Scribd
Upload
73 views7 pages

Lab: ARMA (1, 1) Process: T T T T

arima

Uploaded by

som
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
73 views7 pages

Lab: ARMA (1, 1) Process: T T T T

arima

Uploaded by

som
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Lab: ARMA(1, 1) Process

In this lab we consider an ARMA(1, 1) process of the form:

yt = a1 yt 1 + t + 1 t 1
Where t is a white noise process, mean zero and variance 2.
1. Use the ARMA11 worksheet to generate 20 observations from an ARMA(1, 1)
process with parameters a1 = 0.5, 1 = 0.5. Examine several instances of the
process on the chart provided.
2. Use the Yule -Walker equations to derive the autocovariance terms s = E(yt , yt-s).
Hence derive the ACF terms t for t = 1, 2, . . . ,20
3. Use the ACF function to compute the sample autocorrelations. Compare the
theoretical and sample ACF's on the correlogram provided.
The ACF function has two parameters ACF(Y, L) where Y is the range of time
series observations and L is the lag parameter. It returns the result l.
4. Use the PACF function to compute the theoretical and estimated partial
autocorrelations and compare them on the correlogram provided.
The PACF has two parameters PACF(ACF, L) where ACF is the range of actual
(or theoretical) autocorrelations and L is the lag parameter. The function returns
the result LL, the partial autocorrelation coefficient at lag L.
5. Experiment with different values of the process coefficients and examine the
effects on the form of the ACF and PACF.

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 1

Solution: ARMA(1, 1) Process


1. The layout of the lab worksheet ARMA11 is shown below.

The white noise process t is


generated using the Excel
RAND() function (which
generates a random number
between zero and 1), and then
converting this to a standard
normal variate using the
NORMSINV function.

The process yt is generated using the


formula:
yt = a1 yt-1 + t + 1 t-1
Where a = = 0.5

Enter the formula for the first white noise term 1 in cell C8 as
NORMSINV(RAND()). Copy this formula down into the remaining cells in the
column, corresponding to time periods 2 through 20.
In cell D8, enter set the formula for the first process observation y1 = D8. Then, in
cell D9, enter the Excel formula for the second term as follows:
= a*D8 + C9 + b*C8
Copy this formula down into the remaining cells in this column. The time series chart
is drawn automatically.
By pressing the function key F9, you can generate different instances of the process.
Try various values of the parameters a and (both positive and negative).
A typical example is shown below.

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 2

ARMA(1, 1) Process y t = ay t-1 + t + t-1


2.0
1.5
1.0
0.5
0.0
-0.5 1

11

16

-1.0
-1.5
-2.0

2. We set up the Yule-Walker equations as follows:


0 = E(yt yt)

= a1 E(yt-1 yt)+ E(t yt) + 1 E( t-1 yt)


= a11 + 2 + 1[t-1(a1yt -1 + t + 1t-1)]
= a1 1 + 2 + 1(a1+ 1 )2

1 = E(yt yt -1)

= a1E(yt-1 yt -1)+ E(t yt -1) + 1 E( t-1 yt-1)


= a1 0 + 1 2
.
.
.
= a1E(yt-1 yt -s)+ E(t yt -) + 1 E(t-1 yt-s)
= a1 s-1

s = E(yt yt-s)

Solving these equations simultaneously for 0 and 1 yields:


0 =

1 + 12 + 2a1 1 2

(1 a12 )

Hence, 1 =

and

1 =

(1 + a1 1 )( a1 + 1 ) 2

(1 a12 )

(1 + a1 1 )(a1 + 1 )
and s = a1s -1 for s > 1
(1 + 12 + 2 a1 1 )

3. With parameters a1 = 0.5, 1 = 0.5 the theoretical ACF is as follows:


1 = (1 + 0.52)(0.5 + 0.5) / (1 + 0.52 + 2(0.52 )) = 0.714
In the worksheet the formula is entered into cell E8 as follows:
=(1+a*b)*(a+b)/(1+b^2+ 2*a*b)
Subsequent autocorrelations are calculated using the relationship s = a1s-1.
Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 3

Hence, 2 = 0.5 1 = 0.357


The formula is entered in cell C9 as: =a*E8. This formula is copied down into the
remaining cells in the column.
The sample autocorrelations are calculated using the ACF function. Click the
Function Wizard from the menubar and select User Defined functions, ACF as
shown in the screenshot below.

