whred into

Hello. I think I would like to put together a small automated trading system. There are still a lot of things I need to learn about, so I thought I would come to ET and ask around.

I am mostly interested in making an AI-based Forex trading system. If I am going to trade it myself, it will generate "EOD" signals at 10pm. If I am going to make something for C2, it could be an intraday system. I have historical daily & intraday Forex data to train a system on, but I do not yet have a real-time feed. It will be a while before I need real-time data, but I want to know what will be available for when I'm ready.

My primary question right now is, what are the better intraday Forex feeds, and what are there costs? If I develop an intraday system, I would probably be looking for 5M bars available over an API (ASCII or XML). My system pulling the data down would be a Linux system, and not a Windows system. I'm not sure yet if I am going to need index data. My FX broker has a $600/month feed, and I am starting to look at eSgl.

I'm not exactly a newby. I've occassionaly helped professional traders with their own systems, so I'm a little past the point of wondering if my task is even possible. My primary strength is a gut instinct about what will probably work. I can share ideas about self-learning systems unless there is a practical reason why I shouldn't, or a legal reason I can't.

My primary fear is that my mass of historical data will somehow be incompatible with the real-time data I would be subscribing to. I have already had to fix a lot of "holes" in my EOD data, and I suspect there will be even more in the intraday data.

On the technical side, I will be using the open-source FANN neural network library, unless I have a technical reason to roll my own. There appear to be a few open-source projects that I will be able to leverage. Does anyone here have any experience with QuantLib or TA-Lib?

I was also curious if anyone has any experience with C2, and if they had problems trading with them on the smaller time-scales (5M, 1M, or tick). I have a bunch of historical 1M and tickdata, but I'm guessing that would cost a lot and be hard to trade. Sorry for being a little vague, I'm not exactly sure what ET's rules are on mentioning the names of other websites or companies.

Let me know if you have any suggestions or questions.
 
Back
Top