On optimum strategies for minimizing the exponential moments of a loss function