ARMA model building with AIC and SIC

[petimi]

Hi

I am trying to build an ARMA (5,5) model that will report the AIC and SIC values but can't get it.

I ran the program:

for !i=1 to 5
for !j=1 to 5
equation e1_!i_!j.ls y c ar(!i) ma(!j)
next
next

But the output it gives does not report the AIC and SIC values.

I would like the output to report just the AIC and SIC values

Please help me. :(

Many thanks

[EViews Gareth]

You should take a look at this post:
http://forums.eviews.com/viewtopic.php?f=5&t=220&p=633

[petimi]

Hi Gareth

Thanks for the reply.

I looked at that post before writing to you but still I could not write the appropriate program to run.

I have been on it for a long time with no result.

Please help?

[petimi]

Hi

Please I would like you to help me this problem:

In generating the ARMA (5,5) models with the program:

for !i=1 to 5
for !j=1 to 5
equation e1_!i_!j.ls y c ar(!i) ma(!j)
next
next

The output it generates is, take for instance ARMA (4,4), will be(AR4,MA4) values, but it is the full output i.e. (AR1, AR2, AR3, AR4, MA1, MA2, MA3, MA4) that I require, while also reporting the AIC and SIC values.

What adjustment should be made to the program? :?

Many thanks

[startz]

Hi

Please I would like you to help me this problem:

In generating the ARMA (5,5) models with the program:

for !i=1 to 5
for !j=1 to 5
equation e1_!i_!j.ls y c ar(!i) ma(!j)
next
next

The output it generates is, take for instance ARMA (4,4), will be(AR4,MA4) values, but it is the full output i.e. (AR1, AR2, AR3, AR4, MA1, MA2, MA3, MA4) that I require, while also reporting the AIC and SIC values.

What adjustment should be made to the program? :?

Many thanks

The problem is that

Code: Select all

ls y ar(2)

means

y=phi(2)y(-2)+epsilon

not
y=phi(1)y(-1)+phi(2)y(-2)+epsilon

If you really just want to go up to 5 lags, it'll be easier to copy and paste and run 25 regressions than to figure out the program.

[petimi]

Thanks for the reply.

How do I copy and paste and run the 25 regressions? :?

Thanks in advance

[startz]

Thanks for the reply.

How do I copy and paste and run the 25 regressions? :?

Thanks in advance

Open a program window and enter

Code: Select all

ls y c ar(1) ar(2) ar(3) ar(4) ar(5) ma(1) ma(2) ma(3) ma(4) ma(5)

Copy that line and paste it 24 times. Then go into the copies and delete higher order terms. Be sure that your options are set to allow more than one untitled equation.

[petimi]

Hi

Thanks for your help.

I followed the steps your suggested, when I clicked on run in the program window, it runs all the models but does not save them. It only saves the very last one i.e., ARMA(5,5).

[startz]

Hi

Thanks for your help.

I followed the steps your suggested, when I clicked on run in the program window, it runs all the models but does not save them. It only saves the very last one i.e., ARMA(5,5).

You probably need to look at Options/Window and Font options and unclick Allow only one untitled.

[petimi]

Thank you very much.

It worked! :)

[petimi]

Hi Startz

I posted this question in the past and was wondering if you could help me with it.

I'm currently trying to estimate the CKLS continuous time model in EVIEWS 6 using the same Nowman(1997) methodology, but keep getting an error.

The data I'm using is the 1 Month Interbank rate from August 2001 up to March 2009, at daily intervals. (I named this series 'r')

The Logl object I have declared in EVIEWS is the following:

@logl logl1

res = r - (exp(c(2)))*r(-1) - (c(1)/c(2))*(exp(c(2)) -1) // error term

var = (c(3)^2 / 2*c(2))*(exp(2*c(2)) -1)*(r(-1)^(2*c(4))) // variance term

logl1 = -(2*log(sqr(var)) + res^2 / var) // logl function

where c(1) is alpha, c(2) is beta, c(3) is sigma and c(4) is gamma

The syntax seem to pass ok after I click on 'Estimate', but a message pops up declaring, "Missing values in @LOGL series at current coefficients at observation 1" and the estimation procedure stops immediately. Is there an additional syntax I need to include?

Thanks in advance.

Petimi

[startz]

Hi Startz

I posted this question in the past and was wondering if you could help me with it.

I'm currently trying to estimate the CKLS continuous time model in EVIEWS 6 using the same Nowman(1997) methodology, but keep getting an error.

The data I'm using is the 1 Month Interbank rate from August 2001 up to March 2009, at daily intervals. (I named this series 'r')

The Logl object I have declared in EVIEWS is the following:

@logl logl1

res = r - (exp(c(2)))*r(-1) - (c(1)/c(2))*(exp(c(2)) -1) // error term

var = (c(3)^2 / 2*c(2))*(exp(2*c(2)) -1)*(r(-1)^(2*c(4))) // variance term

logl1 = -(2*log(sqr(var)) + res^2 / var) // logl function

where c(1) is alpha, c(2) is beta, c(3) is sigma and c(4) is gamma

The syntax seem to pass ok after I click on 'Estimate', but a message pops up declaring, "Missing values in @LOGL series at current coefficients at observation 1" and the estimation procedure stops immediately. Is there an additional syntax I need to include?

Thanks in advance.

Petimi

Just a guess, but probably r(-1) doesn't exist for the first observation. Mind you, most EViews adjust the smpl, so that may not be it. Try generating res and var as series and see what they look like.

[petimi]

Thanks for the response.

Was very helpful.