Stability Diagnostics - Recursive Estimation and AR terms

[Mila]

Hi,
I was using the stability diagnostics to get recursive coefficients for some parsimonious OLS equations that have AR terms (lagged dependent variables) in them. Then I read the Eviews Help Guide, and noticed the statement: "The recursive estimates view is only available for equations estimated by ordinary least squares without AR and MA terms". Does anyone know if this means I literally cannot use recursive estimation option for equations with AR terms? I can't put my finger on why not - since it's just adding observations one at a time/increasing the sample size. Plus I can still generate the graphs and series.
I would really appreciate any advice on this.
Thanks!

[EViews Glenn]

There are several sets of recursive diagnostics. Of which are you referring?

[Mila]

Hi, thanks for your message.
It is the option that follows from stability diagnostics, "Recursive Estimates (OLS only)", and then all 6 options within that tab (Eviews 7.1). I think perhaps the Guide means that it doesn't work for purely ARMA regressions? I tried a univariate ARMA regression, and sure enough I wasn't able to use the above diagnostics option. But I suppose if I have a multivariate OLS regression that happens to have AR terms, it is fine?

Mila

[EViews Glenn]

The reason it's not offered for AR models is that the latter are nonlinear estimators. The recursive coefficients view uses one-step updating to find the solution to the linear least squares estimator without inverting the augmented moment matrix. Nonlinear estimation would involve full estimation for each augmented sample.

[Mila]

Hi Glenn,
Thanks for your clarification. I think I sort of get what you mean (please excuse my technical ignorance). Just to clarify: Does that mean that despite being able to get the recursive coefficients with my regressions (with AR terms), they are not the accurate nonlinear estimates?
Many thanks for your help!

[startz]

EViews treats lagged dependent variables and explicit AR terms differently. This may explain some of what's going on.

[Mila]

Thanks Startz! That might be it I think. My equations are with lagged dependent variables, so I'm still "allowed" to proceed with the recursive estimation options. I guess I will the results as "kosher" then.