August 6, 2013

Some don't seem to ask the most basic of questions. One would be: why should one “need” to re-optimize in the first place? High on the list of answers: I would bet that the “automated” trading strategy broke down for some reason or other. This could only mean that the trading strategy has been losing money for some time and still is; and for sure, needs to be stopped, or at least modified in some way, if not replaced entirely.

If the automated strategy was not losing, evidently, the strategy was winning. Then why would one want to change it? The only acceptable answer to this would be to improve and do even better, otherwise, it would not make much sense: hey boys and gals, let's improve on this script so that it makes much less money, sure...

When re-optimizing, one is mostly re-shuffling parameter values thereby producing an incrementally different version of the previous running program. The outcome going forward could be a performance improvement or prove to be another failure. The more you optimize on a given theme, the more you should tend to reach a local optima in backtests, but this local optima might correspond to absolutely nothing if implemented going forward; meaning that it might not change the real problem which was that the original trading script was poorly designed and needed to be changed or replaced no matter what.

Another reason to re-optimize would be to adapt to changing market conditions. This would imply that you have means to identify “changing market conditions”. But if you know that market conditions are going to change, why not program them in your trading script in the first place? Why wait til after market conditions changed? Also, it should be noted that after having adapted to these “market conditions” you might still face more “changing conditions”.

But still, those re-optimizing to “follow” changing market conditions do so because the trading strategy was losing money as it evolved in time and hope that adaptive methods to unforeseeable market conditions could improve performance results. But there is a dichotomy here: adapting to “unpredictable” “changes” in market “conditions”; what does that look like?

What could regulate short-term automation if randomness is a significant part of the equation? This is like what kind of prediction can I make “now” knowing that the near-term future is quasi-random in nature? Should I make a quasi-random decision or what? And if so, am I not trading randomly on random-like changing market conditions?

Technically, the only real good reason to optimize a trading strategy is if it was already generating good results. And in doing so, meaning optimizing trading procedures with potential, you could improve your overall long-term performance level. Using payoff matrix notation, I would translate this into Σ(H(new).*ΔP) > Σ(H(old).*ΔP). May I say I prefer: Σ(H(new).*ΔP) >> Σ(H(old).*ΔP) for those that can see the difference.

Created... August 1, 2013,    © Guy R. Fleury. All rights reserved.