Multicollinearity / Orthogonalizing

[thesis]

Dear all,

Unfortenately 2 of my independent variables are highly correlated (0.75) I would like to estimate my regressions in an alternative way and therefore want to use orthogonalization.

However I do not EXACTLY know how to do this. Does anybody have any experience with this?

I was thinking about this process:

- First run an OLS regression with the 2 correlated Independent variables A and B, A is the dependent and B the independent in this regression
- Then Proc > Make Residual series. Name these residual series.
- Use this residual series as an alternative of one of the correlated independent variable as the new independent variable.

So is this the right process? Anybody any recommendations? Or a complete different solution for multicollinearity.

Thanks! Kind regards,

[startz]

What are you trying to accomplish when looking for a solution to multicollinearity?

[thesis]

I would like to ignore the multicollinearity however the signs(+ and -) are not appropriate so I would like include all the independent variables ( and not just drop one) and just run my model.

Additionally I think orthogonalizing might be interesting since it kind of filters the additional effect of that certain independent variable and I am interested in that specific additional effect!!

[startz]

The regression you've already run gives the best unbiased estimates of the coefficients. If you don't want unbiased coefficients, the procedure you suggest is reasonable. It's somewhat like taking principal components of A and B and then using those in the regression.

[thesis]

Ok, but HOW to do this in Eviews?!