Math isn't a requirement. There are plenty of examples of a dart-based portfolio (sometimes thrown by monkeys) beating or matching a beskpoke, artisanal, investment professional stock picks.

What a great description of decisions driven by LLMs.

even so, unless I misunderstand... LLM are most likely to pick what they've seen the most, ie: what is the most common recommendation it's seen during training; barring rlhf anyway

I published the all the methodologies and data, do take a look if you're curious.