Estimation of exp smoothing parameters

kiber_master · Postby **kiber_master** » Tue Oct 11, 2011 10:10 pm

Hello!

I've got in User's Guide that exponential smoothing estimate best values for smoothing parameters by minimising SSR. What exactly method do you use? If it is grid search, is it effective? If one of nonlinear optimisation methods - does it always get the best coefficients. And I actually can't find, what value of tolerance you provide exactly in this method.

Thank you.

kiber_master · Postby **kiber_master** » Wed Oct 12, 2011 3:28 am

Well, I found the following:

"EViews uses (a version of) grid search for the exponential smoothing routine." (UG II, p.759)

So, it isn't exactly a grid search, am I right?

EViews Gareth · Postby **EViews Gareth** » Wed Oct 12, 2011 6:43 am

It is grid search.

kiber_master · Postby **kiber_master** » Wed Oct 12, 2011 10:42 pm

Thanks a lot!

kiber_master · Postby **kiber_master** » Wed Nov 02, 2011 9:44 pm

Sorry, but how can it be simple grid search, when I didn't get the best result putting E, E, E for Holt-Winters - Additive ?
Please, check the workfile, SSR is better for 1,0,0, not for 1,0.74,0.

EViews Gareth · Postby **EViews Gareth** » Thu Nov 03, 2011 7:52 am

It is a grid search. Just not a particularly good one. If you want more accurate estimates of the parameters, you might want to use the ExpSmooth Add-in that calls the R Forecast package. Although you'll have to think a little about the seasonality of your data for that.

kiber_master · Postby **kiber_master** » Fri Nov 11, 2011 12:08 am

>If you want more accurate estimates of the parameters
And what accuracy do you provide?

EViews Gareth · Postby **EViews Gareth** » Fri Nov 11, 2011 8:43 am

I'm not sure I understand your question.

kiber_master · Postby **kiber_master** » Sat Nov 12, 2011 3:50 am

I mean, when you make a grid, what is the step? 0.01?

EViews Gareth · Postby **EViews Gareth** » Sat Nov 12, 2011 8:56 am

EViews.com