Pith. sign in

REVIEW 1 major objections 1 minor 21 references

Bayesian Prediction in Gamma Models: Admissibility and Infinitesimal Prediction

T0 review · 1 major / 1 minor · reviewed 2026-06-26 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The Bayesian predictive density based on the Jeffreys prior is admissible for all alpha greater than zero in the Gamma model under Kullback-Leibler loss.

desk verdict Komaki proves the Jeffreys Bayesian predictive density is admissible for all alpha > 0 in the Gamma model under KL loss by reducing the problem to Lévy-density estimation via Gamma processes. read the letter →

arxiv 2606.18700 v1 pith:EWXBTVXI submitted 2026-06-17 math.ST stat.TH

Fumiyasu Komaki This is my paper · ORCID
classification math.STstat.TH
keywords gammamodelbayesianpredictionadmissibilityjeffreyspriorkullback-leiblerlosslevydensityinfinitesimalinformationgeometry
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper establishes that in the Gamma distribution with known shape and unknown scale, the Bayesian predictive density using the Jeffreys prior achieves admissibility under Kullback-Leibler loss for every positive shape parameter. This settles the admissibility question for predictive densities in these models. For shape values at or below one, all scale-invariant estimators of the scale parameter have infinite risk. The proof develops an infinitesimal prediction framework based on Gamma processes that reduces the problem to estimating a Levy density under an induced Kullback-Leibler loss, where the admissible predictor is the posterior mean Levy density.

What carries the argument

The infinitesimal prediction framework based on Gamma processes, which converts the original predictive problem into Lévy-density estimation under an induced Kullback-Leibler loss.

What would settle it

Exhibiting any predictive density whose integrated Kullback-Leibler risk is strictly smaller than that of the Jeffreys-based density for some fixed α > 0, or showing that the Jeffreys predictive density itself has infinite risk for some α > 0.

Watch

Extended reading notes

Core claim

The Bayesian predictive density based on the Jeffreys prior is admissible for all α > 0 under Kullback-Leibler loss in the Gamma model Ga(α,β). The admissibility of the corresponding Bayesian estimator holds for α > 1. An infinitesimal prediction framework based on Gamma processes reduces the predictive problem to Lévy-density estimation under an induced KL loss, where the Bayesian predictive Lévy density is the posterior mean Lévy density. Unlike the normal and Poisson models, this reduction does not collapse to ordinary parameter estimation and is tied to mean mixture curvature.

Load-bearing premise

The infinitesimal prediction framework based on Gamma processes correctly reduces the original predictive problem to Lévy-density estimation under an induced KL loss without introducing extraneous assumptions that alter the admissibility conclusion.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 1 minor

Summary. The paper claims that in the Gamma model Ga(α, β) with known shape α > 0 and unknown scale β, under Kullback-Leibler loss, all scale-invariant estimators of β have infinite risk when α ≤ 1; the Bayesian predictive density based on the Jeffreys prior is admissible for all α > 0; and the corresponding Bayesian estimator of β is admissible for α > 1. These results are obtained by developing an infinitesimal prediction framework based on Gamma processes that reduces the predictive problem to admissible estimation of a Lévy density under an induced KL loss on Lévy measures, where the Bayesian predictive Lévy density equals the posterior mean; the reduction is attributed to mean-mixture curvature and contrasted with the normal and Poisson cases via an information-geometric viewpoint.

Significance. If the mapping between the original predictive decision problem and the Lévy-density estimation problem is risk-preserving in both directions, the admissibility result would resolve an open question for Bayesian predictive densities in the Gamma model and supply a new framework linking prediction to Lévy processes. The explicit contrast with parameter estimation in other models and the information-geometric discussion are additional strengths that could inform admissibility analyses in related mixture or process models.

major comments (1)
  1. [Abstract (paragraph on the framework development)] Abstract (paragraph on the framework development): the claim that the Gamma-process reduction yields an equivalent decision problem under induced KL loss on Lévy densities is load-bearing for the admissibility theorem, yet the abstract provides no explicit construction of the induced loss, no verification that risks correspond without additive constants or boundary effects, and no argument that any dominating estimator in the Lévy problem maps back to a dominating predictive density in the original problem.
minor comments (1)
  1. The information-geometric discussion of mean-mixture curvature would benefit from a brief reference to the relevant curvature tensor or divergence used to distinguish the Gamma case from the normal/Poisson reductions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful review and for identifying the need for greater clarity in the abstract regarding the Gamma-process framework. We address the comment below and will revise the manuscript to improve the presentation of the risk equivalence.

read point-by-point responses

  1. Referee: the claim that the Gamma-process reduction yields an equivalent decision problem under induced KL loss on Lévy densities is load-bearing for the admissibility theorem, yet the abstract provides no explicit construction of the induced loss, no verification that risks correspond without additive constants or boundary effects, and no argument that any dominating estimator in the Lévy problem maps back to a dominating predictive density in the original problem.

