Detecting "Near singular matrix" before OLS regression
Posted: Fri Mar 04, 2011 3:47 pm
Hi,
I am trying to estimate a linear equation y = X b + e by OLS over different sample periods.
For a certain sample period, EViews 7 reported error "Near singular matrix error. Regressors may be perfectly collinear."
I would like to skip the problematic sample and avoid the near singular matrix error. Is there a way to automatically do that?
(I have tried to pre-test the X matrix using @rank(X) (with default tolerance), but X is of full rank even when EViews reports the error.)
Thanks a lot for your help.
JF
I am trying to estimate a linear equation y = X b + e by OLS over different sample periods.
For a certain sample period, EViews 7 reported error "Near singular matrix error. Regressors may be perfectly collinear."
I would like to skip the problematic sample and avoid the near singular matrix error. Is there a way to automatically do that?
(I have tried to pre-test the X matrix using @rank(X) (with default tolerance), but X is of full rank even when EViews reports the error.)
Thanks a lot for your help.
JF