I'm just starting out the cheap way: generating different strategies and running stochastic simulations.

No need to be stochastic (with deterministic strategies, anyway). There are only 128 different ways to place the hats, faster and more accurate to just test each of them once.