Converting AR MA terms to a series

[diggetybo]

Hello,

As I am using Eviews 6, I don't have a built-in function for viewing scaled/standardized coefficients.

To manually do this, I need to standardize the data, so my question is how do I code a series to be generated from my equation?

I tried the following, which is not working...

Code: Select all

series auto_reg = @eq01@ar(1)

Once all the terms are in a series, I can standardize them. As it stands though, my ARMA terms are only in the equation itself.

[EViews Gareth]

I'm not quite sure what you're after conceptually. There is no "ARMA" series.

[diggetybo]

Thanks for the chance to clarify,

I'm conceptually looking to compare the coefficients of my ARIMA model. Instead of using the partial derivative based coefficients, I want to see them in standardized form to obtain a unitless framework to compare the coefficients of the variables that would have otherwise been computed in various different units, which is not always ideal when making direct comparisons.

Transforming the pre-existing series is simple enough, the following code will standardize the data for a series called "var1":

Code: Select all

series var1_standardized = (var1 - @MEAN(var1)) / @STDEVS(var1)

However, for AR/MA terms, as you've said, there are no such series. So how can I proceed with standardizing the data?

Recall that theoretically, an ARMA(p,q) model is computed as:

ARMA Model.png (32.69 KiB) Viewed 8675 times

So the fact that eviews has the AR/MA coefficients as part of the estimation output, it should imply that it has computed the above for every time t, except for the first few, depending on how many ARMA terms there are, i.e. for AR(2) there will not be anything for t=1 or t=2.

On a separate note, I'm not 100% sure what units would even accompany ARMA terms, delta y?

Nonetheless, there should be a series somewhere. I just don't know how to code it into existence, from which point I can standardize it.

Let me know if you're not clear, I can try to reiterate.

[EViews Gareth]

That's an econometric issue, rather than an EViews one, so I'm not going to comment.

However I will point out that you should be careful on your definition of an ARMA process. If Y is the dependent variable in your regression, then your formula for the ARMA model is not right. It is not Y that follows the ARMA process, it is the error term that does.

[startz]

If I may be allowed to chime in, in what may be a most unhelpful manner.

First, in an ARMA model all the variables have the same units of measurement, so I don't understand how standardized coefficients would be any different from the usual coefficients.

Second, y does follow an ARMA process in the picture given.

[diggetybo]

Well ok,

thank you eviews gareth and startz.

I may have overlooked that.

I suppose it is not necessary to view the ARMA terms at every time t to create a series. If anyone knows if that's possible though, please post the code here for future reference.

thanks again.

[EViews Gareth]

Second, y does follow an ARMA process in the picture given.

It does. But if you have an equation:

Code: Select all

Y C X AR(1 to p) MA(1 to q)

then Y does not follow the ARMA process given by the formula above. The ARMA process is the error term, not the dependent variable.

[startz]

Well ok,

thank you eviews gareth and startz.

I may have overlooked that.

I suppose it is not necessary to view the ARMA terms at every time t to create a series. If anyone knows if that's possible though, please post the code here for future reference.

thanks again.

What do you mean by "the ARMA terms?" The ARMA coefficients are displayed for you. In the example you give--but note Gareth's caveat--the "ARMA series" is just the forecast value of y.

[diggetybo]

I just meant converting the AR(1) into its formula derived numbers. I'm willing to drop that issue for now, I may have wrote 10 pages about nonsense :(

I know that AR/MA terms can't really be interpreted via coefficients, but rather impulse response and other dynamics. But I'm looking for a way to compare the AR/MA portion of the R squared value with the other coefficients.

You said that in an ARMA model, all units are the same, even if the model has non-ARMA terms? Like in the eviews tutorial here:

http://www.eviews.com/Learning/timeseries_c.html

If we compare the coefficients from slides 19 and 21, they have changed notably.

As this is an econometric issue, maybe we could move this post to the econometric discussion section, and you don't have to help, but I'd appreciate a quick tip.

What are the units for the non ARMA terms here on slide 21? Log(m1), Log(cpi), Log(ip)? The functional form is level_log, but does that interpretation still hold? unit change in y given a percentage change in x? Or have the ARIMA terms irreversibly changed that?

I noticed most papers don't comment much on ARMA coefficients, but I've always assumed there was a way to compare ARMA terms with the other explanatory variables.

Please help if u dont mind

thanks a million

[startz]

Perhaps the following will help. You can run the model with and without ARMA terms and with and without the other explanatory variables and compare the R^2.

[diggetybo]

Hey guys,

I hate to keep drawing this out.

So I will be as concise as possible.

The picture I uploaded may have caused some confusion. My apologies for that. Also, As startz pointed out, ARIMA models are atheoretical and do not always warrant meaningful interpretations of the coefficients. Which scared me away from my initial question about coding the AR and MA into a series, I thought I had a fundamentally flawed model. Which would make what eviewsGareth said all the more true, this is more of an econometric issue.

However, I ran this buy my professor. He pointed out that my model of...

Code: Select all

y x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 ar(1) ma(1) ma(2)

...still has valid coefficients for the non-arma terms, and they still retain the partial derivative interpretation as in OLS.

So my question for eviews are as follows: for a model like mine, what happens in eviews 7 and 8 if we were to click...

View ..... Coefficient Diagnostics ....... Scaled Coefficients??

Question 1: Does eviews not standardize everything? Namely, what about the ARMA terms, would they be 'standardized' or be the same? This is my goal, to understand how eviews renders AR MA terms.

Question 2: Is generating a series from the AR MA terms really that nonsensical? As I've said, one would assume that those coefficients are calculated from a series, just like any other coefficient be it x1 or x9 or what have you. That is to say, AR(1) is simply resid(-1) or y(-1) depending on your specification. The point being, again, I have eviews 6, and if the answer to question one is "yes" they are standardized, then I have to manually standardize the data, meaning I need all the series.

Question 3: Alternatively, would eviews automatically standardize the AR MA terms if I have standardized all the other series from which the AR MA terms are automatically computed?

So it all boils down to, with my older version of eivews, I'm not sure how much work I have to do manually, and what eviews does for me.

Let me know if there is any confusion, I can try to be more clear.

Thank you for your commitment.

[EViews Gareth]

1) EViews doesn't allow standardized coefficients to be computed if you have ARMA terms (for precisely this reason).
2) Yes.
3) See 1).

[diggetybo]

Very well,

This issue has been put to rest. Thanks again eviewsGareth.

:D