Genetic evolution – combining long only with short only
10 replies
mikeyc
9 years ago #112942
Hi Mark,
I generated a set of strategies that trade long only, and another set that trade short only. I then loaded these as an initial population for genetic evolution, and set the strategy settings to trade long and short, and unchecked entry and exit symmetry.
Under these conditions I was expecting the evolution to combine the best long and short strategies, but every single resulting strategy created in the databank is long or short only, not long and short.
Any ideas on why this doesn’t work?
Thanks,
Mike
Patrick
9 years ago #127759
Hi Mike,
this is because evolution works different way. check this article: https://strategyquant.com/articles/how_strategyquant_works
If you want to trade long and short, the initial population has to trade long and short.
Best regards
Patrick
Mark Fric
9 years ago #127772
yes,it doesn’t work that way Mike. I’ll think about this idea, it should be generally possible to use long only strategy, combine it with short only strategy and have strategy with rules to both sides.
But long and short rules influence each other, this strategy would trade differently that you’d expect.
Mark
StrategyQuant architect
matka
9 years ago #127788
Interfering with strategies in this way make me think about symmetry. Does symmetry matters any more if we can manipulate it so easily? But isn’t symmetry the only countermeasure for trend fitting bias?
mikeyc
9 years ago #127791
I’m concerned that SQ doesn’t handle symmetry in an intelligent way, especially for custom indicators.
Even for built in SQ indicators, if SQ selects a rule for long entry such as:
Price > KeltnerChannel_Upper(71, 2.0) + 51 * ATR(33)
What should the symmetrical rule be for short entry?
I would say:
Price < KeltnerChannel_Lower(71, 2.0) – 51 * ATR(33)
I’m not sure SQ3 is that intelligent to create the correct opposite entry/exit rule….
Mark Fric
9 years ago #127801
yes, symmetry works for build-in indicators, but not for custom indicators – because when you have custom indicator A you’d have to specify which custom indicator is a negated version of A.
Maybe it will be possible in the new version 4, but I’m more inclined to just allow SQ to be extended with new programmable indicators and not use custom indicators with data import like in current version at all.
Mark
StrategyQuant architect
matka
9 years ago #127807
Sorry I was talking about this symmetry
Symmetry criterion to maximize strategy symmetry. Symmetry value is in %, and it is measuring how much is the Profit/Loss for Long direction similar to Short direction. For example, if strategy makes $600 on Long trades, and $400 on Short trades, symmetry in this case is 66%. ($400 is 66% from $600). If the strategy makes the same profit on both directions, the symmetry will be 100%. If one of the directions produces loss or 0 profit, the symmetry will be 0%.
Maybe I am wrong but this is the only weapon against long time trend curve fitting (eg. having only longs on fdax). I think cutting and pasting strategy parts like this is not in line with symmetry criterion concept.
mikeyc
9 years ago #127810
yes, symmetry works for build-in indicators, but not for custom indicators – because when you have custom indicator A you’d have to specify which custom indicator is a negated version of A.
Maybe it will be possible in the new version 4, but I’m more inclined to just allow SQ to be extended with new programmable indicators and not use custom indicators with data import like in current version at all.
I agree Mark, I’m inclined to handle custom indicators are programmable calculations, the data import solution is very cumbersome and inflexible.
I would re-implement all my custom indicators as code for SQ4.
you should cache the results in SQ4 memory for speed though, so it isn’t called over and over again for the same parameters.
Patrick
9 years ago #127814
guys, i dont know if you have any results already, but its better to focus on building strategies than looking for mistakes….just use SQ how it is, its good enough…
mikeyc
9 years ago #127815
guys, i dont know if you have any results already, but its better to focus on building strategies than looking for mistakes….just use SQ how it is, its good enough…
I disagree, it is foolish to trust your money to a system without understanding what it produces and how it produces it.
It is also very important to improve the system at every opportunity, to give a greater trading edge and confidence in trading.
Patrick
9 years ago #127816
my view is different, i invest money to PC and software and i want to earn it back. It is like bussiness. Thats why i focus on things, that brings money first-which is now to produce as many good strategies as possible and educate myself in ea trading.
i think it has no sense to create this strategies because they will probably not go trough robustness tests. so i think its loosing time when i can create one side strategy or symmetry strategies.
SQ works similiar as other programs for algotihmic trading and people are earning money with ea created by these programms. so there is no reason to not trust to SQ software if you trade robust strategies.
Viewing 10 replies - 1 through 10 (of 10 total)