I know this has been discussed in the past, but I can’t seem to find the thread. What are people’s thoughts on simple vs. complex ranking systems and trading systems. Is a system with hundreds of nodes more robust than one with only 5 or so simple factors? Are overall trading systems with more rules more robust than ones with 2 or 3 buys and/or sell rules?
The answer I have found is that too many ranking nodes dilutes the effectiveness to the point where the output is unpredictable, meaning easy to optimise but likely to disappoint.
Commonly used metrics “price to something” ratios are, on the other hand too simplistic, I have had success by structuring more complicated formulas as nodes. Think about how the order-by happens - if it is in a formula then it is using “ordinary” mathematics, but if it is split into nodes then it uses the fuzzy logic sorting, including normalisation and etc. which I don’t think produces as good result.
My best performing Port over the last 2 years uses a ranking system with only 12 nodes that I created in Jan. 2013. However, my second best performing Port over the same 2 years uses a ranking system created in 2009 which has 123 nodes. It was created by simply adding together 4 ranking systems that had great performance in back test from 03/31/2001 through 07/20/2009 when I created it. So I think it maters more the quality of the factors & function in the system than the number of nodes.
Do you still have the performance from these systems and from 2015–2023?
Is it many of the same ones I asked about here? Fewer parameters make better models. - #16 by Whycliffes