|
The problem of multicollinearity
It's interesting to read through these beta-tests of this form feature, which I assume is based on D. Schwartz's 'Improve or Decline'. Please correct me if I'm wrong. If someone could interpret and explain these figures and the theory behind them in greater detail, I would be grateful.
One of the things that struck me here, was Ernie's comments about the high degree of correlation between these new indicators and the Sartin primary factors. To me this is the key problem that affects all the experimentation with assigning weights that surrounds the Methodology - almost everything can be too easily correlated with the primary factors, TE, BL/BL etc.. It seems to be that for any new factor, the most important aspect to test for and determine would be its independence from the primary factors - how much it actually adds to their well-established ROI. As far as I know now, the only other truly independent factor is E/L.
Any feedback appreciated.
Cheers,
B.Jennet
|