Pavel Nazarov
Statements
“I have read and understood the instructions and policies for this assignment, including those
related to the use of AI and other external resources.”
“I completed this assignment independently.”
“I did not use AI on this assignment.”
Question 1 (a)
Apply the backshift operator B to the given AR(3) equation to represent it as an autoregressive polynomial:
𝜙(B) = 1 − 0.6B − 0.37B! + 0.21B"
Factor the polynomial:
𝜙(z) = 0.21z " − 0.37z ! − 0.6z + 1
Find roots of 𝜙(z) using the following R script:
Therefore roots z# are:
z$ = 1.4286, z! = −1.6667, z" = 2
Convert roots to factors of the form (1 – ri * z) where r# = 15z# and convert 𝜙(z) into its factored form:
1 1 1
𝜙(z) = 61 − z7 61 − z7 61 − z7
1.4286 −1.6667 2
𝜙(z) = (1 − 0.7z)(1 + 0.6z)(1 − 0.5z)
Causality:
Since |z$ |, |z! | and |z" | are all greater than 1, they lie outside of the unit circle, which is a necessary condition for a
process to be causal. The process is causal if and only if all roots of 𝜙(z) = 0 lie outside of the unit circle.
Question 1 (b)
Given that the factorization yields 𝜙(z) = (1 − 0.7z)(1 + 0.6z)(1 − 0.5z), and given that roots are r1 = 0.7,
r2 = – 0.6 and r3 = 0.5, and given that the process is causal, autocorrelation function 𝜌% (h) = c$ r$&' + c! r!&' + c" r"&'
will take the form of:
𝜌% (h) = c$ 0.7' + c! (−0.6)' + c" 0.5' , h≥0
To find c$ , c! , c" , we will use a two-step process. Step 1 is to use the recursive AR(p) Yule-Walker Equation
𝜌% (h) = 𝜙$ 𝜌% (h − 1)+. . . + 𝜙( 𝜌% (h − p) for h = 1, . . . , p − 1 to solve for 𝜌% (h). Step 2 is to use the values for
𝜌% (0). . . 𝜌% (p − 1) to solve for c$ . . . c( .
Pavel Nazarov
Step 1.
CoeYicients 𝜙$ = 0.6, 𝜙! = 0.37 and 𝜙" = −0.21 are based on an AR(3) process described by
X) = 0.6X)&$ + 0.37X)&! − 0.21X)&" + 1.
Given that 𝜌% (h) = 𝜌% (−h) for all h, 𝜌% (h) for h = 1, 2 becomes:
𝜌% (1) = 𝜙$ 𝜌% (0) + 𝜙! 𝜌% (1) + 𝜙" 𝜌% (2)
D
𝜌% (2) = 𝜙$ 𝜌% (1) + 𝜙! 𝜌% (0) + 𝜙" 𝜌% (1)
With 𝜌% (0) = 1, the above simplifies to:
𝜌 (1) = 𝜙$ + 𝜙! 𝜌% (1) + 𝜙" 𝜌% (2)
D %
𝜌% (2) = 𝜙$ 𝜌% (1) + 𝜙! + 𝜙" 𝜌% (1)
𝜌 (1) − 𝜙! 𝜌% (1) = 𝜙$ + 𝜙" 𝜌% (2)
D %
𝜌% (2) = (𝜙$ + 𝜙" )𝜌% (1) + 𝜙!
𝜌 (1)(1 − 𝜙! ) = 𝜙$ + 𝜙" 𝜌% (2)
D %
𝜌% (2) = (𝜙$ + 𝜙" )𝜌% (1) + 𝜙!
𝜌 (1)(1 − 𝜙! ) − 𝜙" 𝜌% (2) = 𝜙$
D %
−(𝜙$ + 𝜙" )𝜌% (1) + 𝜌% (2) = 𝜙!
Convert the system of equations to a matrix form:
1 − 𝜙! −𝜙" 𝜌% (1) 𝜙
E FE F = E $F
−(𝜙$ + 𝜙" ) 1 𝜌% (2) 𝜙!
