My research methods tend to be convoluted, just like my thought processes. But basically, I am optimizing on OOS data, to simply look for clues and inspiration to reverse engineer a better method of choosing OOS parameters for a specific trading system. The thought being that if one never re-optimizes on the OOS data, and studies the results, how would you ever know that your system truly chose the "best" values, during the IS step, for the OOS data?
The danger of course is one of curve fitting, and engineering the trading system based on this "glimpse into the future". But my approach is general enough that this should not happen. It is simply a method to gain clues on how well the system is choosing parameters, as opposed to relying solely on the equity curve, and associated reports. I am still in the middle of this experiment, and have yet to come to any solid conclusions. It merely looks promising at this juncture for one of my unorthodox systems, yet failed to yield anything fruitful for my other, more normal, systems. --- In [email protected], Howard B <howardba...@...> wrote: > > Greetings Ozzy, and all -- > > Optimization on the out-of-sample data? Tell us more. > > Thanks, > Howard >
