I have just found this forum and think that perhaps I can add some of my own experiences to this.
I have been walking the NN (and wavelets) path some 7 years now - only the last 2 and a half years have I managed to actually make some profits.
For me, NN's are good (wavelets a tool without which nobody can make a success) - and then, the market will change pace.
When historical data is used for training (your NN), there must be an assumption that in future the market will behave in the same way - this is not the case because, we have short term and long term behavior patterns.
By it's very nature, NN's will tend to train more on short term patterns - they occur much more frequently.
As such, in some way, my problems may be the opposite to most other traders - during periods of consolidation, my system will select turning points with a high degree of accuracy and, as a result my account will rocket up.
During trends, the system will attempt to continue with the pattern - indeed, the market will continue in the direction of the trend.
As a result, my account will grind to a halt.
Talking about the EUR, last year, I made exceptional profits - then, from Aug-Sep the market started trending (to the 1.36 mark) and the system would have emptied my account within about 3 months - as it is, I took a 23% dragdown during this period (before turning off...)
This is exceptionally hard - most traders are dancing on the tables, making good profits while the trend is going...
However - in between (the trends) there still may be months (even years) during which a proper NN system may yield exceptional profits - in March, I made up my dragdown within 2 weeks and since then, trading has been reasonably good...