Re: [Gretl-devel] midas (was: Re: next release)
On Fri, 7 Jul 2017, Sven Schreiber wrote:
Am 07.07.2017 um 15:52 schrieb Sven Schreiber:
> Am 07.07.2017 um 01:06 schrieb Allin Cottrell:
>> We're planning to release gretl 2017c soon (I'm thinking of Monday, July
>> 10), so if anyone wants to update a translation, or has a bug report
>> pending, this would be a good time to act.
This time here's something that might be relevant for this release. Consider
the following lines ripped from a script which was based on Allin's example
(gdp_midas.inp):
midasreg dep const ownlag ; mdsl(dIPL, 0) mdsl(dOL, 0)
printf "midas coefficients:\n%10.5f\n", $model.midas_coeffs[,1]
The midasreg line seems to run OK, so having two mdsl(...) parts seems legal.
However, at the second line gretl (latest snapshot) tells me:
"midas_coeffs": kein solches Ding [no such thing]
Is it a bug or am I going beyond the Midas spec here?
Ah, that needs to be documented. $model.midas_coeffs is a special
thing, available just when one midas spec is included: it's intended
(mostly) for producing a plot of the implicit coefficient on
successive high-frequency lags. Such a plot is available when you
estimate a MIDAS model via the GUI, under "/Graphs/MIDAS
coefficients".
The "midas_coeffs" object is a two-column matrix with lag in column
1 and implicit coefficient in column 2 (that is, the leading
coefficient on the MIDAS term times the weight at that lag). This
could be generalized for the case of multiple midas terms by making
it an array of matrices, but that didn't seem worthwhile to me. Also
note that with U-MIDAS or straight Almon polynomial the "leading
coefficient" is absent (or identically 1.0, if you like) so the idea
is not so useful.
But if you just want the coefficients from a MIDAS model, you'll
find them in $coeff as usual.
Allin