EViews.com

Hello,

I wonder if there is a way to extract values of MA(1)

I am running the following regression:

equation reg
reg.LS x c MA(1)

and I wonder if there is a way to extract the values of MA(1), I want to have them as a series or a vector.

Thanks in advance.

It isn't clear what you mean by "values of MA(1)"

When I run this regression:

equation reg
reg.LS x c MA(1)

E-views regresses x on a constant and MA(1). E-views calculates the values of MA(1) to be able to run the regression; however, I never see what those values are (unless I am missing something very obvious) all I get is the results. What I want is to get the values that E-views uses for the regressor and have them as a series or vector.

I am not sure if the best way to get the values of MA(1) for x is to extract them from the regression or maybe there is a way to calculate MA(1) for a certain variable.

Does this make more sense?

I'm still not entirely sure I follow. MA(1) is not some magic variable, rather it is a specification of the error terms in the equation. Perhaps you are asking for the residual series? You can get that with proc->make residuals.

When EViews estimates MA(1) process it runs this regression":
x = c(1) +c(2)e(t-1) + e(t)

and I wanted to get the e(t-1) that Eviews estimates to run the regression. But perhaps it is not so easy.

I think I should point out the reason why I was asking this. I tried to run a maximum likelihood estimation for ARMA(1,1) process

(In the code it is just x and z, two different variables)
But I ran into a problem, because I can't replace z with AR(1) or MA(1), eviews doesn't like that syntax. However, I could estimate AR(1) by replacing z with x(-1), and it worked fine. But I couldn't figure out how to replace "MA(1)" with a variable.

So basically, I need to figure how to estimate maximum likelihood for MA(1) process (If I can do MA(1) I should be able to do ARMA(1,1). I think I either need to get the e(t-1) that Eviews uses or know the syntax to tell Eviews to do it right in maximum likelihood code.

Thanks a lot for your help.

I think you might need a deeper understanding of estimating ARMA models. But in the no-backcasting case, the resid series will be equal to x-c(1)-c(2)*e(t-1).

I tried getting e(t-1) by

e(t-1) = [res - x + c(1)]/( -c(2))

but when I run the regression with x c e(t-1) I don't get the same results as when I do x c MA(1).

Which is a bit odd, because I can duplicate y perfectly from a x c y regression, while using the same exact technique.

I guess I really don't get this MA(1) stuff.