These systems are too large and too chaotic to predict well, at least with human minds and with AI at its current level of sophistication. The latter also suffer from various biases -- the most obvious being the bias against surprising developments -- which hinder them.
Asimov's psychohistory needs something much more powerful than we have yet built.
So it seems to me that you can get interesting "status quo extended" scenarios, but as actual models they are near useless.