Hi, I am considering building an AI web service that would help analyze trading data. It would be something similar to AzureAI / AutoML but simpler (no experience in AI required). A service that’s custom build for traders and investors with some very specific features that traders need. I am a system trader myself. I use AI to help me select trades and trading systems. I work mainly from the system’s console, which can be difficult for non-programmers. A web service would be a lot simpler to use, just point and click in a browser. Let me know if you would be interested.
For me, Andrew Ng's explanation seems ok: it's a paradigm shift. With algo you have input -> create algo -> get results. With AI: input -> results -> create algo.
You're unclear about what business problems you are trying to solve: Alpha extraction Risk management ... something else? Also I scanned your recent post history, I've assumed you learnt the last decade since that drawdown that the major red herring in this business is curve fitting.
That would be up to a user. The problems I've solved* for my self are: 1. selecting trades that have more chance to be profitable, and 2. selecting trading system that has more chance to be profitable. Every user would have to decide for him/her self and bring their own data. *) kind of, there is some progress but it's not a Holy Grail.
If you're just doing the project as a hobby and you really don't care if it succeeds or fails, that's one thing. But if you're serious about providing traders with a cutting-edge AI service that you want to turn into a real business, then here's some advice. Most guys that have built specialty trading software over the years make one critical mistake: They put all of their time, money, and effort into building the software but along that timeline they never even thought once about how they were going to market it. So when the product eventually gets finished, they have virtually no money whatsoever to advertise their software so traders will actually know about it and more importantly, how to use it. So eventually the whole project just withers away slowly. They also vastly underestimate the amount of marketing time that's required. Reputation is everything in this business, so you can't just throw up an ad for a month or two and then just stop. If you do it that way, you'll just shoot yourself in the foot because you'll be quitting before anybody even knows who you are. In most cases, it requires years of nonstop marketing and product demonstration to even get a new piece of software noticed and respected by the public.
"more chance to be profitable" tells me all I need to know. In order to help you: how do you separate trade systems that have a high chance to appear profitable from trade systems whose expected profit is above zero?
Thanks for the advice. Yeah, I know marketing will be a nightmare. The thing, is I don't really care if it's successful. Worst case scenario is that I will go back to running it directly from a console or even IDE. It's just that it *could* be useful to others and I'm trying to gauge the interest before spending months building something that nobody wants.
Based on things like expectancy, profit factor and so on. AI can figure out rules that are slightly better than my guess work.