Has anyone had luck with high frequency using either neural nets or genetic Pr?

We want to predict short term tick movement using one of these two methods and I am at the beginning of our modeling curve.


Any words of wisdom to speed up our process would be great.

Thanks

Hello,
I have developed a trading system based on fuzzy logic and a neural network, all custom built. My process is:
1) Strategies created from combination data mining/pattern matching , linear statistics, TA. - Back and forward testing of these untrained/unfiltered strategies to prove a solid foundation.
2) Fuzz logic filter decides which TA values and patterns in strats are suitable given current volatility.
3) Neural net simulation (using TABU learning algorithm) predicts which inputs result in a positive or negative movement.

I have had very good results in paper trading so far. I am in the process of fine tuning my overall trading system ( modifying the framework for more performance). I am setting up a co-loco at equinix at the moment and working out some legal/tax details.

Of course as I've yet to trade live I can't say I've made money, but I think it looks fairly promising. My one piece of advice would be to not overtrain your models!


Good luck!

Scalping to me is high frequency trading.
I prefer scalping.

However, I do all kinds of trading.
Yes, it works.

I am always profitable. Why? There are average people out there.
And then there are smart people out there.

Your success is dependent on you and the people you learn from.

I use to believe that 95% of traders will lose. I use to think that I will lose. However, this is not true. It is because I am not average.

Use whatever you feel comfortable with and are profitable with.

I'm playing around with a neural network in matlab and wanted to ask for advice with what inputs to select for the network:

- I'm using the following inputs:
- O(last)
- H(last)
- L(last)
- C(last)
- O(current)
- V(last)

for all JPY crosses, which are all exported into a csv file by mt4.

Last refers to last data set for a given time frame. I chose weekly, so last would refer to last weeks O H L C V, and current would be this weeks open, O(current).

I'm doing this for all JPY crosses targeting GJ C(current) as the only required output, the goal being to forecast this weeks (current) closing price. Not to say that this is what i will use as TP, but rather tell me if the NN indicates that price should go a little or a lot higher / lower.

I want to use this along with PA, TA, and some fundamentals.

What would some other inputs you could suggest? These guys use a bunch of indicators, but i am of the opinion that the information in the indicator is strictly a derivative of price, so it wouldn't be adding anything novel to the inputs.

One more thing i thought might be of interest to you. These guys used 6 different moving averages and argue that more inputs does not necessarily equate to better performance. In our case though, i think all JPY crosses are good inputs; what do you think?

Thanks,

Bwana.