PCA scores

[eduardopolimpio]

Hi,

I've performed a PCA in Eviews 7 (covariance matrix and normalized loadings) and tried to reproduce the scores obtained using the eigenvectors matrix (L) and the data in the matrix Y. I tried to use:
matrix factors = Y*L
but the matrix "factors" has different values than the scores obtained under the Proc/Make Principal Components. I couldn't find why is this difference appearing. Can anybody help me?

Regards

Eduardo

[EViews Glenn]

As described in the manual,

Note that when computing scores using Equation (12.33), EViews will transform the Y to match the data used in the original computation. For example, the data will be scaled for analysis of correlation matrices, and partialing will remove means and any conditioning variables.

The manual does, however, incorrectly say that the default computation uses "cov". The default is actually "corr" so that we have to standardize the Y prior to calculation (I'll get the docs fixed up.)

Accordingly, the following replicates the calculations for the (arbitrarily chosen series) LBLA LOAN OIL ORDERS. Substitute with your own as necessary.

Code: Select all

group group01 lbla loan oil orders group01.makepcomp(eigvec=evec) a1 a2 a3 group group02 a1 a2 a3 group group03 (lbla-@mean(lbla))/@stdevp(lbla) (loan-@mean(loan))/@stdevp(loan) (oil-@mean(oil))/@stdevp(oil) (orders-@mean(orders))/@stdevp(orders) stom(group03, ff) matrix scores = ff*evec