Id just like to run something past the general user population in terms of the best way to create genetic strategies. I would be interested to know whether seasoned and successful users agree with how I see it or otherwise how I can improve my approach.
My ultimate goal is to create strategies on different currency pairs which (i) pass the ranking tests and (ii) pass the robustness tests and (iii) combine those different strategies into a portfolio approach.
I understand that we first need an initial population, which you can create and load into the initial population tab. From there, run the genetic evolution algorithm and wait for the breeding frenzy to take place. The initial population needs to be a strong population, do you agree? As strong as possible so that the offspring born into the results tab are good specimens. So, how do we get a strong initial population?
Approach 1 – Identify a number of core ranking options which should apply to each specimen (e.g. net profit>0; SQN>4; Stability>0.6) and then add an additional ranking requirement (win/loss; DD; Monthly Profit etc) for each set of 500. Once the 500 are created, take the top 10% ranking from each set and add to the population. Once you have an initial population of 500 or 1000 use this as initial population and take the top 10% every 1,000 specimens. Refine down. The numbers are approximate and its the methodology Im interested in, really.
Is this a messy way to do it or is there a way to get the 10% automatically?
Also, should we be applying robustness tests for the initial population; will this make it more likely than the final results will also pass the robustness tests?
I realise there is probably no “correct” way to approach it and there is some trial and error, would love to hear how other people approach this.
hello I’m not an expert….. but I have been a trader for over 25 years position trader W.D.GANN , ELLIOT and I used digital filters to clean the data as an approach…. if the trader knew definitely what are the parameters to consider to have a winning strategy over time… would be the holy gral…. but I saw that P/F rett/DD stability… etc. etc. are not enough to be almost certain… then filtering the initial population too much you risk having so many identical strategies
because GA works on the lows and ‘once it finds a good population continues on’ that doing so many almost equal strategies… my opinion is not to filter too much the initial population but to use a large number of population…. I repeat it’s my observation
thank you Gianfranco
i am not using initial population at all, neither any filters for the initial population
i start my genetic building with random initial population without any filtering…i think that the genetics itself is here for me to made this job
Thanks, that is really useful.
Just one more thing that I am unsure about, when using the genetic approach … Are you putting anything into the initial population tab at all, or are you just running the algorithm and having the program populate the results tab directly?