What trains the neural network?

Just curious what any of you neural fans use to train your neural networks. If anyone has an experience with them I would love your input. I am just curious what the common methods are and what may/may not work. Thanks.

I have started reading about neural networks for forecasting stock/forex prices a few days ago, so I am in no way experienced.

But I came across some great documents (bachelor, master thesis, scientific papers, etc.) and usefull books in our universities library.

To forecast the price for (e.g.) the next day, you feed the ANN with the open,high,low,close,volume values of the last 50,100,200... days. In addition to this, it can be useful to add the corresponding values of some moving averages, MACD, RSI or other technical indicators. Another approach is, to add values of correlated markets like gold price, oil price, interest rate, bond rates etc.

I believe this is the hardest work, that can only be achieved by trial and error:
- Which input data?
- How to transform input data prior to feeding the net?
- How many neurons?
- How many layers?
- ...

I am sorry that I could not give you any more help, but I have not really started research and testing right now. I am also interested in other peoples attempts on this.

I am currently working on similar projets.
I've written a library (fann4mt) in MQL that allows me to easily create and use ANN in a few lines of code. It is based on Fann2Mql, itself based on Fann.

In order to forecast the price, I tried to feed the ANN with the last x values of a moving average. I scale the values to the [0;1] scale of course (it's done by the lib).
It's not really accurate, but you can recognize the chart pattern in the forecasted output.
I also tried to feed the difference from one bar to another instead of the value itself. This is the worst method so far, as the results are random.

Last attempt to date mix the RSI, RVI, Stochatic, MA, and Volume.
Not really accurate neither, but still working on it.

You can find the page of the fann4mt library (including download, tutorials and various examples) here: http://fann4mt.thetradingtheory.com/

Let me know if you achieve anything!