We’ve made available FRank and FMedian in ranking systems stock nodes. While it may be counter intuitive to provide a ranking function ‘FRank’ and a function that only returns one value ‘FMedian’ (or a handful depending on the parameters) they have very interesting applications.
Please see the attached image. It used 5 binary functions that return either 100 or zero if the factor or formula for a particular stock is above the median. In essence each node acts like a binary switch giving either full score or 0 if the factor/function value is above the median. We think this has a lot of value as well instead of straight ranking of each node.
In the image the test was done only in the DOW 30 stocks, and only Walmart passes all conditions:
Price2Sales less than median of the DOW
MktCap greater than the median of the DOW
1Y Sales Growth greater than the median of the DOW
PE less than the median of the DOW
1Y Return greater than the median of the DOW
Several other variations of these theme can be done with FRank, with nodes that return 100 if the value is in the top decile, for example.
Let us know your thoughts and ideas and we can follow up with a new systems or portfolio or a case study blog/tutorial.
Marco
NOTE: when editing a ranking node you will not find the functions listed in the drop-downs. They have not been added yet as part of the ‘vocabulary’ for ranking system nodes. For parameter and usage reference see the function in the screener under FUNCTIONS->RANKING & SORTING
Please note that these nodes are quite expensive in terms of processing. We’ve limited the number of rank and median functions in a single ranking system to 20 or less. Let us know if you see severe performance issues.
I disagree. In part, I don’t really understand the thought behind why you think only binary nodes make sense. It’s normal in prediction work to avoid binary explanatory variables unless the underlying information is truly of a binary nature.
But here is how I will try to use FRank:
First, I will try to take advantage of the flexibility of using a formula to replace performance relative to a broader market benchmark with performance relative to stocks with similar market capitalization and maybe similar beta.
Second, I will want to make relative performance a more positive factor when the similar class of stocks have recently been down or flat and a more negative factor when similar class of stocks has lately been on a good run (say over last 50 days). I anticipate these factors will work better as quantitative ones rather than arbitrary binary cutoffs.
“We also think that these functions only make sense in binary 0/1 nodes. Let us know if you disagree and why.”
Marco - I disagree. The FMedian function is a very powerful means for determining industry and sector characteristics. Here is a simple way to evaluate industry strength:
