Help with stepls combinatory?

For questions regarding programming in the EViews programming language.

Moderators: EViews Gareth, EViews Jason, EViews Moderator

danielb
Posts: 3
Joined: Tue Dec 05, 2017 1:22 pm

Help with stepls combinatory?

Postby danielb » Tue Dec 05, 2017 4:23 pm

I have 13 regressors and I have to find the best combination based on the Akaike Information Criterion. Being new to programming in EViews, the closest way I found was by using stepls combinatory, but it yields the model that presents the largest R-square, instead of the lowest AIC. How could I change that? If not possible, any tips on solving my problem? I also saw some programs involving for loops but they don't seem suitable for the case where we have 13 regressors (over 8000 equations). Any help would be extremely appreciated.

Thank you!

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 11411
Joined: Tue Sep 16, 2008 5:38 pm

Re: Help with stepls combinatory?

Postby EViews Gareth » Tue Dec 05, 2017 7:06 pm

13 choose how many?
Follow us on Twitter @IHSEViews

danielb
Posts: 3
Joined: Tue Dec 05, 2017 1:22 pm

Re: Help with stepls combinatory?

Postby danielb » Tue Dec 05, 2017 7:09 pm

It doesn't matter, i could use 1 of them or all 13, i just have to found the smallest AIC. That's the problem: i need to run through ALL possible combinations of the 13... the ones that only use 1, the ones that use 2, and so on.

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 11411
Joined: Tue Sep 16, 2008 5:38 pm

Re: Help with stepls combinatory?

Postby EViews Gareth » Tue Dec 05, 2017 7:19 pm

ok, that's not cominbatorial then.

I have a hunch that for a fixed number of regressors, minimising AIC is the same as maximising R^2 (in a linear regression). So you could just run STEPLS for each number of regressors, then compare the AICs from the chosen ones.
Follow us on Twitter @IHSEViews

danielb
Posts: 3
Joined: Tue Dec 05, 2017 1:22 pm

Re: Help with stepls combinatory?

Postby danielb » Tue Dec 05, 2017 7:22 pm

That's what i thought too... and i did exactly that. I used stepls combinatorial to run all models and compared the AIC's. Unfortunately, our hunch is not right: maximizing R-squared is not necessarily the same thing as minimizing AIC, so I could not rely on that result. :( :(


Return to “Programming”

Who is online

Users browsing this forum: No registered users and 11 guests