Page 1 of 4

Switching regression Problem

Posted: Thu Jan 04, 2018 1:02 pm
by nicotrader
Hi all,

i have a problem with eviews 10 in particular with switch regression.

Then if I estimate a model and specific a sample, ex. 1-10000 observation of 12000 dataset the model rejects an output of probability state (STATIC FORECAST).
But if I estimate a model with same sample (1-10000 observation) but with 12500 dataset the model rejects a different output of probability state (STATIC FORECAST), then is impossible to estimate a reliable model.
If I add an observation in out-of sample the output in-sample result change!


please help.

Re: Switching regression Problem

Posted: Thu Jan 04, 2018 1:18 pm
by startz
May simply be a coincidence. Check that the starting values are the same for both estimates

Re: Switching regression Problem

Posted: Thu Jan 04, 2018 6:09 pm
by nicotrader
values ​​change I have tried innumerable estimate. ...

Re: Switching regression Problem

Posted: Fri Jan 05, 2018 12:13 am
by nicotrader
I have just made a further series of estimates, the problem exists.
Surely there is a bug problem, TO SOLVE URGENTLY.

Re: Switching regression Problem

Posted: Fri Jan 05, 2018 3:56 am
by EViews Gareth
Could you provide an example?

Re: Switching regression Problem

Posted: Fri Jan 05, 2018 11:14 pm
by nicotrader
I add an update.
The model output also changes when I use the same estimation model.
Incredible bug....



Model

https://ibb.co/bQuf6b

output

https://ibb.co/jV09zw

output

https://ibb.co/ds0WCG

Re: Switching regression Problem

Posted: Sat Jan 06, 2018 12:06 am
by EViews Gareth
I tried to recreate what I think you are explaining, but every thing looked ok...

Code: Select all

create u 12262 series y=nrnd smpl 1 10361 equation eq1.switchreg(type=markov, nstates=3) y c smpl @all eq1.fit yf pagestruct(end=13000) eq1.fit yf2 show yf yf2

Re: Switching regression Problem

Posted: Sat Jan 06, 2018 8:23 am
by nicotrader
I tried many times error is present, in addition you do not have the time series I use.

Re: Switching regression Problem

Posted: Sat Jan 06, 2018 9:28 am
by EViews Gareth
in addition you do not have the time series I use.
Would it be possible to rectify that?

Re: Switching regression Problem

Posted: Sat Jan 06, 2018 10:45 am
by nicotrader
sorry, rectification what?

Re: Switching regression Problem

Posted: Sat Jan 06, 2018 11:41 am
by EViews Gareth
Could you give us the data?

Re: Switching regression Problem

Posted: Mon Jan 08, 2018 12:27 am
by nicotrader
data.xlsx
data
(232.2 KiB) Downloaded 863 times

Re: Switching regression Problem

Posted: Mon Jan 08, 2018 9:26 am
by EViews Gareth
If I run the following:

Code: Select all

wfopen data.xlsx smpl 1 10361 equation eq1.switchreg(type=markov, nstates=3) series01 c smpl @all eq1.fit yf1 pagestruct(end=13000) eq1.fit yf2 show yf1 yf2
I obtain identical results.

Re: Switching regression Problem

Posted: Mon Jan 08, 2018 3:06 pm
by nicotrader
Sure????

This is the forecast output with in sample 10361 and out of sample 12536.


https://ibb.co/nezghR

totally different to in sample 10361 out of sample 12200.....

Re: Switching regression Problem

Posted: Mon Jan 08, 2018 3:25 pm
by EViews Gareth
Did you run the code I provided and achieve the same results or different ones?