Abstract
Suppose that is a sample of size n with log likelihood , where θ is an unknown parameter in having a prior distribution . We need not assume that the sample values are independent or even stationary. Let be the maximum likelihood estimate (MLE). We show that is asymptotically normal with mean and covariance , where . In contrast, is asymptotically normal with mean θ and covariance , where is Fisher’s information. So, frequentist inference conditional on θ cannot be used to approximate Bayesian inference, except for exponential families. However, under mild conditions in probability. So, Bayesian inference (that is, conditional on ) can be used to approximate frequentist inference.
For any smooth function, we obtain posterior cumulant expansions, posterior Edgeworth–Cornish–Fisher (ECF) expansions and posterior tilted Edgeworth expansions for , as well as confidence regions for of high accuracy. We also give expansions for the Bayes estimate (estimator) of about , and for the maximum a posteriori estimate about , as well as their relative efficiencies with respect to squared error loss.
Citation
Christopher Withers. Saralees Nadarajah. "Expansions for posterior distributions." Braz. J. Probab. Stat. 37 (1) 73 - 100, March 2023. https://doi.org/10.1214/22-BJPS561
Information