The arithmetic vs. geometric mean distinction is a good one to note. Since you mention risk-adjusted return, do you favor any particular approaches to optimization? ReSolve makes a pretty strong case for numerical optimization, summarized by this decision tree based on prior beliefs: