Bayesian Analysis Bayesian Computations推荐.pdfVIP

Bayesian Computation s Bayesian analysis requires computation of expectation s and quantiles of prob- ability distribution s that arise as posterior distributions . Modes of th e densi- ties of such distribution s are also sometimes used . Th e standar d Bayes esti- mate is th e posterior mean , which is also th e Bayes rule under th e squared error loss. It s accuracy is assessed using th e posterior variance , which is again an expected value . Posterior median is sometimes utilized, and t o provide Bayesian credible regions, quantiles of posterior distribution s are needed . If conjugate prior s are not used, as is mostly th e case these days , posterior dis- tributions will not be standar d distribution s and hence th e required Bayesian quantities (i.e., posterior quantitie s of inferential interest) cannot be computed in closed form. Thu s special techniques are needed for Bayesian computations . Example 7.1. Suppose X is N{0^a^) with known cr^ and a Cauchy(/i,T) prior on 6 is considered appropriat e from robustnes s considerations (see Chapter 3, Example 3.20) . Then niO\x) oc exp {-{9 - x)y{2a^)) (r^ + (^ - / i ) ) , and hence th e posterior mean and variance are E^(0\x) = 7^ r^ , and n e x p ( - i ^ ) ( r 2 + ( ^ - ^ ) ^ ) - d ^ Vrnx) = - ^ y^ / iE(0\x)? . n e x p ( - ( ^ ) ( r ^ + (^-M)^)-rf ^ Note tha t th e above integrals cannot be computed in closed