Also - I don't know what your running implementation of ONS is. I'm working with something similar to what Sphin posted. I would love to see what you are using if it is different.

BUT - what I wanted to say is that I get significantly better results if I lag the HH and LL functions by 1. I tested it on 14 pairs (from 2011 to 2015). On average I get 7% higher AR when testing with no money management set. I did the whole test on M1 data and tried it with TICKS only for two pairs. Those tests confirmed the previous results. I hope I haven't overlooked anything important.