|
After collecting and examining more data, I think that perhaps results presented earlier were too optimized.
This is always a danger when seeking to model complex systems, and must always be kept in mind.
I have found that better results can be achieved by reducing the sensitivity of the networks (ie, signalling trades less frequently, when they get a much closer match) and increasing exit levels.
I am now using a stop-loss of 70, and taking profits after 90.
anomaly: Yes, I think I can get even better results if I use individual levels for each pair. I haven't coded this up yet, but plan to research at some point.
Also, like you mentioned, I have found my nets to be most successful at predicting EUR/JPY.
gamma_jammer: I don't collect that data right now, but its a good idea and I will prolly start logging that also.
|