A while back, I wrote my personal trading platform using neural networks. I’m not exactly the best trader in the world, and modelling a trading strategy (for example, the best indicators and parameters for a certain instrument at a certain timeframe). So although the system “worked”, I think it was let down on my technical trading ability.
Fast forward to now, and I’m working on making the system more user friendly (i.e., not just me using it), and I’d like to see if people would actually be interested in utilising something like this. I know the fundamentals of the system are sound; how the networks are trained, their architecture, and how the data flows through the system, and I did artificial intelligence at University to back that up to a certain extent or another.
I’m at the stage now where the system works, and I’ve got friends and family testing it out, but before I do anything else with it, I wanted to see if the trading community itself would be interested in something like this? I’ve found it difficult to explain to traders about how neural networks “work”, and it is difficult to explain to AI people how trading works, so to say this is niche is a bit of an understatement really.
There are some free neural networks ea’s on the internet, but they are not as profitable as you may think. But if you have a unique approach on the subject, I studied neural networks in college and I am a trader and ea programmer and I can help you with what you need.
To answer, Yes of course we would be interested but we want to learn as well. It doesnt help if its black box. Else we would just be beta testing your strategy
I’ve spent quite a while building a platform, not a strategy, and I’m happy discuss a lot of what I have found over that period to a certain extent. If anyone wanted to try out the system, they would be setting up their own strategies, not testing mine. The system allows people to construct models through neural networks, and those would be owned by you, not me.
A few high lev things I have found, which will probably not be a surprise to people; nn’s don’t deal very well with the tails of a traditional normal distribution of market conditions (the Swiss de pegging CHF for example), and they don’t deal well with noisy timeframes (less than 1H).
The reason why most nn based ea’s don’t work is that they are pre-trained on a data set and then distributed to users. You have to follow the market in terms of training, and you have to also carry out forward testing of those networks on unseen data by the networks to optimise and kill off ones that are not “locked in” to the market.
Neural Network application in itself is not a difficult thing, how you get around these issues, and reduce that error rate before attaching those to a live market are the little things that make a big difference.
I may well regret this, but I think the system is ready for some beta testers to give it a go. If any of you are interested, you can go to inklin – bring the noise to create a neural network model of a strategy. It is still rough around the edges, but any feedback or questions would be most welcome.