AUDJPY shaved candle system - swing and a miss...

I completed my backtest on the AUDJPY which ended dismally. The system performed well throughout 2010-2012, but then fell apart in late 2012 and throughout 2013.

I'm quite annoyed since the backtest started out so well, but then began to smack me in the face, time and time again, as I plowed through the home stretch. That's the nature of trading. 

My system is nominally profitable, but there's a steep decline in the equity curve in the last year (this is for all trades from 8AM-2PM GMT). 

Here's the equity curve if we trade against the trend. (I used RSI(14) to measure the trend, with RSI(14) > 50 suggesting a bullish trend, and RSI(14) < 50 suggesting a bearish trend).

The drawdown isn't so bad. 

TIME ANALYSIS

Hour / Market session

I found a few interesting results when I sorted my trades by time. Here are my results sorted by hour:

Here's the same information as a bar chart:

We can see that the shaved candle signal performs best during the European session, which I found surprising, being the AUDJPY. I had hoped that this would perform best during Asian / Australian trading hours but I guess not.

Weekday

Here are the results for each weekday.

I'm not sure if this information is particularly useful. I expected Monday to perform poorly since Mondays traditionally tend to be quiet, but I also expected Friday to perform poorly for the same reason as well. 

Conclusion

This isn't a system. But what I have learned is that the European session seems to be best for intra-day trades. My next plan is to test the EURUSD during 08:00 to 14:00 GMT. Because I'm only testing six hours per day, this backtest should be alot quicker than the AUDJPY. I would especially like to see how large shaved candles perform when going against the trend.