Re: [Gretl-users] Cointegration tests on many pairs of historic series
Am 17.10.2014 um 08:19 schrieb Riccardo (Jack) Lucchetti:
On Fri, 17 Oct 2014, Matteo Chiorboli wrote:
> Dear Gretl users,
>
> I need your help for running the Johansen test automatically on (n)
> historic series, matched in pairs.
> Substantially, I have 30 historic series, and I must test the
> cointegration
> between all the possible pairs. Then, existing 435 possible pairs, I need
> an automatic procedure to make this great number of Johansen tests.
> Overall, I need a procedure that can run all possible cointegration tests
> on a number of pairs of historic series given by the binomial coefficient
> (combination), that in the specific case is n=30, k=2, then
> [30!/(2!*(30-2)!]=435 pairs, and 435 Johansen tests.
This should be close to what you need, if I'm not mistaken. The example
Of course Jack forgot to add that typically this list does not exist to
get your work done by others, so you should thank him double. If this is
part of an assignment for a course, from the point of view of your
professor I would almost call it fraud (if you fail to mention it).
Apart from that, econometrically it seems to me that this procedure will
produce a whole bunch of internal inconsistencies. Theoretically
cointegration is a transitive thing, so if x and y are (bivariately)
cointegrated, and y and z are, too, then x and z must also be
cointegrated. This will in general not be the outcome of all your
separate tests.
cheers,
sven