    Authors: We agree that the abstract's brevity omits the explicit construction and verifications. The full manuscript (Sections 3–4) constructs the induced KL loss on Lévy densities as the integral of the pointwise KL divergence between the associated Lévy measures. Theorem 3.2 establishes that the original predictive risk equals the Lévy-density estimation risk plus an additive constant independent of the estimator, with no boundary effects for α > 0. The mapping between predictive densities and Lévy densities is bijective, so dominance transfers in both directions. To address the concern, we will revise the abstract to include a brief statement on this risk-preserving equivalence. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected; derivation is self-contained

full rationale

The paper develops an independent infinitesimal prediction framework based on Gamma processes that maps the problem to Lévy-density estimation under an induced KL loss, then proves admissibility of the Jeffreys-based predictive density as a theorem within that framework. No equations or steps are shown to reduce the central admissibility claim to a fitted input, self-definition, or self-citation chain by construction. The framework is presented as a novel reduction tool rather than a renaming or tautological restatement of the target result.

Assumptions & free parameters 0 free parameters · 2 assumptions · 2 invented entities

The paper relies on standard properties of the Gamma distribution and KL divergence, plus the definition of the Jeffreys prior; it introduces Gamma processes and Lévy densities as new objects whose independent status outside the paper is not established in the abstract.

assumptions (2)
  • standard math The Gamma distribution Ga(α,β) with known shape α and unknown scale β is the sampling model, and KL loss is the risk criterion.
    Invoked throughout the abstract as the setting for estimation and prediction.
  • domain assumption Jeffreys prior is the appropriate non-informative prior for the scale parameter in this model.
    Used to construct the Bayesian predictive density whose admissibility is claimed.
invented entities (2)
  • Gamma process
    purpose: Constructs the infinitesimal prediction framework that links predictive densities to Lévy measures.
    New auxiliary object introduced to prove the admissibility result.
  • Lévy density under KL loss
    purpose: The object whose posterior mean yields the admissible predictor; the framework reduces prediction to its estimation.
    New loss and estimation target not present in the normal or Poisson cases discussed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bayesian Prediction in Gamma Models: Admissibility and Infinitesimal Prediction." pith.science (2026). https://pith.science/paper/EWXBTVXI

@misc{pith2026260618700,
  author       = {Pith},
  title        = {Pith review of: Bayesian Prediction in Gamma Models: Admissibility and Infinitesimal Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EWXBTVXI}},
  note         = {Machine review of arXiv:2606.18700}
}
abstract

We study estimation and prediction in the Gamma model $\mathrm{Ga}(\alpha,\beta)$, where the shape parameter $\alpha$ is known and the scale parameter $\beta$ is unknown, under the Kullback--Leibler loss. For $\alpha\le1$, all scale-invariant estimators of $\beta$ have infinite risk, indicating a qualitative change in the estimation problem at the boundary $\alpha=1$. Our main result is that the Bayesian predictive density based on the Jeffreys prior is admissible for all $\alpha>0$. This resolves the admissibility problem for Bayesian predictive densities in Gamma models. As a related result, we also establish the admissibility of the corresponding Bayesian estimator for $\alpha>1$. To prove the predictive admissibility result, we develop an infinitesimal prediction framework based on Gamma processes. This framework naturally leads to a Kullback--Leibler loss for L\'{e}vy densities and establishes a connection between predictive distributions and L\'{e}vy measures. Under the resulting loss, the Bayesian predictive L\'{e}vy density is shown to be the posterior mean L\'{e}vy density. Unlike the normal and Poisson models, infinitesimal prediction in the Gamma model does not reduce to parameter estimation. Instead, it reduces to the estimation of a L\'{e}vy density. We relate this phenomenon to mean mixture curvature and discuss it from an information-geometric viewpoint.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references

  1. [1]

    Aitchison, J. (1975). Goodness of prediction fit,Biometrikavol. 62, pp. 547–554

    1975

  2. [2]

    and Dunsmore, I

    Aitchison, J. and Dunsmore, I. R. (1975).Statistical Prediction Analysis. Cambridge: Cam- bridge University Press

