Asymptotic Variance as an Approximation to Expected Loss for Maximum Likelihood Estimates