Re: [Gretl-devel] our handling of daily data
Am 26.05.2014 21:07, schrieb Allin Cottrell:
Sven has raised the question of the handling of daily data in gretl;
see the threads starting from
http://lists.wfu.edu/pipermail/gretl-users/2014-May/010037.html
I'm glad of that: it's time we clarified what we do now, and what we
should do in future. (But please note, I'm mostly talking here about
5-day financial-market data; other sorts of daily data might require
different handling.)
Sorry, this is long, but I'd encourage those who work with daily
data to read on...
I'm only able to react to part of your post, like Jack also said.
Anyway, it's (now) easy enough to strip out weekends, which leaves
the more interesting question of how to deal with holidays.
I think it's fair to say:
(a) most econometricians who wish to apply time-series methods to
daily financial market data will, most of the time, want to ignore
holidays as well as weekends, treating the data as if these days did
not exist and the actual trading days formed a continuous series,
but
yes
(b) for some purposes it may be important to be able to recover
information on (e.g.) which days were Mondays or which days followed
holidays.
How are these needs best supported by econometric software? I can
see two possibilities:
And I have remembered that I wanted to check with Eviews. I have now
taken a look at Eviews 8, but I think this area has not changed much
since v7 or v6.
Eviews introduces the concepts of "regular frequency" and "irregular
frequency" time series, where "irregular" is exactly what we're talking
about with 5-days daily and holidays: the physical time interval is not
constant between observations, but we "don't care". When you want to
specify an Eviews workfile as having irregular frequency, you need to
supply a date ID series. For regular frequency of course a start date is
sufficient.
Interestingly I could force an irregular-frequency 5-days daily to be
converted to regular-freq 7-days daily with NAs, and the gaps would show
up in the graphs, but afterwards Eviews would still give me the
correlogram of that gappy series. Probably by internally treating it as
5-days daily, but that's speculation so far. The correlogram also still
worked after I manually had set a virtual May-1st obs to NA as well.
When I estimate a simple regression with this 7-days daily workfile with
many NAs, Eviews just does it and reports the start and end dates plus
the number of obs used (haven't checked the count) and the information
"adjusted sample"; interestingly the residual graph doesn't show any
gaps (in contrast to the simple time series graph in the beginning),
whereas the spreadsheet view of the resid vector has all the expected
NAs -- don't know how they connect their points. Lags in the regression
also just works.
BTW, all this is not confined to daily data, I re-defined the gappy
workfile as monthly, and correlogram and OLS still works as before. So
it seems that Eviews in general very "tolerant" of non-contiguous
time-series. (Whether it makes sense for monthly or quarterly is another
issue I guess...)
So that's the situation. Part of it sounds a bit like Allin's ideas.
thanks,
sven