12:11 p.m.
Hi all,
with a fairly recent snapshot I've loaded the shipped example dataset
b-g.gdt, which gretl at the bottom announces as 5-day daily and
incomplete. OK - does "incomplete" mean gappy here? And there is no
information about the starting or ending dates, and possibly not on
where the gaps are, either. (Not sure about this one because it's a lot
of observations.)
Does anybody know more about this data?
thanks
sven
12:15 p.m.
On Wed, 13 Jul 2022, Sven Schreiber wrote:
Hi all,
with a fairly recent snapshot I've loaded the shipped example dataset
b-g.gdt, which gretl at the bottom announces as 5-day daily and
incomplete. OK - does "incomplete" mean gappy here? And there is no
information about the starting or ending dates, and possibly not on
where the gaps are, either. (Not sure about this one because it's a lot
of observations.)
Does anybody know more about this data?
The dataset info sheet reports:
This data set contains daily percentage nominal returns for the
Deutschemark/British Pound exchange rate. The sample period is Jan 3, 1984
to Dec. 31, 1991 for a total of 1974 observations. The original source is:
Bollerslev, T. and Ghysels, E., "Periodic Autoregressive Conditional
Heteroscedasticity", Journal of Business and Economic Statistics, Vol 14,
1996, pp. 139-151.
The data set is available from the JBES Data Archive, at
ftp://www.amstat.org/jbes/View/
Y = Daily percentage nominal returns, 100 * [ln(Pt) - ln(Pt-1),
where Pt is the bilateral Deutschemark/British pound rate
constructed from the corresponding U.S. dollar rates.
The gretl dataset is marked as undated, because although we know the
starting and ending dates, the individual observation dates are not
given and some days are skipped owing to holidays.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
12:31 p.m.
Am 13.07.2022 um 14:15 schrieb Riccardo (Jack) Lucchetti:
On Wed, 13 Jul 2022, Sven Schreiber wrote:
> Does anybody know more about this data?
The dataset info sheet reports:
This data set contains daily percentage nominal returns for the
Deutschemark/British Pound exchange rate. The sample period is Jan 3,
1984
to Dec. 31, 1991 for a total of 1974 observations. The original source
is:
Thanks, I should have thought of looking there myself.
The gretl dataset is marked as undated, because although we know the
starting and ending dates, the individual observation dates are not
given and some days are skipped owing to holidays.
Well, it's not really "undated" in gretl's terminology, I think,
because
it comes as a daily time-series dataset. There's also the accompanying
variable "dayskip" which tells you when the previous day (or presumably,
_at least_ the previous day, perhaps two over the weekend) was not a
trading day. Wouldn't it be possible to reconstruct the time information
from there? I mean we have the starting date, and we have gretl's
weekday() function, and because of "dayskip" we know when there's a gap.
OK, so I don't remember the exact current status quo with "gappy time
series" in gretl. For inserting the dates information maybe one would
have to fill the series with NAs, and that would prevent you from using
standard time-series operations and methods. So perhaps you would want
to keep two copies of the dataset, one with the gaps filled and another
one like b-g.gdt with the gaps being ignored.
thanks
sven
2:16 p.m.
On Wed, Jul 13, 2022 at 1:31 PM Sven Schreiber <svetosch(a)gmx.net> wrote:
Well, it's not really "undated" in gretl's
terminology, I think, because
it comes as a daily time-series dataset. There's also the accompanying
variable "dayskip" which tells you when the previous day (or presumably,
_at least_ the previous day, perhaps two over the weekend) was not a
trading day. Wouldn't it be possible to reconstruct the time information
from there? I mean we have the starting date, and we have gretl's
weekday() function, and because of "dayskip" we know when there's a gap.
<hansl>
nulldata 2086
setobs 5 1984-01-03 # ends at 1991-12-31
mondays = sum(weekday($obsmajor,$obsminor,$obsmicro)==1)
scalar dayskip_sum = 456 # from b-g.gdt
nhols = dayskip_sum - mondays
scalar yrs = 1991 - 1984 + 1
eval nhols / yrs
</hansl>
nols/yrs is 4.875. So you'd have to place 4 or 5 holidays per year.
