Interesting article...didn't you overfit here? You found the numbers to choose from the same data on which you then retroactively tested your method.. If I have understood you correctly, you got an over-optimistic answer simply because of overfitting here...