Trying to predict the stock market with AI


(Kenny Grant) #1

An interesting thought. With the recent progress in machine learning (which is now very effective for limited tasks like recognising pictures, faces or patterns in limited datasets), there may be areas of the stock market reaction to news which are predictable enough to make money from.


(Alex Sherwood) #2

It seems like this would eventually become self-defeating - eventually the AI would be trying to predict another AI’s reaction to the latest news. But if you can make it work now, while it is still human’s reacting to the news - & I’m sure there’s companies that’re doing this already - it’d be pretty cool.


(Kenny Grant) #3

I think people have the same concerns about momentum investing (following the market), but according to people I’ve spoken to who do that sort of thing, there are nowhere near enough bots to have any significant worries about them taking over the market. People are pretty irrational and unpredictable, and influenced by other things unrelated to the market/companies, so they probably make it more chaotic and difficult to predict if anything.


(Dave Smith) #4

It would probably be relatively straightforward to train a neural network to recognise chart patterns, what decision to make based on these chart patterns is another matter.


#5

Not sure it would be self-defeating, one company would create a better AI, which includes what other AIs are doing in it’s calculations, then another would create an even better AI, which includes what that AI is doing in its calculations etc…

Basically you get a load of companies racing to make the best AI, which should (hopefully) increase how good AIs are at other tasks too, leading to much better AIs.


#6

Isn’t this what most of the market trading has been for years already? ie the high-frequency trading stuff. Lots of algorithms, made by people or machine learning.