zt−2 + Zt,
2
CONTINUED OVERLEAF/
where Zt is a purely random process with E [Zt] = 0 and Var [Zt] = σ2z .
i. What type of process is defined here? [1 MARK]
ii. Is the process stationary, invertible, or both? Explain your answer.
[4 MARKS]
(e) Consider the model
Xt = β0 + β1t+ β2t
2 + λ1Zt−1 + λ2Zt−2 + Zt,
where β0, β1, β2, λ1 and λ2 are all non-zero and finite, and Zt is a purely random
process with E [Zt] = 0 and Var [Zt] = σ2z .
i. Calculate the mean and variance of Xt. Is it weakly stationary? Explain your
answer. [4 MARKS]
ii. Let Yt = ∇Xt be the first order difference of Xt. Write out Yt as a function
of Zt. [2 MARKS]
2. (a) Consider the time series process Xt given by
Xt = β0 + β1t+ Yt,
where Yt is a weakly stationary AR(1) process with lag one coefficient α = 0.7.
i. What would be a naive way to estimate β1, assuming that the error process
Yt is independent? How could this be done in R? [1 MARK]
ii. Give one drawback of the naive approach from i. in the estimation of β1
compared with a more robust approach that allows for the correlation present
in Yt. How could a more robust approach be performed in R? [2 MARKS]
iii. As part of a simulation study you may want to assess parameter estimation.
Detail two standard metrics that can be used to measure the quality of pa-
rameter estimates.
[2 MARKS]
(b) Consider the time series process given by
Xt = exp
(
β0 + β1t+ β2t
2 + St + Zt
)
,
where Zt is a purely random process with E [Zt] = 0 and Var [Zt] = σ2z , and St is
a seasonal component with period d, such that St = St+d. Show how this process
can be made weakly stationary. [5 MARKS]
3
CONTINUED OVERLEAF/
(c) The autocorrelation function (ACF) and partial autocorrelation function (PACF)
have been computed for a time series with 700 observations and are shown in the
table below.
Lag τ 1 2 3 4 5 6 7 8 9 10
ACFτ 0.57 0.42 0.28 0.10 0.08 0.03 -0.01 0.01 0.01 -0.01
PACFτ 0.76 0.68 0.12 0.06 0.05 0.03 0.020 0.01 0.01 0.01
Would you prefer an AR(p) or an MA(q) model for this process? If so, what would
be the corresponding order of the chosen model? [3 MARKS]
(d) Given below is R code to simulate time series data.
time <- 1:500
beta0 <- 0.3
beta1 <- 0.55
Z.sd <- 10
Y <- arima.sim(list(order=c(1,0,0), ar=0.6), n=500, sd=Z.sd)
X <- beta0 + beta1 * time + Y
Write out the time series model in clear mathematical notation. Provide details
on any trend, seasonality, or short-term correlation induced by the model.
[7 MARKS]
3. (a) Consider the ARIMA(1, 1, 0) process. Is this a stationary process? [2 MARKS]
(b) Two ARIMA processes are given below:
A : ARIMA(3, 2, 1)
B : ARIMA(1, 0, 2)
Write out both processes A and B in backshift notation. Hence write out the
model equations for both processes, expressing Xt in terms of Zt and previously
occurring quantities. [4 MARKS]
(c) Consider the time series model
Xt = β0 + β1t+ β2t
2 + Zt t = 1, . . . , 30,
where Zt is a purely random process with E [Zt] = 0 and Var [Zt] = σ2z . This
model was fitted to data with 30 time points which provided estimated regression
parameters βˆ0 = 4, βˆ1 = 0.8, βˆ2 = 0.6, and an estimated error variance of σˆ
2
z = 1.2.
i. Calculate the one-step ahead prediction x30(1) for time t = 31. [2 MARKS]
4
CONTINUED OVERLEAF/
ii. Calculate the 95% prediction interval for x30(1). [6 MARKS]
(d) Consider the AR(2) model
Xt = 0.75Xt−1 − 0.125Xt−2 + Zt,
where Zt is a purely random process with E [Zt] = 0 and Var [Zt] = σ2z = 2, and
xn = −1, xn−1 = 3, and xn−2 = 2. Calculate the one, two and three steps ahead
forecasts and their associated error variances. [6 MARKS]