Making Value Work

[yuvaltaylor]

Here’s an experiment. Create a ranking system with, say, 50 or 100 factors. Weight one of them at 100% and the rest at 0%.

Then create a screen on the Russell 3000 universe using that ranking system with just two rules:

rankprev(52) > 80
eps%chgttm > 15

Press the “totals” key and write down the number. Then weight a different factor at 100% and repeat.

The factors with the highest totals will be the ones that have worked the best in the last year.

Repeat for a few previous “as of” dates to weed out flukes.

Does that seem like a sensible way to model for future growth? Obviously, the factors would have to make good financial sense.

If anyone has any suggestions for improving this experiment, I’d welcome them.

Yuval, this may answer a major question I have had about testing emerging factors.
Have you found this method to be effective?

Well, I was using it to answer a very specific question: what factors best predict future EPS growth? But then I changed my approach pretty significantly: for a subsequent study I simply got masses of data and looked at the correlation between various factors and future EPS growth.