There is no clear total answer to that because it's going to depend on each different portfolio. We do know from a Michael Kitsus article that at least for ordinary portfolios about 20% relative deviation from where they started tends to be a decent place to do a rebalancing as long as you're not scheduling the rebalancing too often. The math isn't quite that heavy although if you did go into the particulars of the efficient frontier calculation it would get pretty hairy. Real wrath of God type stuff.