On Parametric Bootstrapping and Bayesian Prediction (Q4677103)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 2169663
Language Label Description Also known as
default for all languages
No label defined
    English
    On Parametric Bootstrapping and Bayesian Prediction
    scientific article; zbMATH DE number 2169663

      Statements

      On Parametric Bootstrapping and Bayesian Prediction (English)
      0 references
      0 references
      0 references
      0 references
      20 May 2005
      0 references
      Kullback-Leibler loss function
      0 references
      asymptotic expansion
      0 references
      Poisson regression
      0 references
      The i.i.d. observations \(x^N=(x_1,\dots,x_N)\) have a PDF \(p(x\,|\,\omega)\) w.r.t. a measure \(\mu\), where \(\omega\) is an unknown parameter. The problem is to predict the distribution \(q(y\,|\,\omega)\) of some future observation \(y\) (\(q\) in this setting is not necessary equal to \(p\)). A Kullback-Leibler divergence is used as a loss function. The Bayesian \(q_\pi\) and bootstrap predictive \(q^*\) distributions are considered. E.g., \[ q^*(y\mid \hat\omega(x^N))=\int q(y\mid \hat\omega(Z^N))p(Z^N\mid \hat\omega(x^N))\mu(d Z^N), \] where \(\hat\omega(x^N)\) is the MLE estimate for \(\omega\) by \(x^N\). Asymptotic expansions for \(\hat q_\pi\) and \(q^*\) are derived. It is demonstrated that they are asymptotically equivalent and asymptotically dominate estimative distributions (i.e., the distributions of the form \(q(y\mid \tilde\omega(x^N)\), where \(\tilde\omega\) is an estimate for \(\omega\)). Normal \(N(\omega,\omega^2)\), binomial, negative binomial distributions and Poisson regression are considered as examples.
      0 references

      Identifiers