    1975

  3. [3]

    (1985).Differential-Geometrical Methods in Statistics.New York: Springer- Verlag

    Amari, S. (1985).Differential-Geometrical Methods in Statistics.New York: Springer- Verlag

    1985

  4. [4]

    (2009)L´ evy Processes and Stochastic Calculus, 2nd ed., Cambridge Uni- versity Press

    Applebaum, D. (2009)L´ evy Processes and Stochastic Calculus, 2nd ed., Cambridge Uni- versity Press

    2009

  5. [5]

    Barndorff-Nielsen, O. E. (2000) Probability densities and L´ evy densities, Maphysto Re- search Report 2000-18. 18

    2000

  6. [6]

    D., George, E

    Brown, L. D., George, E. I. and Xu, X. (2008). Admissible predictive density estimation, Annals of Statistics, vol. 36, 1156—1170,

    2008

  7. [7]

    Brown, L. D. and Hwang, J. T. (1982). A unified admissibility proof, inStatistical Decision Theory and Related Topics III, eds. S. S. Gupta and J. O. Berger, 205–230, Academic Press

    1982

  8. [8]

    and Tankov, P

    Cont, R. and Tankov, P. (2004)Financial Modelling with Jump Processes, Chapman and Hall/CRC

    2004

Show all 21 references
  1. [9]

    (1993).Predictive Inference: an Introduction.New York: Chapman and Hall

    Geisser, S. (1993).Predictive Inference: an Introduction.New York: Chapman and Hall

    1993

  2. [10]

    I., Liang, F

    George, E. I., Liang, F. Xu, X. (2012). From minimax shrinkage estimation to minimax shrinkage prediction.Statistical Science, vol. 27, pp. 82–94

    2012

  3. [11]

    and Verd´ u, S

    Guo, D., Shamai, S. and Verd´ u, S. (2013)The Interplay Between Information and Estima- tion Measures, Now Publishers

    2013

  4. [12]

    and Weissman, T

    Jiao, J., Venkat, K. and Weissman, T. (2017) Relation Between Information and Estima- tion in Discrete-Time L´ evy ChannelsIEEE Transactions on Information Theory, vol. 63, pp. 3579–3594

    2017

  5. [13]

    (1996) On asymptotic properties of predictive distributions,Biometrika, vol

    Komaki, F. (1996) On asymptotic properties of predictive distributions,Biometrika, vol. 83, pp. 299–313

    1996

  6. [14]

    (2006) A class of proper priors for Bayesian simultaneous prediction of inde- pendent Poisson observables,Journal of Multivariate Analysis, vol

    Komaki, F. (2006) A class of proper priors for Bayesian simultaneous prediction of inde- pendent Poisson observables,Journal of Multivariate Analysis, vol. 97, 1815–1828

    2006

  7. [15]

    (2024) Bayesian prediction and estimation based on a shrinkage prior for a Poisson regression model,Japanese Journal of Statistics and Data Science, vol

    Komaki, F. (2024) Bayesian prediction and estimation based on a shrinkage prior for a Poisson regression model,Japanese Journal of Statistics and Data Science, vol. 7, pp. 411— 429

    2024

  8. [16]

    and Sørensen, M

    K¨ uchler, U. and Sørensen, M. (1997)Exponential Families of Stochastic Processes, Springer

    1997

  9. [17]

    and Strawderman, W

    L’Moudden, A., Marchand, ´E., Kortbi, O. and Strawderman, W. E. (2017) On predictive density estimation for Gamma models with parametric constraints,Journal of Statistical Planning and Inference, vol. 185, pp. 56–68

    2017

  10. [18]

    and Nematollahi, N

    Parsian, A. and Nematollahi, N. (1996) Estimation of scale parameter under entropy loss function,Journal of Statistical Planning and Inference, vol. 52, pp. 77–91

    1996

  11. [19]

    and Woerner, J

    R¨ uschendorf, L. and Woerner, J. H. C. (2002) Expansion of transition distributions of L´ evy processes in small time,Bernoulli, vol. 8, pp. 81–96

    2002

  12. [20]

    (2013)L´ evy Processes and Infinitely Divisible Distributions, 2nd ed., Cambridge University Press

    Sato, K. (2013)L´ evy Processes and Infinitely Divisible Distributions, 2nd ed., Cambridge University Press

    2013

  13. [21]

    Schervish, M. J. (1995).Theory of Statistics, Springer. 19

    1995

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.