[R] Granger causality: lag selection

[John C Frain]

This is probably a question about econometric methods rather than one about
R,  While I would probably use an information criterion this would only be
my first step in determining an appropriate lag length.  One needs to check
1. The whiteness of the residuals
2. Non-normality and
3. Structural Change.
One then may need to amend the selected lag to take account of any problems
that are revealed by the relevant tests.  Only when you are satisfied that
your model is correct you should proceed with your  causality tests.
These matters are well covered in texts such as Lutkepohl (2005), New
introduction to multiple time series analysis, Springer (see Chapter 4)

Looking at the example you quote I might say that the automatic first
differencing of I(1) series and doing granger causality tests on the
differenced series is not a valid option if there is cointegration.  When
there is cointegration the usual test statistics do not have the standard
distributions.  This is covered in section 6.6 and chapter 7 of Lutkepohl's
book.
John C Frain
3 Aranleigh Park
Rathfarnham
Dublin 14
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
mailto:frainj using tcd.ie
mailto:frainj using gmail.com


On Fri, 2 Nov 2018 at 09:00, John <miaojpm using gmail.com> wrote:

> Hi,
>
>    I see an R user chooses the lag of Granger causality by finding out the
> lag for the most significant result
>    https://www.r-bloggers.com/granger-causality-test/
>
>    Is it generally legitimate to do so, without determining the lag first
> by AIC or BIC?
>
>    Thanks,
>
> John
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

	[[alternative HTML version deleted]]