Subject: Re: ML for MI question
Thank you for that list!
I'm glad someone did it, i.e. throwing all the random forests, neural net, new stuff against trying to predict stock returns from fundamentals, "Empirical Asset Pricing via Machine Learning" paper:
https://papers.ssrn.com/sol3/p...
But IMHO, the goal seems a little mis-guided.

The LLM stock screen is amusing.
But I wasn't thinking of asking an existing LLM to provide a stock screen, but to train an "AI" to identify "dog" stocks from e.g. "cheetah" stocks.

This is just blue-skying:

I spent a couple of minute online, so no doubt there's something better I missed, but this one at least exemplifies a ML framework that one can fairly easily train on one's own data
https://developer.apple.com/ma...

An amusing story of a guy using the above Apple thingy to identify his cat, using his rather small "database" of pics of his own cat, from a bunch of online cat pics
https://www.linkedin.com/pulse...

An initial thought is to translate all finanical data such as P/B ratios etc into categories e.g. "very low, low, medium, medium high, high" because LLMs aren't great with numbers/arithmetic. This
should be carefully done because e.g. certain sectors tend to have low P/B while others high, etc, so a little thought put into the categorization could be useful.
After translating numbers into categories, then can use text AI's.

With the Apple one, or something similar, could use its "text classification" mode, "word tagging" mode, or "tabular" mode. In classification mode: instead of "great movie" or "bad movie" from reading reviews it'd output "great investment" or "bad investment" from reading (categorized) fundamentals etc. In "word tagging" mode: instead of "iphone" it'd tag "good investment", etc.

Cats from dogs?
https://www.zdnet.com/article/...
I'm not sure you'd want to fake an existing purpose-built image classifier into doing the finance problem, but maybe one could try. Instead, you might have much better success if build your own AI using some tool like the Apple one to identify, from financial features, the dog stocks from cheetah stocks.