I have implemented a algorithm to find (by brute force) the best combinaison of indicators for crossover.
Here is some example of gains with these combinaisons :
I think that if I use the same technique but over 3 or 4 years of data, the result should be interesting because the probability that the gain is here by chance is lower.