9:43 a.m.
Hi,
I am trying to replicate the following triple exponential model (additive) from the famous
Hyndman book. https://otexts.com/fpp3/holt-winters.html. Gretl predictions are given
below.
Trips_W_add Trips_W_trend Trips_W_seas
2018:1 13.13124 11.01523 2.210526
2018:2 11.11609 11.10974 0.100856
2018:3 11.07205 11.20426 -0.037689
2018:4 11.58887 11.29877 0.384615
2019:1 13.50930 11.39329 2.210526
2019:2 11.49414 11.48780 0.100856
2019:3 11.45011 11.58231 -0.037689
2019:4 11.96693 11.67683 0.384615
2020:1 13.88735 11.77134 2.210526
2020:2 11.87220 11.86586 0.100856
2020:3 11.82817 11.96037 -0.037689
2020:4 12.34498 12.05488 0.384615
Hyndman predictions:
2018 Q1 1 12.9
2018 Q2 2 11.2
2018 Q3 3 11.0
2018 Q4 4 11.2
2019 Q1 5 13.4
2019 Q2 6 11.7
2019 Q3 7 11.5
2019 Q4 8 11.7
2020 Q1 9 13.9
2020 Q2 10 12.2
2020 Q3 11 11.9
2020 Q4 12 12.2
They are not matching. Also, gretl is not printing the final alpha, beta, gamma values.
Any help? data is given below.
https://drive.google.com/open?id=1BnKMYc6vO1fi987f5P6LcEGgV9JXcYxV&us...
12:16 p.m.
Am 04.03.2026 um 10:43 schrieb ramakrishnasalagrama(a)gmail.com:
Hi,
I am trying to replicate the following triple exponential model (additive) from the
famous Hyndman book.https://otexts.com/fpp3/holt-winters.html. Gretl predictions are given
below.
Hi, such replications and comparisons are really very useful, so it's
great you're bringing this up.
...
They are not matching. Also, gretl is not printing the final alpha, beta, gamma values.
Any help? data is given below.
From the subject of your post, you're using the contributed package
Winters version 1.1. If I understand the help doc of that package
correctly, it's using fixed parameter values -- assuming that what
you're calling alpha,beta,gamma is the same thing as
delta1,delta2,delta3 (=level/slope/seasonality) in the Winters package.
So the "final" values should be the same as the initial ones, AFAICS.
One could try to use the values reported by Hyndman with the Winters
package, or did you do that already?
The package that you were using previously, fcModels, apparently chooses
the parameters automatically. I'm not saying one package is better than
the other, I know too little about their internals.
cheers
sven
2:13 p.m.
Hi,
Fc models indeed chose the values automatically, but it is not providing out of sample
predictions. So, one has to use fcmodels to get parameters and use these in winters1.1
package to get forecasts. This is really irritating and some one has to fix these issues.
Forecasts by Winter's TES (Triple Exponential Smoothing)
-----------------------------------------------------
The parameter is based on min. MSE
Level alpha: 0.310
Trend gamma: 0.110
Seasonality beta: 0.040
Original series: Trips
Forecast series: Trips_TES
-----------------------------------------------------
Full data range: 1998:1 - 2020:4 (n = 92)
Current sample: 1998:2 - 2020:4 (n = 91)
Naive Trips_TES
RMSE 1.4840 0.4244
MSE 2.2021 0.1801
MAE 1.1624 0.3172
MAPE 11.8303 3.3247
Theil's U 1.0000 0.3031
=====================================================
Generated list List01
2:52 p.m.
On 04/03/2026 15:13, ramakrishnasalagrama(a)gmail.com wrote:
Hi,
Fc models indeed chose the values automatically, but it is not providing out of sample
predictions. So, one has to use fcmodels to get parameters and use these in winters1.1
package to get forecasts. This is really irritating and some one has to fix these issues.
I understand your disappointment. Since Ignacio has retired from the
scene since a few years, I suggest that you could adopt his package and
modify it to your liking.
If you need assistance, I'm sure the community will be glad to give
you all the help you need.
-------------------------------------------------------
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
-------------------------------------------------------
6:36 p.m.
On Wed, Mar 4, 2026 at 9:50 AM Riccardo (Jack) Lucchetti
<p002264(a)staff.univpm.it> wrote:
On 04/03/2026 15:13, ramakrishnasalagrama(a)gmail.com wrote:
> Hi,
>
> Fc models indeed chose the values automatically, but it is not providing out of
sample predictions. So, one has to use fcmodels to get parameters and use these in
winters1.1 package to get forecasts. This is really irritating and some one has to fix
these issues.
I understand your disappointment. Since Ignacio has retired from the
scene since a few years, I suggest that you could adopt his package and
modify it to your liking.
If you need assistance, I'm sure the community will be glad to give
you all the help you need.
Another approach would be to complete the wes() function in fcModels.
I see that some of the other forecasting methods in the package offer
out-of-sample forecasts, and I guess that apparatus could be re-used
for wes(). Or one "cannibalize" elements of Winters and fcModels to
make Ignacio's Winters call the parameter search currently in fcModels
if the caller doesn't specify the parameters -- I think this option
looks quite promising.
Allin
17
days inactive
17
days old
4 comments
4 participants
participants (4)
-
Cottrell, Allin -
ramakrishnasalagrama＠gmail.com -
Riccardo (Jack) Lucchetti -
Sven Schreiber