If you're genuinely modeling complex systems with butterfly effects and uncertainty, you should sometimes be wrong in both directions. Sometimes things should be worse than predicted, sometimes better. If you're consistently wrong in one direction, that's not complexity - that's bias.