R help - Jun 2007 - Evaluating predictive power with no intercept-statistics question - not R question

I realize that the following has been talked about on this list many
times before in some related way but I
am going to ask for help anyway because I still don't know what to do. 

Suppose I have no intercept models such as the following :

Y = B*X_1 + error
Y = B*X_2 + error
Y = B*X_3 + error
Y = B*X_4 + error

and I run regressions on each ( over the same sample of Y ) and now I
want to evaluate which X has the greatest predictive power.
I'm fairly certain that R squared is not applicable because of the lack
of an intercept but I was wondering what was ? 
Any references to this particular problem or suggestions are
appreciated. I honestly believe that including an intercept is incorrect
For my particular problem. Thanks.

Maybe I could put all the X's in one regression and some kind of
topdownselect or StepAIC algorithm for example ?  Thanks.
--------------------------------------------------------

This is not an offer (or solicitation of an offer) to buy/se...{{dropped}}