fitness function for Genetic Algorithm

New comment
 

Hi all,

I'm revising the fitness function of my genetic algorithm implementation and so far I've only had the profit to be the only criteria. That means the fittest lifeform/strategy is the one with the highest pip count.

Clearly this needs improvement. So for now I have:

- profit, in relation to the maximum profit of all lifeforms in the population

- maximum number of consecutive loosing positions, in relation to the maximum number of consecutive loosing positions of all lifeforms in the population

- maximum loss of consecutive loosing positions, in relation to the lifeform's profit


I would like to hear your opinion on that. What's your definition of a _successful_ automated strategy?

 

You'll perhaps get as many different ideas of 'successful' as members here.

Mine is 'preservation of capital' e.g. I'm not prepared to risk $100 to make $5 profit (that's how some methods can claim 'over 90% success rate')

Another factor to consider is sometimes called 'velocity' - how quicker the EA turns over its capital (related to how many times the EA trades) e.g. for the same risk, $10 profit 100 times a month is better than $100 profit 5 times a month, even though the second method has a higher 'profit factor'

I also like a method that is not too sensitive to 'tuning' e.g. If varying my SL from say 10 to 11 to 12 pips results in loss $1000 to gain $5000 to loss $2000, I'd be very suspicious of thinking that 11 pip SL is going to be the Holy Grail.

 
Does it make money without keeping me up at night.
New comment