[pystatsmodels] PACF>1

[Georgios Boumis]

Hello,

I run the following code and it seems I get an error message along with a
PACF>1 for lag 6. I cannot figure out what is wrong...

[image: ΣτιγΌιότυπο 2018-10-18, 8.42.29 ΌΌ.png]

[image: ΣτιγΌιότυπο 2018-10-18, 8.42.36 ΌΌ.png]

[image: ΣτιγΌιότυπο 2018-10-18, 8.42.44 ΌΌ.png]

[j***@gmail.com]

Post by Georgios Boumis
Hello,
I run the following code and it seems I get an error message along with a
PACF>1 for lag 6. I cannot figure out what is wrong...
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.29 ΌΌ.png]
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.36 ΌΌ.png]
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.44 ΌΌ.png]

Please copy text instead of adding pictures whenever possible, i.e. copy
paste from whatever console that you are using.

I think the underlying problem is that you are simulating a non-invertible
process. The first MA coefficient is 2, so I guess the MA roots are much
too large.
I never checked how this would affect the various computations for an ARMA
process.
AFAIK, all or most of this requires stationarity and invertibility of the
ARMA process.

Josef

[Georgios Boumis]

Yeah this ARMA process is non invertible. If I change the method of PACF
calculation to 'ols' for example, then it seems to give me PACF that are
under 1. So I guess is something with the yule walker solver.

Post by j***@gmail.com

Post by Georgios Boumis
Hello,
I run the following code and it seems I get an error message along with a
PACF>1 for lag 6. I cannot figure out what is wrong...
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.29 ΌΌ.png]
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.36 ΌΌ.png]
[image: ΣτιγΌιότυπο 2018-10-18, 8.42.44 ΌΌ.png]

Please copy text instead of adding pictures whenever possible, i.e. copy
paste from whatever console that you are using.
I think the underlying problem is that you are simulating a non-invertible
process. The first MA coefficient is 2, so I guess the MA roots are much
too large.
I never checked how this would affect the various computations for an ARMA
process.
AFAIK, all or most of this requires stationarity and invertibility of the
ARMA process.
Josef