Factor model restrictions

[miksterz1]

Hello,

I am wondering if there is a way in Eviews to impose restrictions on the factor structure matrix when estimating a factor model. I am interested in extracting factors, some of which should be common to the whole data set (the usual procedure achieves this), but also some which are only common to a subset (not able to do this yet). Hence, I need to tell the model to generate a factor that only loads on a subset of my data. Is this possible in Eviews?

[miksterz1]

Bumping this. To put it more simply, is there a way to restrict certain factor loadings to be zero?

[EViews Glenn]

I can't think of a way to do this.

[miksterz1]

Is it something you guys are looking in to for future versions? You can't achieve identification in factor models without imposing restrictions on the loadings matrix.

[EViews Glenn]

We currently don't but will give it some thought. Typically, people use the various rotation methods which place different kinds of restrictions on the loadings matrix. But we hadn't given much thought to user-specified restrictions. Any references you wish to provide for context and validation would be useful.

[miksterz1]

Hi Glenn,

Thanks for your response. As you know, the factors and loadings are not uniquely identified in factor models. Here is an example of a technical paper explaining the different kinds of restrictions that facilitate identification:
http://www.columbia.edu/~sn2294/papers/rotate.pdf

And here is a more applied example:
http://www.aeaweb.org/articles.php?doi= ... 3769206278

[EViews Glenn]

Followup

http://forums.eviews.com/viewtopic.php?f=4&t=6625

Bottom line. Our factor guy had forgotten that he implemented target rotation. However, target rotation will not enforce the zeros if the restrictions are not just identifying.