Substitute the values of coefficients and run the following script to solve it:
0.63 0.21 𝜌% (1) 0.6
G IE F=G I
−0.39 1 𝜌% (2) 0.37
𝜌% (1) = 0.7336
𝜌% (2) = 0.6561
Step 2:
For h = 0, 1, and 2, 𝜌% (h) = c$ 0.7' + c! (−0.6)' + c" 0.5' shall take the forms of:
𝜌% (0) = 1 = c$ + c! + c"
𝜌% (1) = 0.7336 = 0.7c$ + (−0.6)c! + 0.5c"
𝜌% (2) = 0.6561 = 0.49c$ + 0.36c! + 0.25c"
Pavel Nazarov
Convert to matrix form:
1 1 1 c$ 1
J 0.7 −0.6 0.5 K Jc! K = J0.7336K
0.49 0.36 0.25 c" 0.6561
Use R Script to solve the system of equations:
𝒄𝟏 = 𝟏. 𝟔𝟓𝟏𝟖
𝒄𝟐 = 𝟎. 𝟎𝟖𝟕𝟗
𝒄𝟑 = −𝟎. 𝟕𝟑𝟗𝟕
𝝆𝑿 (𝒉) = 𝟏. 𝟔𝟓𝟏𝟖 × 𝟎. 𝟕𝒉 + 𝟎. 𝟎𝟖𝟕𝟗 × (−𝟎. 𝟔)𝒉 + (−𝟎. 𝟕𝟑𝟗𝟕) × 𝟎. 𝟓𝒉 , h≥0
Question 1(c)
The answers from the manually computed formula and ARMAacf built-in function match exactly.
Pavel Nazarov
Question 2(a)
The ARMA(2,1) process above has:
AR coefficients:
𝜙$ = 0.2
𝜙! = 0.08
MA coefficient:
𝜃$ = 0.7
To test for causality, we need to focus on the AR polynomial to find its roots.
AR Polynomial is:
𝜙(z) = 1 − 𝜙$ z − 𝜙! z !
𝜙(z) = 1 − 0.2z − 0.08z !
Solve for 𝜙(z) = 0
0.08z ! + 0.2z − 1 = 0
−0.2 ± Z0.2! − 4 × 0.08 × (−1)
z=
2 × 0.08
−0.2 + 0.6 −0.2 − 0.6
z$ = = 2.5; z! = = −5
0.16 0.16
Calculate the reciprocal roots for the factored form of (1 − r$ z)(1 − r! z):
1 1
r$ = = 0.4; r! = = −0.2
2.5 −5
Factored form of the AR polynomial:
𝝓(𝒛) = (𝟏 − 𝟎. 𝟒𝒛)(𝟏 + 𝟎. 𝟐𝒛)
The process is causal because moduli of both AR roots |z$ | and |z! | > 1, therefore they lie outside the unit circle,
which is a requirement for the process to be causal. The process is causal if and only if all roots of 𝜙(z) = 0 lie
outside of the unit circle.
The MA polynomial is:
𝜽(𝒛) = 𝟏 + 𝟎. 𝟕𝒛
The root of the MA polynomial z = – 1 / 0.7 = – 1.43. with modulus of z being greater than one, this ARMA(2,1) is also
invertible.
Pavel Nazarov
Question 2(b)
X) = 𝜓/ W) + 𝜓$ W)&$ + 𝜓! W)&! . ..
1
𝜓(z) = 𝜓/ + 𝜓$ z + 𝜓! z . . . = f 𝜓0 z 0
!
02/
3(5)
For a causal ARMA(p, q) process the weights 𝜓0 come from a relationship 𝜓(z) = 7(5).
1
1 + 0.7𝑧
f 𝜓0 z 0 =
1 − 0.2z − 0.08z !
02/
(1 − 0.2z − 0.08z ) f 𝜓0 z 0 = 1 + 0.7𝑧
!
02/
(1 − 0.2z − 0.08z ! )(𝜓/ + 𝜓$ z + 𝜓! z ! . . . ) = 1 + 0.7𝑧
(𝜓/ + 𝜓$ z + 𝜓! z ! . . . ) + ( −0.2𝜓/ z − 0.2𝜓$ z ! − 0.2𝜓! z " . . . ) + ( −0.08𝜓/ z ! − 0.08𝜓$ z " − 0.08𝜓! z 8 . . . ) = 1 + 0.7𝑧
Polynomial equality requires coefficient-by-coefficient equality. If a/ + a$ z + a! z ! . . . = b/ + b$ z + b! z ! . .. then
a/ = b/ , a$ = b$ , a" = b" . ..
