ARMA model with implicit dependent variable

[Novi]

Hi all,
these days I am working with an ARIMA(12,1,2) model, which I estimated with an implicit dependent variable D(r), so my equation looks like:

equation e1.ls D(R) AR(1) AR(2) AR(3) AR(4) AR(5) AR(7) AR(9) AR(10) AR(12) MA(1) MA(2)

A new feature of eviews 7 seems to be that it's directly possible to forecast the level of the variable r, rather than the differenced variable d(r)...nevertheless I am trying to manually check the static forecasts and I keep coming up with different results from what eviews gives me...does anyone know what Eviews does EXACTLY to forecast the level of r from an equation expressed in differences?

What I do is:

Code: Select all

e1.fit(d) yhats1 genr rf_indirect_manual=na smpl @first 2007m8 rf_indirect_manual=r smpl 2007m8+1 @last rf_indirect_manual=rf_indirect_manual(-1)+yhats1

How come that my rf_indirect_manual is always different from the eviews forecast rf_1step? this is obtained as:

Code: Select all

e1.fit rf_1step

[EViews Gareth]

Code: Select all

create u 100 series y=nrnd series x=nrnd equation eq1.ls d(y) c x 'static eq1.fit yf_s eq1.fit(d) dyf_s smpl 2 @last series ym_s = y(-1) + dyf_s 'dynamic smpl @all eq1.forecast yf_d eq1.forecast(d) dyf_d smpl 1 1 series ym_d = y smpl 2 @last ym_d = ym_d(-1) + dyf_d show yf_s ym_s yf_d ym_d

[Novi]

Thanks for the prompt reply! :)

Anyway, what actually works in this simple specification with an exogenous variable x does not apply (at least for me) when I have a pure arma specification and no exogenous terms....

would you mind writing a similar example with equation eq1.ls d(y) c ar(1) ma(1) instead? that would clarify things a lot....

[EViews Gareth]

Code: Select all

rndseed 1 create u 100 series x=nrnd series y=nrnd equation e1.ls d(y) c ar(1) ma(1) e1.fit yf e1.fit(d) dyf smpl 2 @last series ym = y(-1) + dyf show yf ym

[Novi]

Ok....
But then why this thread:
http://forums.eviews.com/viewtopic.php?f=7&t=465 (in particular the final paragraph)

says something different regarding the forecasts from an arma model?
there it seems that the ma part also has to be taken into account while in your code it doesn't....

[EViews Gareth]

How you forecast an ARMA model is entirely different from how you treat the implicit dependent variable having done an ARMA forecast. You asked about the latter, not the former.

It doesn't matter how you got your forecast of D(Y), once you have it, figuring out Y is easy...

[Novi]

You are right, it's easy to get all problems mixed up when you work a lot! Thanks...