[Gretl-devel] Adding observations within a function

Sven Schreiber svetosch at gmx.net
Fri Jul 31 09:22:18 EDT 2015


Am 31.07.2015 um 15:08 schrieb Riccardo (Jack) Lucchetti:
> On Fri, 31 Jul 2015, Sven Schreiber wrote:
> 

>>
>> In my view producing out-of-sample forecasts is one of the central aims
>> of applied time-series econometrics. Gretl then says "sorry I don't
>> support this (fully) with my built-in objects series and lists, it makes
>> me uneasy". Not exactly a first-class treatment, no?
> 
> I have to say, you have a point. However, there's a big difference (in
> terms of internal work needed, checks etc) between
> 
> (a) starting your job with a dataset that extends into the future
> because you know from the start that you'll want to do some forecasting
> (and this is possible right now), and
> 
> (b) starting your job with a certain dataset and then saying "well I
> think I could use a couple more obs, let's extend the dataset"
> 
> Whereas from the user's point of view it's not too much to ask to go for
> (a) instead of (b).
> 

Well yes and no. It would be impossible to write a function package to
do forecasting because of the limitations. So "asking the user to go for
(a)" in this context would seem to imply you cannot use a contributed
package as with other applications, instead the user has to do at least
some of the prep work herself.

I guess it is conceivable in principle that the out-of-sample range
itself becomes an argument of the function call: Suppose gretl had a new
sample-range object type. Then the user/caller could supply the
sample-to-be-added as an argument to the function that does some
out-of-sample forecasting. That way the responsibility of "mucking" with
the dataset remains with the caller, but the function would be capable
of producing out-of-sample results.

cheers,
sven



More information about the Gretl-devel mailing list