Meaning of "Eviews"'s VIF

[mectricsdonk]

Dear Eviews brothers,

Theoretically, the uncentered VIF is the variance of the coefficient divided by the variance of the coefficient if the equation only contained that specific regressor and no constant.

Eviews computes this different, something with inproduct of the regressor times some variance ratio.

What is the meaning of the Eviews VIF? Why is it different. What is the argument to compute it this way?

In my model with only one regressor the VIF seems to explode, what could be the reason?

Thanks mucho!

[EViews Gareth]

The User Guide has pretty good details...

[mectricsdonk]

The User Guide seems pretty useless if it isn't correct...

http://forums.eviews.com/viewtopic.php?f=4&t=3454

Could you please answer my question, why the difference in computing the VIF?

[EViews Gareth]

With a bit of algebra you can show that the uncentered VIF, for simple least squares, is actually just the ratio of the diagonal of inv(X'X) and the diagonal of (X'X).

The inner product bit of the formula I wrote in that post calculates (X'X). The coefficient covariance matix is calculated as sigma^2*inv(X'X). So we use the coefficient covariance matrix as a quick way of calculating Inv(X'X).

[mectricsdonk]

Thank mucho Gareth