Asymptotic approximations to the Bayes posterior risk (Q2641016)

Let \(\Gamma\) be a non-degenerate open interval of the real line R and let \(G_{\gamma}\) (\(\gamma\in \Gamma)\) be a one-parameter exponential family of probability distributions with natural parameter space \(\Gamma\). Let \(x_ 1,x_ 2,...,x_ m\) and \(Y_ 1,...,Y_ n\) be independent random variables, where the \(X_ i's\) have a common distribution function \(G_{w_ 1}\) and the \(Y_ i's\) have a common distribution function \(G_{w_ 2}\) for unknown \(w_ 1\in \Gamma\) and \(w_ 2\in \Gamma\). It is assumed that \(w=(w_ 1,w_ 2)\) is distributed on \(\Gamma^ 2\) with a prior density \(\Pi\). Approximations to the posterior probabilities of w lying in the closed convex subsets of \(\Gamma^ 2\) under the prior and prior density \(\Pi\) are derived. The given result is used to approximate the Bayes posterior risk for testing the hypothesis \(H_ 0:\) \(w\in r_ 1\) versus \(H_ 1:\) \(w\in r_ 2\) using a zero-one loss function, where \(r_ 1\) and \(r_ 2\) are disjoint closed convex subsets of \(\Gamma^ 2\).