Page 1 of 1

AI/ML assorted questions

Posted: May 9th, 2022, 5:00 pm
by Mercadian
Hi All,

I have to admit that I'm not very familiar with the specifics of AI and ML, I have a 100k feet understanding of it and I'm considering whether or not I should devote time to learn some of these tools.

The way I see it its a combination of Applied Statistics, Probability and Computer Science which created their own band/camp.

Probably a very small minded view hence why I'd like to ask some assorted questions:

1. Your personal opinion of AI/ML is it all hype?
2. Do you see valid/sound usages in Finance, Economics?
3.  Any references further reading you found worthwhile? 
4. finally, on a tangent, if you had a time constraint and had to choose between learning AI/ML vs Complexity/Chaos which one would you pick?



Re: AI/ML assorted questions

Posted: May 12th, 2022, 2:04 pm
by Alan
Since nobody answered, my two cents. Re 1, I think the interesting application is autonomous driving, which I think is borderline partially solvable with ML. So-called Level 5 autonomy may require artificial general intelligence, which is nowhere close, but autonomous trucking over interstates from hub-to-hub in the US southwest might work. (TuSimple, etc). Re 2, supposedly banks were using it for credit card fraud detection -- don't know if real or hype. I think an interesting legit finance project would be "postdiction": classifying the cause of significant stock moves in the absence of "known news".

Re: AI/ML assorted questions

Posted: May 12th, 2022, 9:01 pm
by bearish
I think there are various applications of ML techniques in finance, especially if you don’t just try to beat the market (that’s hard!). As a somewhat simpleminded example, most fixed income risk management systems require a bunch of categorical attributes to be available for each bond, loan, derivative or whatever. These would be things like industry/sector and credit rating. How do you deal with missing data, which may be more common than you would expect? A very reasonable approach would be to apply some ML classification scheme based on the well behaved bonds, and then hope that there is not some systematic reason behind the missing data (like all the unrated bonds being from companies who failed to pay their rating bill, or all the N/A industry bonds being from crypto firms - that would be bad).

Re: AI/ML assorted questions

Posted: May 16th, 2022, 1:27 pm
by Mercadian
Hi Alan, Bearish,

Thanks for your thoughts, yeah I've seen some retail banking applications of AI, also some ANNs used for prediction based on text input only (ratings, fed moves etc) as well as networks trained with valuation models to speed up PFE, VA and other risk calculations.

The thing is I'm still on the fence about the soundness of the methods, yes they'll find a place in institutions as they dramatically speed up processes with cost reductions making them irresistible for End of Year reviews,  management consulting pitches and the like.

I think I need to do a lot more research into where it does and doesn't work for the right reasons.

Any thoughts on complexity/chaos theory? I'm very interested in this area as well.

Thanks again.