To come to the help of the poster:
To explain how neural network works is not easy;
in a nutshell, imagine an articial brain as a black box;
you feed it various scenarios and outcomes from the past; heaps of them=> so basically historical data (to be defined)
and teach that brain to detect a positive outcome; after a while , the brain will tend to recognise patterns which should end up positive, or not;
Once trained, an ideal NN for each past scenario should tell you:
in that case, the day after, the ASX will be at xxx or the SP will reach , etc;
you then expect to feed live data to the "brain" and the brain telling you then what is (the most likely) to happen in that case;
Hopefully, this should give you an edge.
BUT
If the brain tells you: "you need to buy XX today and you will profit tomorrow", at no time can you (or the brain) explain why there is such an assumption.
What is important and can restrict the model is:
how wide is your data feed for scenario, do you include weather? Do you include news on the latest bomb in Paris, usually you do not, yet it is obvious this may affect the market.
And how do you restrict your domain?
for example: if my NN just care about AAC share price: Iinclude weather, AUD/US, ASX move...but are you going to fed the latest row between Indonesia and Australia?Will you detect the Mouth and Foot disease alert?
So even restricting the universe may not be enough to remove sizable risk from the prediction;
I hope this helps?
It is very interesting; i have projects right and left, startup work, innovative apps, and a life, but if one day I ever get bored, a NN in amibroker is in my TODO list