Putting Grades Together

Turning a lot of metrics into one metric

Recently, I’ve done a few posts where I give grades for metrics based on how strong a stock is relative to its peers and history. We’ve already looked at two domains (valuation and growth) with a total of ten metrics.

But, how do we put those metrics together if we want to create summary metrics?

The answer is not as simple as one might think for a simple reason. We have an unequal distribution for grading.

See, I wanted to make an A really mean something. Same for an F. Most stocks will have metrics that are clustered around B’s, C’s, and D’s. In fact, that will be the observation 80% of the time based on the distribution we gave it.

If an A is special then if we have a stock with a few A’s and maybe a few C’s then should that stock get an A, B, or C grade? Well, it’s probably not a C since it has strength from its special A’s. Giving it a B would be fine if A’s and C’s meant the same thing but clearly they don’t because A’s are given out 10% of the time and C’s 30% of the time as opposed to 20% and 20%. But, if it has C’s then does it really deserve an A?

This whole situation feels messy and could be resolved if I made the grading distribution flat (or even made a more granular percentile distribution like I have shown previously with my watchlist). But, again, I want an A (or an F) to really stand out and mean something because we want to really highlight when a stock has particular strengths and weaknesses.

Ultimately, we are trying to create a simple overlay (letter grades) that can be quickly informative and meaningful while still having the nice properties that the complex numbers offer underneath.

There are a lot of potential approaches to take but instead of paralysis by analysis we are just going to move forward with our best reasonable guess or idea, see how it works, and then look to iterate and improve it. I suppose then that all this text is me just putting ideas on paper as I continue to work through finding that best first guess!

Who knew making complicated things simple was so…complicated?