Hi,
Kehl Dániel schrieb:
Dear gretl-users,
first of all I have to say that I am a beginner user of gretl.
I have a time-series with a frequency of 15 minutes (electricity
consumption).
~12 000 observations from 1 January 2009.
Is there a way I could set this frequeny? (unfortunatelly I don't use
the script language yet)
the 'setobs' command should also handle non-standard frequencies.
There's also a menu entry for it, roughly data->data set structure or
some such (I'm on German right now).
There is of course a daily seasonality and a weekly too, so I thought
I
should use 672 lags as predictor but was unable to do it with the ARMA
task.
ARMA is nonlinear and a large number of lags will mean numerical
problems. I don't know if there's an upper bound imposed in gretl's
implementation though. did you mean all lags up to 672 or just the 672th
lag itself?
In addition there are some days where the electricity consumption is
much lower due to national holidays. I wanted to use dummy variables to
solve this problem, but the fitting was very poor on these days.
Well you only seem to have each holiday ocurring once, given that your
data start 4 months ago, right? In any case, this sounds like a concrete
econometric modelling and specification problem, not something related
to the actual software used.
good luck,
sven