Re: DM: Re: Why is Singular Vector Decomposition for OLS?

[wang yongjiang]

Inverting X'x when X is highly collinear is not computationally 
stable. SVD
is useful in diagnosing collinearity in X, but it is not 
computationally
efficient. See Chapter 9 and 11 of HAdi, A.S. (1996) "Matrix Algebra 
as a
Tool," Duxbury Press.

Other decompositions such as Cholesky and triangular decompositions 
are
better than SVD. For more details see Chapter 9 of Chatterjee, S.  and
Hadi, A. S. (1988) "Sensitivity Analysis in Linear Regression," John 
Wiley
and Sons.


=======================================================================
  Ali S. Hadi, Professor and Chair   Phone: +1-607-255-2748
  Department of Social Statistics    Fax: +1-607-255-8484
  Cornell University                 E-mail: ali-hadi@cornell.edu
  358 Ives Hall                      http://www.ilr.cornell.edu/~hadi/
  Ithaca, NY 14853-3901              http://ccaix3.unican.es/~AIGroup/
=======================================================================