Exponential tilting in Bayesian asymptotics

• 成果类型: Article
• 署名作者: Kharroubi, S. A.; Sweeting, T. J.
• 署名单位: American University of Beirut; University of London; University College London
• 刊物名称: BIOMETRIKA
• ISSN/ISSBN: 0006-3444
• DOI: 10.1093/biomet/asw018
• 发表日期: 2016
• 页码: 337349
• 关键词: marginal tail probabilities; posterior simulation; approximations; inference; FORMULA

摘要

We use exponential tilting to obtain versions of asymptotic formulae for Bayesian computation that do not involve conditional maxima of the likelihood function, yielding a more stable computational procedure and significantly reducing computational time. In particular we present an alternative version of the Laplace approximation for a marginal posterior density. Implementation of the asymptotic formulae and a modified signed root based importance sampler are illustrated with an example.