Hello gents.

I wanted to ask what is the difference between the approach used in Zorro for building strategies compared to other platforms which used GP to evolve and create strategies starting from a main trading idea as we do.

I found many platforms around and I was wondering what is the difference between our approach and their.

Regards, Francesco.

Genetic algorithms are used by most other platforms, also by early Zorro versions. The difference is that they select the maximum profit peak in the parameter space, while Zorro ignores peaks and selects broad hills or plateaus instead.

Genetic algorithms always produce better backtest results, but we have made the experience that they normally produce worse WFA and trading results.

In simple words with GP I have strong probability of building very curve-fitted strategies that look very promising in backtest, but fail in real market conditions.

Thank you Jcl, Francesco.

hello France,

i believed jcl is talking about genetic algorithm and not genetic programming. programs which used gp, trading systems lab, adaptrade n chaos hunter to name 3 of such software.

of course i might be mistaken....

best regards.

-guan

If I do not say something wrong, I think there are two areas for using GP.

Strategy evolutionary process applies GP to create strategies that meet fitness parameter we chose.

Parameter space optimization used to define the best parameter that optimize a trading strategy.

In both cases I have just a vague general concept.

Regards, Francesco.

GA is more suited for the second thing you mentioned.
Basically, if you can formulate the structure of the solution, use GA. Structure can be a fixed number of parameters to optimize.
On the other hand, if the form of the solution is open ended, you can try GP. But be warned the problem space is very large. Even Zorro is probably not fast enough to give you your fitness measurement efficiently enough.