You need to remember that we already had a big correction in 2015, bigger than the current one (30% down from the top in DAX), and the pathfinder backtest performed well (i.e., was well optimized) for this period.
Isn’t above the point though? If we optimise over a big correction period then the System looks great as the optimiser works out variable values that give us a good profit over that big correction period (I haven’t seen the equity curve). Moving forward with that System through another similar non-optimised correction period (similar but not the same!) then trades may not perform anywhere near as good and so losses mount??
We always need to be aware that it it possible to develop and to optimize a trading system for a price curve that has been generated by tossing a coin. Heads up means price moves up by one unit, tails up means price moves down one unit. In such a price curve we will find trends, trendless phases, and many more characteristics we know.
Of course, even when we develop a sucessful backtest for such a price curve, it will never make any money in the future, because price movements are totally random and will continue to be generated by tossing the coin.
So, in my eyes, the question is : Are we able to find an instrument that statistically shows more trend behavior than can only be described by random price movements ? And can these trends be captured with indicators in the past as well as replicated in the future ?
