On Fri, 23 May 2014, Sven Schreiber wrote:
Am 23.05.2014 18:23, schrieb Allin Cottrell:
> On Fri, 23 May 2014, Sven Schreiber wrote:
>
> Anyway, there are a couple of new things in CVS that should be helpful.
>
> 1) If the current dataset is 7-day or 6-day daily and you apply a
> restriction via "smpl", we check to see whether the restricted dataset is
> still "OK" as 6- or 5-day data. If so, we re-establish its time-series
> credentials.
Nice. Does it affect the "official" periodicity of the workfile ($pd for
example)?
If you open a 7-day data file, the $pd of the dataset in memory will be 7.
If you then subsample to drop weekends, the $pd reported will change to 5.
If you don't give the --permanent option, then in the background you still
have a 7-day dataset, and if you do smpl --full $pd will go back to 7.
And could I pester you again about displaying the variant of daily
periodicity at the bottom of the main window where it now just says
"daily".
That's already done in CVS ;-)
> Along with the generalization of weekday() this means that you
can discard
> empty weekends to get a 5-day dataset in two lines of hansl.
>
> <hansl>
> open <seven-day-data>
> series wd = weekday($obsmajor,$obsminor,$obsmicro)
> smpl wd >= 1 && wd <= 5 --restrict --permanent
> </hansl>
>
Strictly speaking the --permanent option is not necessary here, right?
Right. That's just if you want to save the reduced data, or save on
memory.
And an OT follow-up about the non-weekend holidays: How should they
be
treated? Econometrically I guess in many applications you just remove
and ignore them like the weekends. But workfile-wise? If they are
removed, we lose the time-series property of the workfile (apart from
the clunky workaround without date information). If I keep them I don't
have a contiguous time series anymore.
We handle this automatically for ols/wls (dropping the NAs on the fly and
treating the reduced data as if contiguous). But not for other estimators.
Maybe I should check how Eviews does it to get some inspiration...
That would be interesting to know.
Allin