I frequently see this scenario in my own research: A factor has good 10- or 20-decile linearity***, the difference between the highest quantile and the lowest decile is fairly large, yet none of the deciles have performed better than the S&P 500.
*** Or a good curve if it's the middle values that are superior
Does a "good" factor require that the top quantile outperform the chosen benchmark? Or is the the linearity + large difference between top and bottom the key?
Obviously, the ideal factor would show outperformance. Just wondering about others' interpretations of this scenario and how they may deal with it.
I would say that it is not required to have the top quantile beat the benchmark. S&P 500 has done quite well in the last years, probably better than the universe you're using, so it's hard to beat with single factors.
Especially if it's a 'different' factor (one that probably has low correlation with your other factors), I would not care at all if the raw performance is not that impressive.
1 Like
The top quartile of a good factor will outperform your stock universe. Your stock universe should be your benchmark, not some arbitrary index such as the S&P 500.
2 Likes
My $0.02: Not every factor needs to independently beat the SP500 (or whatever benchmark you are using) at the top percentiles. Stacking several factors like you describe may combine to exceed the benchmark. It's been a while, but I'm pretty sure I have many factors that may not be beneficial in isolation, but in combination are effective in helping sort companies.
Recently the SP500 has been driven by outperformance of largest companies so harder to beat mkt cap weights. Hasn't always been the case. If you go back 10-15 yrs the equal wt SP500 performed better than SP500, so it likely isn't a permanent state of affairs.
2 Likes
Thank you all. Makes sense to me.