Page 1 of 1


Posted: Wed May 13, 2015 1:31 am
by trubador
This thread is about crossvalid add-in.

The add-in performs a k-fold cross validation procedure on an already estimated model. Basically, it randomly divides the data set into a k number of subsets and reestimate the model to forecast each fold using all the remaining folds as the training set. Although it operates on equation objects that are estimated with any chosen method, the approach is feasible only for cross-section data and therefore is not available for time series data (as there are better ways for such purposes). Please see the documentation for more details.

Since the add-in makes use of some new features of EViews introduced with version 9, it will not work for older versions.

Re: Crossvalid

Posted: Tue Jul 14, 2015 5:47 am
by jbonanca

thank you for this add-in. regarding procedures for time-series data, you mention there are better ways -- before i write something from scratch, is there a thread or another package you would recommend i look at? (i've searched but found anything that seems to be right.)

Re: Crossvalid

Posted: Thu Jul 30, 2015 12:22 pm
by trubador
I was not referring to a particular package. In-sample and/or out-of-sample forecasting exercise along with using appropriate evaluation criteria might be a good way to start. EViews has nice features in this respect. You can refer to any time series econometrics textboot and EViews' manuals.

Re: Crossvalid

Posted: Mon Mar 21, 2016 9:13 pm
by diggetybo

Is it possible to view the eviews code used for k-fold cv somehow?

Re: Crossvalid

Posted: Tue Mar 22, 2016 1:50 am
by trubador
You can send me a private message explaining the reason of your interest along with your e-mail address.