I'm asking this question because if it is feasible, then why are not we hearing anything from this? If not, then what are difficulties? Are pattern recognition hard to replicate using computers? Traders' gut feelings are not completely lay out, leading to failure in automate trading system? I believe if money can be made, someone would have solve this by now.
Yeah, your title implies a different thing than your actual question. There's no edge in patterns, there's an edge in trading patterns within a context of current market environment. As you said - gut feel (experience) is big part of the edge. I am not very knowledgeable about ML or neural networks, but I think it should be possible to take a successful day trader and record his/her thought process over a year let's say. This may capture enough contextual awareness to automate it. However, day trading is not scalable enough to be worth it. So that leaves us with people like Soros, PTJ, Kovner, etc and they adopt as the markets change as well as experiencing dry spells. And that leaves us with the fact that we are better off removing discretion all together and building robust models based on emperical evidence. Which is where the money is going I suppose.
What I mean by mirroring traders is to put successful traders thought process into a computer program. Not get the end result from signal service. I understand that day trading is not scalable at a certain point but I believe that a successful day trader can easily make $300K per year. Isn't $300K/year enough to motivate some1 to code or it costs more(millions?) to write that program? Or, successful traders are not successful in the long run, i.e. decades?
A single good programmer who is an expert in such field can easily make that in a year via salary. His/her time would be better spent doing research on something that can yield millions once deployed by large hedge funds. My understanding is that big money looks for stable and predictable growth, not high risky/unpredictable returns. People are unstable.