Bayesian estimation of linear functionals in the case of multidimensional normal distribution (Q1365717)

Let \(x\) be a \(k\)-dimensional random vector, having the density \(f(x\mid \theta )\), \(\theta\in\Theta\), and let \(H(f)\) be some linear functional of \(f(x\mid \theta)\), where \(\theta =(\theta_1, \theta_2, \dots, \theta_m)\) is the \(m\)-dimensional vector of parameters. The construction of a statistical estimator \(\widetilde H(f)\) for the functional \(H(f)\), when the vector \(\theta\) is completely or partially unknown, given a sample \(x_1\), \(x_2, \dots, x_n\) from the distribution \(f(x\mid \theta)\), is an important problem. We often use functionals having the structure \(H(f)=E [\Psi(X)]\), \(H(f)= P(X\in D)\), where the function \(\Psi(X)\) and the region \(D\) are known. If \(\widetilde f(x \mid\theta)\) is a statistical estimator for \(f(x\mid\theta)\), then we have the substitutional estimator \(\widetilde H(f)= H(\widetilde f)\). So the likelihood and unbiased estimators \(\widetilde H(f)= H(\widetilde f)\) and \(\widehat H(f) =H(\widehat f)\) are substitutional estimators for \(H(f)\), where \(\widetilde f(x\mid \theta)\) and \(\widehat f(x\mid \theta)\) are the likelihood and unbiased estimators for \(f(x\mid \theta)\), respectively. We describe an approach for the construction of the substitutional method of obtaining Bayesian estimators \(H_B(f)\) for \(H(f)\). This method is used to obtain Bayesian estimators for the characteristic function, for moments of fixed orders and for the probability of linear inequalities in the case of a multivariate normal distribution.