Your worst short signals may be great long signals (and vice versa)

This is an idea I keep revisiting as I try to design new signals - I'll often be trying to devise a short signal based on a range of indicators and historical data, and will, somewhat at random, come up with an appallingly bad signal. The sort of short signal that, if used consistently over a few years, would cost you thousands of pips. When I stumble on one of these stinkers, I always have to remind myself that what I may actually have on my hands is a great long signal. After charting its results as a trigger of long trades to determine its consistency, whether it has too many big drawdowns, and whether the results look unlikely to be some weird fluke, I may then put it into service on the long side.

I'm posting about this today because I just came across a signal that performs very badly as a short trigger: over 106 trades it lost 888 pips. I designed it using Bollinger Bands with a 10-day period, and it attempts to predict what happens if today's close was below the upper Bollinger, and yesterday's high was above the upper Bollinger. Turns out it's a terrible predictor of short trades, but it may just be a decent signal for long ones. Time to do some more testing...