Little update - loose pinbar backtest

My preliminary backtest on my mechanical pinbar system showed a profit factor hovering between 1.15 and 1.3. This week, I decided to stress-test the system with the entirety of data I had for the EURUSD and NZDUSD from 2001 to 2012. 

 

My profit factor turned out to be a little less than 1.0. It was almost as if I was trading randomly. As to the discrepancy, I'm not sure how to explain it. I suppose you have a system, and it performs okay 90% of the time. However, it may hit a pocket of "bad luck" 10% of the time where consecutive trades fail. Random sampling may not pick up on these pockets, especially if the pockets and/or sampling is too small. Either that, or I was just lucky with my preliminary sampling.

 

I added an RSI(10) filter to measure the trend. Pinbars tend to behave as trend-reversal signals, but they can also act as trend-continuation signals, as seen below:

As you may see, my criteria for a "tradable" pinbar is very loose. I totally ignore the prominence of the pinbar's "nose" with regards to to the previous candle e.g. the noses of the very first and very last pinbars don't even break the previous day's low. The open and close of the candle must also occur only in the top or bottom 50% of the pinbar, which is also very loose. I'll explain more if my backtest continues to be positive.

 

Adding an RSI filter improved my results significantly. Here are the results, filtered with the RSI indicator. Basically if RSI(10)>50 and we have a bullish pinbar, we're trading with the trend. If the pinbar is bearish, we're against the trend. Vice versa when RSI(10)<50.

Trading with a 2:1 reward:risk seems best. I used an ATR(14) * 0.5 for my stop loss. As you can see, going against the trend was very bad. The backtest so far suggests a frequency of 1.5 signals a month per currency pair, which is very good (trading six pairs = 2 signals a week), and compensates for the relatively low profit factor.

 

The next step(s) is to test the other pairs, of course. This may take awhile. Any suggestions on improving this system are welcome.