7. So far we have treated the sample mean vector and covariance matrix as fixed when considering the risk of a portfolio. Stated differently, estimation risk has been ignored. A methodology for taking risk due to estimation error into account was proposed by Greyserman, Jones, and Strawderman

(2006). Assume that the vector of returns R is N(μ, Σ) distributed. Let

(μ(k)

, Σ(k)

), k = 1,...,K, be an MCMC sample from the posterior distribution of (μ, Σ). For each k, let R(k) be N(μ(k)

, Σ(k)

) distributed. Then R(1),..., R(K) is a sample from the posterior predictive distribution of R and take uncertainty about μ and Σ into account and K−1 K

k=1 U{X0(1 + wTR(k)

)}