So, to restate what you are saying in your quote, he looks at the variability of results on the out of sample data, concerning different parameterizations on a single model?
Not that i was asked..but... He's explicit about use of many models, but doesn't elaborate on multi-parametrization of single models. I suspect he uses both; I do. He is clearly a fan of ensemble methods. And for me, intuition and framing the problem (sometimes referred to as feature engineering) is an incredibly important part of the problem. Good to see you around.
Sorry, I get tons of thread subscriptions and spam to wade through. Try my blog email; it is fairly clean. Cheers.
I hope your post to the OP helps him. The "chain of numbers" and "chain of random events" you mention is nearly correct. I read all the posts in this thread. taken together they could solve the mystery you all have on your hands. It is very difficult to "back up in thinking" and get back on the true path. So I guess this misery that people create for themselves will never change. To get a solution it could look like climbing a pine tree to its top. In this thread each person has climbed up some and then crawled out on a limb. Ends of limbs is as far as that person can go in his trading life. As stated here "feature engineering" is what allows a person to climb higher and higher. I probably should quote each person's contribution towards getting to the top of the tree. At least the OP see's that there is a path to the top.