First of all, thanks for the quick answer! Great support.
Let me just put the X-12-ARIMA and TRAMO/SEATS question in another
way: Ok, monthly and quarterly is possible, daily maybe after
revisiting gretl code. But am I right that both programs cannot handle
hourly values at all? I thought that their seasonal adjustment
algorithms just work for something more aggregated than hourly values.
Correct? I tried to use TSW on a small subset of my data but already
failed on specifying the periods in the data file.
Again, thanks in advance for answer. :-)
On Thu, Jul 23, 2009 at 12:03 AM, Allin Cottrell<cottrell(a)wfu.edu> wrote:
On Wed, 22 Jul 2009, Oliver Heering wrote:
> i'd like to fit an ARMA or ARIMA model to hourly timeseries data. As i
> see monthly or quarterly timeseries in examples most of the time, some
> questions arise:
>
> My data summary:
> Type of data: time series
> Frequency: hourly
> Range: 1:01 - 3325:24 (n = 79800) <- yes, a lot i assume?
>
> - is there any possibility to use special labels for my values? I
> imported the timeseries from CSV files which had a timestamp for each
> value. gretl extracted these strings and stored them in
> string_table.txt. How can i use these strings for example in plots?
> Having only 3456:23 as label isn't much of a help, "2006-04-02
> 15:00:00" would be much better. What is the string_table.txt for?
> Haven't found any clue in the docs.
Hello Oliver,
The string table is really meant for string variables, such as are
found in some datasets, rather than for observation identifiers.
(There's a short discussion in cjapter 4 of the User's Guide,
section "Common points on imported data".)
If your CSV file contains timestamps you could add them as "case
markers": /Data/Add case markers. However this won't work at
present to encode something like "2006-04-02 15:00:00", even if
you dropped the redundant trailing ":00:00", since gretl displays
only 8 characters of such markers. That's probably something we
should revisit.
> - I know about X-12-ARIMA and TRAMO/SEATS integration and i also know
> about their limitation of about 720/600 values. Is it true that both
> programs only work for monthly timeseries?
Quarterly is OK too. I'm pretty sure the programs can handle
ARIMA on daily data, but gretl is not set up to pass them the
right commands to do so. Again maybe something to revisit.
> - why can't i compact my data with gretl? hourly -> daily -> weekly ->
> monthly... even daily data cannot be compacted AFAIR.
hourly -> daily would be easy, and daily -> monthly also, provided
we had correct calendar dates for the daily data. There hasn't
been any demand for this to date, but it's something we could add.
Weeks are a mess to handle.
> - a general question: can i resize the plots that gretl produces
> (aspect-ratio changes for example)? or do i have to: 1. save the plot
> as icon 2. copy over the plot commands to gnuplot 3. edit size of
> plot manually by issuing "set term ..." (i want to store the plot as
> file)
No, gretl doesn't support resizing the plots it produces, you'd
have to do that via gnuplot commands.
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
--
mfg,
Oliver