How large is your sample?  One characteristic root of your system is 0.954 which is not far from a unit root.  Thus there is a lot of persistence in your model that may be leading to your result.  Increase your sample and see if you get more satisfactory results.  

Bear in mind that the purpose of Box-Jenkins is to derive a parsimonious approximation to the process.  One does not try to uncover the actual process underlying the data.  The parsimonious representation may also forecast metter than a more elaborate representation.  Even if you knew the p and q for the actual model you would probably find that the forecast of the parsimonious model and the true model were very similar.

John C Frain
3 Aranleigh Park


On Wed, 27 Feb 2019 at 15:25, Raul Gimeno <mrexito@vtxmail.ch> wrote:

Hello

 

I’ve simulated an AR(3) process: Xt = 5 + 0.4Xt-1 - 0.1Xt-2 + 0.6Xt-3  + Ut

 

The PACF gives me only for lag 1 a statistically significant value.

Shouldn’t I get for the first three lags statistically significant values?

If the PACF doesn’t help me to determine the order of the AR process how should I proceed to estimate the process order with gretl?

Thanks

Raul

_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users