Comment Posted Online to “Quantitative Strategies for Selecting Stocks,” by Richard Tortoriello, May 2010 AAII Journal:
We need more articles like this on quantitative methodologies, assumptions and results.
Print this article
Tortoriello requires a linear relationship in his quintiles. Sometimes both the low and high values may have the best (or worst) performance. By dividing into deciles and using decision-tree software, one can discover relationships other than linear relationships. Even in a linear relationship, one would want to know the best cut-off value.
Just because I don’t predict or understand a given relationship is not to say that the relationship does not exist and is predictive. We are easily seduced and blinded by our assumptions. One can mine the data and explore many more possibilities while still guarding against over-fitting by testing and protecting with statistical measures.
The linear relationships confirmed for the four variables don’t say anything about how they would perform in combination. One needs to test for the interaction effect.
To read more, please become an AAII member or CLICK HERE.