Page 1 of 1

Near Singular Matrix Error when performing zivot-andrews test

Posted: Tue Dec 06, 2016 10:12 am
by albertozzio
Hi! Can anybody explain me why when I perform a Zivot-Andrews Unit root test I get a message "Near singular matrix error. Regressors may be perfectly collinear" ?

I have 8 or 9 time-series representing the price of natural gas in different countries at different periods of time. The weird thing is that it only appears when performing the aforementioned test on two series (p_austria and p_slovenia).
It's really important to understand why is there this problem since I need the study for my dissertation! P.S. I am using Eviews 9.

Here I attach the wf1:

Thanks!
database.wf1
(84.88 KiB) Downloaded 231 times

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Tue Dec 06, 2016 2:26 pm
by EViews Glenn
Since you have EViews 9, is there a reason that you are not using the built-in unit root breakpoint tests?

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 4:16 am
by albertozzio
Hi Eviews Glenn!

Thanks for your reply. I am using Zivot Andrews unit root test because I wanna test H0: unit root hyp. + structural change hyp. Basically I am replicating a paper by Renou-Maissant, in which they use this technique.

Which other "built-in" unit root breakpoint testing do you suggest me?

Thanks a lot,

Alberto

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 7:38 am
by EViews Gareth

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 9:08 am
by albertozzio
Hi Gareth,
thanks for your message. These tests are different because Zivot - Andrews uses the dummies to account for structural break and for my thesis I need to use it then.
Do you have any clue about why is it possible to me to perform Zivot-Andrews on some series and not on some else? I mean the model it implements is the same one (with dummies to account for intercept and for trend change), and the data about different countries are over the same period of time.
How can I explain that then?

Bests,

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 10:55 am
by EViews Gareth
We know nothing about the add-in

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 10:58 am
by albertozzio
Who could I ask for then? should I move to another area in the Eviews User Forum?

Thanks

Re: Near Singular Matrix Error when performing zivot-andrews test

Posted: Wed Dec 07, 2016 1:44 pm
by EViews Glenn
I'm not familiar with the paper that you cite, but as Gareth indicated, the options that Zivot-Andrews employ in their paper can be specified in the built-in breakpoint unit root test. For example, the results in Perron (1997), can be replicated in EViews (as in our documentation). See Vogelsang and Perron (1998) International Economic Review, for a nice summary of the specifications. Once you decide your intercept/trend spec and the breaking spec, then it's straightforward to specify the t-test lag selection and minimize t-statistic form of the ZA test.