Hi Hugin, Most state based models that I've worked with or find in literature/internet fall into the 1st description; i.e. conditional gating or regime modelling, whereby some systems have a better chance of thriving under a certain regime. That is not to say the other approach might not be useful either.
Ok. I guess that only some simpler transformation will work when including state decision variables in the model. And if they have a heavily non-linear impact then it will probably not work. One other potentially good thing with not having market state included in the model and instead use filtering/gating is that you "save" a number of parameters in your model.
Keep in mind that evolution does not really pick "superior" species. It picks the species better adapted for the CURRENT environment. We tend to think of evolution as a progression toward an ultimate species (i.e. us) but that is us just being really conceited. The only progression of evolution is toward more complexity, not superior beings. I think if you apply GA to trading you will end up with more and more complexity geared toward a fleeting environment. Which is really just curve fitting. I'm no expert though, I just really like the book "Full House" by Stephan Jay Gould.
A while back, they started saying programers would make more money than traders maybe soon they will start saying geneticists will make more money than programers Help Wanted: Geneticist with good prgramming skills, and a knack for trading
Exactly my contention as well. Best to develop a premise, test it, modify it slightly, and there you go: decent trading strategy that works over time.
I agree that lots of people complain about the GA/GP blender approach to data mining, but I've found that feature selection (or narrowing down the ingredients with some heuristics) tends to work better to start the modelling process. And as pointed out earlier, there have been a lot of improvements with respect to curve-fitting problems in the machine/statistical learning space at least.