TSLS, same equation and differentes results??

[lucem_fert]

Hi!

I'm tryng to run a TSLS estimation with AR(1), the problem is that if I run the same equation it gives me different results!!! Even if I just click on "estimate" and i jut click again in OK, it gives different results!!

The problem is that results are quite different: sometimes the equation gives me a "Inverted AR roots" and then a stationary ARMA... but sometimes it gives a "Inverted AR Roots 1.03; Estimated AR process is nonstationary"... so I cannot definitely rely on that results because i need to get Inverted AR roots.

It's making me crazy!

1) The Equations specification is:

pibcf80 i80 pibcf80(-1) ar(1)

And the instrument list is:

i80 i80(-1) pibcf80(-1) pibcf80(-2).

And i select both of the options "include a constant" or "include lagged regresssor for...". i80(-1) is the instrument for pibcf80.

2) And same problem happens if I don't choose the options or if i choose them and omit the instruments nmanually like just using i80 and i80(-1)...


Is there anybody who can help me to understan it???

Thanks a lot!

[startz]

Technically, an AR(1) model is nonlinear so the results can depend on starting values in the coefficient vector. This is quite unusual though.

You might post either your output or your workfile. And what version and release date of EViews are you using.