Re: [Gretl-devel] multivariate ARCH test for VARs
On Sat, 1 Apr 2017, Sven Schreiber wrote:
Am 01.04.2017 um 06:44 schrieb Artur Tarassow:
> It's very nice to have this type of test natively implemented now.
>
>> modtest [horizon] --arch --multivariate
>>
>> (--multivariate is currently a "secret" option).
>
> However, couldn't we just use the option "--march" instead of
"--arch
> --multivariate" ?
That was my first thought as well, but I guess at some point this method
should become the default (for VARS/systems), and so it wouldn't be good to
introduce a new option now. But it's not obvious (to me) which is the best
arrangement.
Here's one more thought. If the multivariate autocorrelation and ARCH
tests are fully validated, they should probably become the default for
VARs. But as Lutkepohl points out, they both chew up a lot of degrees
of freedom, and he talks about use of univariate tests as a fallback.
So we might go the other way from the current private option and add a
--univariate option to get the univariate versions even when testing a
VAR.
Allin