Therefore, we need to match coefficients by powers of z:
For z / , gather coefficients from the left-side and ride-side of the equality:
𝝍𝟎 = 𝟏
For z$ , … :
𝜓$ z − 0.2𝜓/ z = 0.7𝑧
𝜓$ − 0.2 = 0.7
𝝍𝟏 = 𝟎. 𝟗
For z ! , … :
𝜓! z ! − 0.2𝜓$ z ! − 0.08𝜓/ z ! = 0
𝜓! − 0.2(0.9) − 0.08 = 0
𝝍𝟐 = 𝟎. 𝟐𝟔
For z " , … :
𝜓" z " − 0.2𝜓! z " − 0.08𝜓$ z " = 0
𝜓" − 0.2(0.26) − 0.08(0.9) = 0
𝝍𝟑 = 𝟎. 𝟏𝟐𝟒
Pavel Nazarov
For z 8 , … :
𝜓8 z 8 − 0.2𝜓" z 8 − 0.08𝜓! z 8 = 0
𝜓8 − 0.2(0.124) − 0.08(0.26) = 0
𝝍8 = 𝟎. 𝟎𝟒𝟓𝟔
For z : , … :
𝜓: z : − 0.2𝜓8 z : − 0.08𝜓" z : = 0
𝜓: − 0.2(0.0456) − 0.08(0.124) = 0
𝝍: = 𝟎. 𝟎𝟏𝟗𝟎𝟒
The above sequence is consistent with the general recursion formula for an ARMA(p, q):
𝜓0 = 𝜃0 + 𝜙$ 𝜓0&$ + 𝜙! 𝜓0&! +. . . +𝜙( 𝜓0&(
ARMAtoMA R Script precisely confirms the above calculations:
Question 2(c)
Use the following sequence of steps to compute the autocorrelation (ACF) function 𝜌% (h) for an ARMA(p, q)
process, where p = 2 and q = 1.
Step 1: compute the autocovariance function (ACVF)
Step 2: use the covariance values 𝛾 to calculate correlation values 𝜌
Step 3: find the general formula and compute the constants c
Step 4: devise the final general formula 𝜌% (h) for h ≥ 1
Pavel Nazarov
Step 1
For an ARMA(p, q) process the ACVF satisfies the following conditions:
For h = 0:
( ;
!
𝛾(0) = f 𝜙# 𝛾(i) + 𝜎 f 𝜃0 𝜓0
#2$ 02/
For 1≤ ℎ ≤ 𝑞:
( ;
!
𝛾(h) = f 𝜙# 𝛾(h − i) + 𝜎 f 𝜃0 𝜓0&'
#2$ 02'
For h > q:
(
𝛾(h) = f 𝜙# 𝛾(h − i)
#2$
Use the given and computed values of 𝜙, 𝜃 and 𝜓 to set up the system of ACVF equations with p = 2 and q = 1.
For h = 0:
𝛾(0) = 𝜙$ 𝛾(1) + 𝜙! 𝛾(2) + 𝜎 ! 𝜃/ 𝜓/ + 𝜎 ! 𝜃$ 𝜓$
𝛾(0) = 0.2𝛾(1) + 0.08𝛾(2) + 𝜎 ! (1 × 1 + 0.7 × 0.9)
𝜸(𝟎) = 𝟎. 𝟐𝜸(𝟏) + 𝟎. 𝟎𝟖𝜸(𝟐) + 𝟏. 𝟔𝟑𝝈𝟐
For h = 1:
𝛾(1) = 𝜙$ 𝛾(0) + 𝜙! 𝛾(1) + 𝜎 ! 𝜃$ 𝜓/
𝛾(1) = 0.2𝛾(0) + 0.08𝛾(1) + 𝜎 ! 0.7 × 1
𝟎. 𝟗𝟐𝜸(𝟏) = 𝟎. 𝟐𝜸(𝟎) + 𝟎. 𝟕𝝈𝟐
For h = 2:
𝛾(2) = 𝜙$ 𝛾(1) + 𝜙! 𝛾(0)
𝜸(𝟐) = 𝟎. 𝟐𝜸(𝟏) + 𝟎. 𝟎𝟖𝜸(𝟎)
Step 2
𝛾(0)5 𝛾(h)
Define g = 𝜎 ! . Autocorrelation is a normalized autocovariance function: 𝜌% (h) = s𝛾(0).
