AR MA terms as explanatory variables

[Alexx]

Hi,
sorry for asking a really basic question, but I just have to:
If I want to estimate a series (say GDP) as an arma(1,1) process, I just go ahead and run "gdp c ar(1) ma(1)". This is clear.
Now if I want to estimate gdp as a function of its own lag, an ma term and say a further explanatory variable (consumer confidence), i.e. gdp=c+a1*gdp(-1)+a2*u(t-1)+a3*CC+u(t) what do I do?
Tellling eviews to run "gdp c ar(1) ma(1) CC" obviously is not the way to do it (on the basis of the user guide explanation).
Should I then run "gdp c gdp(-1) ar(-1) CC" ??


Am I the only one who thinks this is really confusing? Textbooks generally seem to suggest that an ARMA model is one, where the dependent variable X (whatever it is) is a function of a constant, its own lag and one or more lags of the error term (ut-1 etc). Yet eviews seems to define this completely differently. (AR and MA terms are only relating to the error term)

Anyway, the question with two question marks above is the most important one for me right now.
Thank you.

[startz]

Hi,
if I want to estimate gdp as a function of its own lag, an ma term and say a further explanatory variable (consumer confidence), i.e. gdp=c+a1*gdp(-1)+a2*u(t-1)+a3*CC+u(t) what do I do?

Code: Select all

gdp c gdp(-1) ma(1) cc

Textbooks generally seem to suggest that an ARMA model is one, where the dependent variable X (whatever it is) is a function of a constant, its own lag and one or more lags of the error term (ut-1 etc). Yet eviews seems to define this completely differently. (AR and MA terms are only relating to the error term)

EViews uses the standard definition, which does apply to the error terms. The confusion is because ls y ar(1) and ls y y(-1) are equivalent, although the equivalence breaks down in more complicated models.

[Alexx]

Thank you for the answer.