A NOTE ON THE ORTHOGONAL REPARAMETRIZATION IN BAYESIAN INFERENCE FOR THE EXTREME VALUE DISTRIBUTION

Josemar RODRIGUES[1]

Francisco LOUZADA NETO[2]

§    ABSTRACT: Consider the problem of inference about the scale parameter j in presence of location parameter f under the extreme value distribution. In a Bayesian framework, the marginal posterior and the profile posterior may be of interest. In this paper, using the Jeffreys prior density and the Laplace approximation, we show that these posteriors are invariant under the orthogonal reparametrization (COX & REID, 1987) adopted for the location parameter f . Also, from the data generated by the extreme value distribution, we show that the marginal and profile posteriors do not give conflicting information about the scale parameter j.

§    KEYWORDS: Marginal posterior; profile posterior; orthogonality; reparametrization.



[1]ICMSC-USP - 13560-970 - São Carlos - SP.

[2]DESs-UFSCar - 13565-905 - São Carlos - SP.