We congratulate the authors for this interesting contribution to the wide world of objective priors (see Consonni et al., 2018, for a recent review). The authors tackle the problem of providing an objective prior which is model-free and based on the sole knowledge of the parameter space. We think that the main result can be a useful practical tool for objective Bayesian analysis in many applications and can open new ideas about objective priors. With our discussion, we hope to shed light on some aspects of the proposed approach, which is based on seeking a prior such that a combination of the log-score and of the Hyvarinen scoring rule is constant. In particular, we briefly comment on the following points: -extensions of the proposed approach using different scoring rules, and objectiveness and invariance of the proposed prior densities; -double use of the the Hyvarinen scoring rule, both for the derivation of the prior and to replace the likelihood function in models known up to the normalisation constant.
Invited discussion on: On a class of objective priors from scoring rules (with discussion)
Federica Giummolè;
2020-01-01
Abstract
We congratulate the authors for this interesting contribution to the wide world of objective priors (see Consonni et al., 2018, for a recent review). The authors tackle the problem of providing an objective prior which is model-free and based on the sole knowledge of the parameter space. We think that the main result can be a useful practical tool for objective Bayesian analysis in many applications and can open new ideas about objective priors. With our discussion, we hope to shed light on some aspects of the proposed approach, which is based on seeking a prior such that a combination of the log-score and of the Hyvarinen scoring rule is constant. In particular, we briefly comment on the following points: -extensions of the proposed approach using different scoring rules, and objectiveness and invariance of the proposed prior densities; -double use of the the Hyvarinen scoring rule, both for the derivation of the prior and to replace the likelihood function in models known up to the normalisation constant.File | Dimensione | Formato | |
---|---|---|---|
discussion.pdf
accesso aperto
Descrizione: Versione dell'editore
Tipologia:
Versione dell'editore
Licenza:
Creative commons
Dimensione
364.34 kB
Formato
Adobe PDF
|
364.34 kB | Adobe PDF | Visualizza/Apri |
I documenti in ARCA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.