Christmas, the 4th of July, and New Year are good candidates. Not sure
how you'd place the rest.
Allin
2:37 p.m.
Am 13.07.2022 um 16:16 schrieb Cottrell, Allin:
nols/yrs is 4.875. So you'd have to place 4 or 5 holidays per
year.
Christmas, the 4th of July, and New Year are good candidates. Not sure
how you'd place the rest.
Hm, but something's weird with the dayskip description. If the first obs
is Jan 3rd, 1984 (1984-01-03), then according to weekday() the first
couple of obs where dayskip is 1 would be Fridays (weekday() is 5
there). But the series' label says "no trading took place on the
previous day". Should this perhaps rather say "on the next day?".
And about placing the other holidays: You "just" would have to loop
through the series from the beginning, and every time you hit a
dayskip==1 which does not belong to a weekend, you insert a new NA. This
would break down if there were two or more consecutive non-weekend
holidays. Can this happen? OK, this is clearly not very important, but
maybe I can supply the needed information at some point in the future.
thanks
sven
2:48 p.m.
On Wed, Jul 13, 2022 at 3:37 PM Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 13.07.2022 um 16:16 schrieb Cottrell, Allin:
> nols/yrs is 4.875. So you'd have to place 4 or 5 holidays per year.
> Christmas, the 4th of July, and New Year are good candidates. Not sure
> how you'd place the rest.
>
Hm, but something's weird with the dayskip description. If the first obs
is Jan 3rd, 1984 (1984-01-03), then according to weekday() the first
couple of obs where dayskip is 1 would be Fridays (weekday() is 5
there). But the series' label says "no trading took place on the
previous day". Should this perhaps rather say "on the next day?".
No. It's explicitly Mondays and other days not preceded by a trading
day. See http://www.econometrics.com/intro/garchdat.htm .
1984-01-03 was a Tuesday, The dayskip value of 1 at observation 4
presumably refers to Monday the 9th, with the preceding Friday being
skipped as well as the weekend.
Allin
3:09 p.m.
Am 13.07.2022 um 16:48 schrieb Cottrell, Allin:
On Wed, Jul 13, 2022 at 3:37 PM Sven Schreiber
<svetosch(a)gmx.net> wrote:
> Am 13.07.2022 um 16:16 schrieb Cottrell, Allin:
>> nols/yrs is 4.875. So you'd have to place 4 or 5 holidays per year.
>> Christmas, the 4th of July, and New Year are good candidates. Not sure
>> how you'd place the rest.
>>
> Hm, but something's weird with the dayskip description. If the first obs
> is Jan 3rd, 1984 (1984-01-03), then according to weekday() the first
> couple of obs where dayskip is 1 would be Fridays (weekday() is 5
> there). But the series' label says "no trading took place on the
> previous day". Should this perhaps rather say "on the next day?".
No. It's explicitly Mondays and other days not preceded by a trading
day. See http://www.econometrics.com/intro/garchdat.htm .
1984-01-03 was a Tuesday, The dayskip value of 1 at observation 4
presumably refers to Monday the 9th, with the preceding Friday being
skipped as well as the weekend.
OK, if already at the beginning there's an interruption, then this makes
sense. Perhaps Jan 6th as being Three Kings' Day in some involved
countries is the background explanation there.
-s
1:10 p.m.
On Wed, Jul 13, 2022 at 1:11 PM Sven Schreiber <svetosch(a)gmx.net> wrote:
Hi all,
with a fairly recent snapshot I've loaded the shipped example dataset
b-g.gdt, which gretl at the bottom announces as 5-day daily and
incomplete. OK - does "incomplete" mean gappy here? And there is no
information about the starting or ending dates, and possibly not on
where the gaps are, either. (Not sure about this one because it's a lot
of observations.)
Please take a look at the "Dataset info" item (under the Data menu)
for this dataset (and any other datasets supplied with gretl about
which you have questions).
Allin
878
days inactive
878
days old
7 comments
3 participants
participants (3)
-
Cottrell, Allin -
Riccardo (Jack) Lucchetti -
Sven Schreiber