Convert 𝛾(2) to ACF form [Equation 1]:
𝛾(2) = 𝜙$ 𝛾(1) + 𝜙! 𝛾(0) ; divide both sides by 𝛾(0)
𝜌% (2) = 𝜙$ 𝜌% (1) + 𝜙!
𝝆𝑿 (𝟐) = 𝟎. 𝟐𝝆𝑿 (𝟏) + 𝟎. 𝟎𝟖
Pavel Nazarov
Convert 𝛾(1) to ACF form [Equation 2]:
0.92𝛾(1) = 0.2𝛾(0) + 0.7𝜎 ! ; divide both sides by 𝛾(0)
0.7𝜎 !
0.92𝜌% (1) = 0.2 +
𝛾(0)
0.7𝜎 !
0.92𝜌% (1) = 0.2 +
g × 𝜎!
0.7
0.92𝜌% (1) = 0.2 +
g
0.2 0.7
𝜌% (1) = +
0.92 0.92g
𝟎. 𝟐𝒈 + 𝟎. 𝟕
𝝆𝑿 (𝟏) =
𝟎. 𝟗𝟐𝒈
Convert 𝛾(0) to ACF form [Equation 3]:
𝛾(0) = 0.2𝛾(1) + 0.08𝛾(2) + 1.63𝜎 ! ; divide both sides by 𝛾(0)
1.63𝜎 !
1 = 0.2𝜌% (1) + 0.08𝜌% (2) +
𝛾(0)
1.63𝜎 !
1 = 0.2𝜌% (1) + 0.08𝜌% (2) +
g × 𝜎!
1.63
1 = 0.2𝜌% (1) + 0.08𝜌% (2) +
g
Recursively solve for g, 𝝆𝑿 (𝟏) 𝒂𝒏𝒅 𝝆𝑿 (𝟐) using Equations 1 and 2:
1.63
1 = 0.2𝜌% (1) + 0.08(𝟎. 𝟐𝝆𝑿 (𝟏) + 𝟎. 𝟎𝟖) + ⟸ subsitute 𝜌% (2)
g
1.63
1 = 0.2𝜌% (1) + 0.016𝜌< (1) + 0.0064 +
g
1.63
0.9936 = 0.216𝜌% (1) +
g
𝟎. 𝟐𝒈 + 𝟎. 𝟕 1.63
0.9936 = 0.216 6 7+ ⟸ subsitute 𝜌% (1)
𝟎. 𝟗𝟐𝒈 g
0.0432g + 0.1512 1.63
0.9936 = +
0.92g g
0.16435 1.63
0.9936 = 0.04696 + +
g g
1.79435
0.94664 =
g
𝒈 = 𝟏. 𝟖𝟗𝟓𝟔
0.2𝑔 + 0.7
𝝆𝑿 (𝟏) = = 𝟎. 𝟔𝟏𝟖𝟖
0.92𝑔
𝝆𝑿 (𝟐) = 0.2𝜌< (1) + 0.08 = 𝟎. 𝟐𝟎𝟑𝟔
Step 3
General solution for 𝜌< (ℎ) for h ≥ 1 is: 𝜌< (ℎ) = c$ r$' + c! r!' . Using roots calculated in question 2(a),
𝜌< (ℎ) = c$ 0.4' + c! (−0.2)'
To calculate the constants, solve a system of equations 𝜌< (ℎ) for h = 1 and h = 2:
𝜌< (1) = c$ 0.4$ + c! (−0.2)$ = 0.6188
𝜌< (2) = c$ 0.4! + c! (−0.2)! = 0.2036
Pavel Nazarov
0.4 −0.2 c$ 0.6188
G IG I = G I
0.16 0.04 c! 0.2036
c$ = 1.364
c! = −0.366
Step 4: ACF formula
𝝆𝑿 (𝒉) = 𝟏. 𝟑𝟔𝟒 × 𝟎. 𝟒𝒉 − 𝟎. 𝟑𝟔𝟔 × (−𝟎. 𝟐)𝒉 𝒇𝒐𝒓 𝒉 ≥ 𝟏
Question 2(d)
ARMAacf matches the manually computed formula, save for minute rounding error of the magnitude of 10^-4.