The parameters used in the function are the range of values yt in cells D8 to D27,
and the time lag T in the corresponding cells B8 to B27. The first value is
calculated in cell F8, as shown below.

The formula in cell F8 should be copied down to the remaining cells in the column
to calculate the autocorrelations for T = 2 to 20.

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 4

The ACF chart is automatically updated, showing both the theoretical and sample
ACF function values. An example is shown below.
ACF for ARMA(1,1) Process
0.80
0.60
0.40
Theoretical

0.20

Estimated

0.00
-0.20

6 7

9 10 11 12 13 14 15 16 17 18 19 20

-0.40
Lag

4. The theoretical partial autocorrelation coefficients are given by:

11 = 1 , 22 = (2 - 12) / (1 - 1 2)
and
s 1

ss =

s s 1 s j
j =1
s 1

1 s 1 j
j =1

sj = s 1, j ss s 1,s j

j = 1, 2, ...s 1

Hence, 11 = 1 = 0.714 ,
and 22 = (0.357 - 0.7142) / (1 - 0.714 2) = -0.313
Thereafter the partial autocorrelations are derived iteratively:

21 = 11 - 22 11 = 0.714 - (-0.313)*0.714 = 0.9375

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 5

Hence
2

33 =

3 2, j 3 j
j =1
2

1 2, j j
j =1

= [0.179 - (0.9375)(0.357)-(-0.313)(0.714)]/ [1 - (0.9375)(0.714)-(-0.313)(0.357)]


= 0.152
We can use the PACF function to calculate the theoretical PACF coefficients,
using the theoretical ACF as the first input parameter. Click the Function Wizard
and selection user defined functions, PACF. The input parameters are as shown
below:

Likewise we can calculate the estimated PACF, this time using the estimated ACF
as the first input parameter.
The resulting PACF correlogram is shown below.

PACF for ARMA(1,1) Process


1.000
0.500
Theoretical

0.000
-0.500

9 10 11 12 13 14 15 16 17 18 19 20

Estimated

-1.000
Lag

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 6

5. With parameter a1 < 0, the ACF follows an oscillating decay pattern, like the one
below:
ACF for ARMA(1,1) Process
0.40
0.30
0.20
0.10
0.00
-0.10
-0.20
-0.30
-0.40
-0.50

Theoretical
1

2 3

7 8

9 10 11 12 13 14 15 16 17 18 19 20

Estimated

Lag

With parameter a1 > 0, the ACF follows a geometric decay pattern, with positive
or negative coefficients depending on whether (a1 + 1) > 0 or < 0. The rate of
decay is governed by the coefficient a1. The more positive the value, the slower
the decay.
An example is shown below:
ACF for ARMA(1,1) Process
0.30
0.20
0.10
Theoretical

0.00
1

-0.10

2 3

7 8

9 10 11 12 13 14 15 16 17 18 19 20

Estimated

-0.20
-0.30
Lag

With parameter 1 < 0, the PACF follows an exponential decay pattern, with sign
= Sign (a1 + 1), like the one below. With parameter 1 > 0, the PACF follows an
oscillating decay pattern (see example with 1 = 0.5).
PACF for ARMA(1,1) Process
1.00
0.50

Theoretical
Estimated

0.00
1

9 10 11 12 13 14 15 16 17 18 19 20

-0.50
Lag

Copyright 1999-2001 IISEC

ARMA(1, 1) Process

Page 7

You might also like