Asymptotic expansions of posterior distributions in a non-regular model (Q690157): Difference between revisions

Let \(f(x)\) be a uniformly continuous density function, defined on an interval \((0,\beta)\) such that \(f'(x) \to \alpha(x \to +0)\), and consider a family \(\{f(x-\theta);\;\theta \in R\}\). Let \(x_ 1,\dots,x_ n\) be i.i.d. random variables with density function \(f(x-\theta_ 0)\). Further, let \(\hat\theta_ n\) be the maximum likelihood estimator of \(\theta_ 0\). \textit{M. Woodroofe} [Ann. Math. Statistics 43, 113-122 (1972; Zbl 0251.62018)] showed the asymptotic normality of \(\alpha_ n(\hat\theta_ n-\theta_ 0)\) \((\alpha^ 2_ n=(\alpha/2)n \log n)\) when \(f\) is ``nonregular''. On the other hand, \textit{R. A. Johnson} [Ann. Math. Statistics. 41, 851-864 (1970; Zbl 0204.530)] obtained the almost sure asymptotic expansions of the centered and scaled posterior distributions when \(f\) is ``regular''. In this paper the author shows, using the Woodroofe method, that the analogous asymptotic expansions to the Johnson's results (but with probability greater than \(1-\varepsilon)\) hold in the nonregular case.