This is a principle i apply in virtually almost every algo, unless I see very stable returns over a long time period (10+ years).
I don't really get what you're saying either, however I'd like to phrase it differently: markets move through different market regimes. In this case, the algo was optimizied specifically for a certain market regime, which contains certain non-random elements. However, once this particular market regime changed, the algo tries to look for the same non-random elements, which inherently are no longer there, and therefore finds random elements instead. Comments?
I only read the last page of this thread. I am sorry I wasn't aware your strategy worked for 3 years. In a typical parameter space a wide plateau is obviously preferable to a peak surrounded by red. I made several incorrect assumptions but moving past that and beyond my pet peeves about trading terminology I suggest altering your search from a parameter space if indeed that's what you were optimizing, and instead focusing on strong logic which you may already have been doing ...but this time require the logic be viable across a wide spectrum of the parameter space.