Creating software to predict stock markets

Also there is a product called SAP Predictive Analytics. As far as I know it comes below $1000 and you can feed into it any amount of data. So you can take the price series and all the additional information calculated in excel then load the data to this tool. If it works than you know what to program but if no models work, perhaps you save yourself a few months.
 
nonlinear5 said:
Java technology has moved on. For example, the JavaFX technology allows you to write java code and deploy it to pretty much any imaginable platform:

-- any desktop (Windows, Linux, Mac OS, Solaris)
-- web browser (IE, FireFox, Chrome, Safari) to run as a RIA web app
-- mobile devices (Android, iOS, Rasberry PI)
-- embedded devices (armv6/armv7 such as ATMs, security systems, storage appliances, imaging devices, etc.)



That was applicable in the 1990s. Not anymore. I run a Java app which handles and processes about 10 million market data messages every daily trading session just fine. Unless you compete in the HFT space where you count nanoseconds, you don't have to think about the speed of modern Java/C# runtimes.
More...

programming language should be the least of your worry, pretty much all modern language will do the job

you need to focus on the entire system design, which is far more complex to accomplish
 
I started to document my experience of developing algorithms for 'stock market prediction' or more specifically for generating strategies that successfully exploit price volatility. I started with ensembles of voting neural networks which I have written about in my blog:

https://theartificiallyintelligenti...5/12/13/algorithms-ai-and-volatility-trading/

Subsequently I have started to develop ensemble genetic programming approach but haven't got round to writing about that yet. (Unlike the neural network approach where I need to decide which technical indicators to use (as inputs into the neural net), genetic programming decides for itself which indicators are most relevant by evolving different buy/sell decision trees). For either approach having a voting ensemble - and having some mechanism for adapting the constituents of the ensemble - seems to be key to getting anywhere... at least for what I'm doing.
I do this for interest (though always open to offers to do it for a living!) as my PhD was in AI area and I'm probably a bit of a geek. Like all the best areas for applying AI, its a simple problem to state (i.e. make money) but quite difficult to create something that's successful... and lots of different AI algorithms can be applied to the problem. Obviously if you get something working well, you don't need to sell anything to anyone, just let your software run. Not quite there yet though!
 
Myself, with some help from two friends, built this site from scratch except for the charting package and an open source library for calculating trading relating functions.

http://fasterbull.com/

It can do some cool stuff and I use it for my own trading. Not really trying to make it commercially viable right now. All that being said, it took A LOT of time. Whatever you decide to do, take your time estimate for completion and quadruple it. If you are cool with that level of commitment, go for it.

Good luck!
fan27
 
Why are you reivnenting the wheel? I just want to know the decision behind that. There are tons of web based backtesters, stand-along programs and so on.
 
You must log in or register to reply here.
Back
Top