I have a question about how to best handle NAs in ranking systems. Most public ranking systems like Core Combination use “Percentile NAs Negative”. Now I wonder why “Percentile NAs Neutral” is used in the Core Combination Europe version?
I would be grateful for recommendations on how you handle NAs in your ranking systems?
In my experience, “Percentile NAs Neutral” tends to yield better performance and have a more logical foundation than treating NAs as negative.
For example, consider a factor that looks at EPS surprises for a long ranking system. If there are no analyst estimates for a given stock, it would yield NA, but would that be considered a bearish signal for that stock? No, most would consider it neutral, hence why I now set up all my ranking systems as “Percentile NAs Neutral”.
Personally, I think it should be the default setting. Or at the very least, if you use the “Save As” function on a ranking system, have it default to the setting of the original ranking system. I’ve been burned a couple times copying a “Percentile NAs Netural” ranking system but forget to change the dropdown from negative to neutral.
Good point here. I stumbled upon this just this weekend actually. For the longest time I default to “NA’s negative”; I run long only, so keeping all the NA’s at the opposite end of the spectrum makes sense.
But - as you point out, N/A is not always a bad thing.
I made some mods to one of my main strategy in changing over GICS to RBICS, and noticed 24 yr CAGR increased 5% (40-45%)! Wow. But I realized the ranking system defaulted back to NA’s middle, (was previously negative).
If you want to get fancy I suppose you could deal with NAs differently for each factor/rule using ISNA.
One reason to use percentile NAs negative is if you’re using traditional price-to-X value factors. P/E, for example, should always be run with percentile NAs negative. Why? Because you don’t want all the stocks with negative earnings to be in the middle of the stocks with positive earnings! In addition, many people think that a company with N/A for a common profitability or margin ratio should be punished for lack of information. Personally, I avoid formulas like P/E and use earnings yield instead, and all my systems are NAs neutral.
I notice that enterprise value has increasingly been used in formulas I’m seeing. I don’t always see it being modified if it is a negative value in the denominator though. Isn’t a negative enterprise value possibly a good thing and shouldn’t formula factors using it be modified in such cases?
EV returns N/A if negative. If you want to modify a ratio that uses EV so that companies with negative EVs come out on top, I suggest something along these lines: EBITDATTM / Min(EV, 5), with higher values being better. Very few companies have EV = NA unless it’s negative, so this is a relatively safe approach. This Min command will force all NAs to equal 5.