arxiv: 2605.10495 · v1 · submitted 2026-05-11 · 📊 stat.ME · econ.TH

Recognition: 2 theorem links

· Lean Theorem

Robust Bayes Acts under Prior Perturbations: Contamination, Stability, and Selection Paths

(2) Ludwig-Maximilians-Universit\"at M\"unchen, Christoph Jansen (1), Georg Schollmeyer (2) ((1) Lancaster University Leipzig, Germany, Germany)

Pith reviewed 2026-05-12 04:58 UTC · model grok-4.3

classification 📊 stat.ME econ.TH

keywords robust Bayesprior perturbationsstability measuresrobustness radiuscontamination needdecision theoryportfolio selectionmodel uncertainty

0 comments

The pith

Two complementary stability measures quantify how robust Bayes-optimal acts are to prior perturbations in finite decision problems.

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

The paper introduces the robustness radius, which is the largest perturbation to a reference prior that keeps a given act Bayes-optimal, and the contamination need, the smallest perturbation needed to make an act Bayes-optimal under a nearby prior. These measures are formulated as linear programs and can be computed efficiently using bisection methods that exploit monotonicity. The authors then develop a cost-adjusted stability criterion that creates a family of decision rules parameterized by a regularization parameter, leading to selection paths that show transitions between stability-focused and cost-focused choices. This framework is illustrated in a portfolio selection problem using historical ETF returns, where it helps account for uncertainty in economic regimes and refines standard expected utility approaches.

Core claim

In finite decision problems, the robustness radius and contamination need provide quantitative assessments of an act's stability under prior perturbations. These are solved via linear programming and bisection. A cost-adjusted version yields parametric selection rules whose paths reveal regime shifts. Application to portfolio strategies under regime uncertainty demonstrates how robustness considerations modify classical Bayes decisions.

What carries the argument

The robustness radius and contamination need, defined as optimization problems over prior perturbations and characterized through linear programming formulations that exploit monotonicity for efficient computation via bisection.

If this is right

The cost-adjusted stability criterion generates a parametric family of decision rules indexed by a regularization parameter.
Selection paths reveal structural transitions between stability-driven and cost-driven regimes as the parameter varies.
Robustness-based selection refines classical expected utility by incorporating considerations of prior misspecification.
In the portfolio choice example, different strategies exhibit distinct robustness and contamination profiles under heterogeneous belief specifications.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

These stability notions could be tested in other finite decision settings, such as medical treatment choices under uncertain priors.
The bisection computation method suggests the approach scales well to moderately sized finite problems.
Selection paths might be used to analyze sensitivity in dynamic decision environments where priors evolve over time.

Load-bearing premise

Prior perturbations can be quantified such that the resulting optimization problems remain linear programs with the required monotonicity properties.

What would settle it

A counterexample in a small finite decision problem where the linear programming solution for the robustness radius does not correspond to the actual maximum perturbation preserving Bayes-optimality of the act.

Figures

Figures reproduced from arXiv: 2605.10495 by (2) Ludwig-Maximilians-Universit\"at M\"unchen, Christoph Jansen (1), Georg Schollmeyer (2) ((1) Lancaster University Leipzig, Germany, Germany).

**Figure 1.** Figure 1: Contamination need and the robustness radius of the different portfolios 𝑎1, . . . , 𝑎6 under all prior specifications. Interpretation of the results. The results in [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗

**Figure 2.** Figure 2: Selection paths of the cost-adjusted stability criterion for the portfolios 𝑎1, . . . , 𝑎6 along the different choices of 0 ≤ 𝜆 ≤ 3. istics are preferred, whereas larger values of 𝜆 increasingly favor portfolios with lower variance-based costs. Overall, the results demonstrate that robustnessbased analysis refines classical expected utility comparisons by incorporating the stability of optimality under ch… view at source ↗

read the original abstract

This paper develops a quantitative framework to assess the robustness of Bayes-optimal decisions in finite decision problems under model uncertainty. We introduce two complementary stability notions for acts: the robustness radius, measuring the largest perturbation of a reference prior under which an act remains Bayes-optimal, and the contamination need, quantifying the minimal perturbation required for an act to become Bayes-optimal under some nearby prior. Both concepts are characterized via linear programming formulations and computed efficiently using bisection methods exploiting monotonicity properties. Building on these stability measures, we propose a cost-adjusted stability criterion that integrates robustness considerations with act-specific selection costs, yielding a parametric family of decision rules indexed by a regularization parameter. We analyze how optimal act selection evolves along this parameter and derive selection paths that reveal structural transitions between stability-driven and cost-driven regimes. The framework is applied to a portfolio choice problem under uncertainty between different economic regimes. Concretely, using data on historical ETF returns, we compute robustness and contamination profiles for six portfolio strategies and analyze their behavior under heterogeneous belief specifications. The results illustrate that robustness-based selection refines classical expected utility by accounting for prior misspecification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives LP-based measures for how far a prior can be perturbed before a Bayes act loses optimality, plus cost-adjusted selection paths shown on ETF data.

read the letter

The main takeaway is that this work defines two complementary stability measures for Bayes acts in finite decision problems. The robustness radius is the largest prior perturbation under which the current act stays optimal, and the contamination need is the smallest perturbation that makes some other act optimal. Both are turned into linear programs and solved by bisection on the monotonic value functions. The paper then folds these into a cost-adjusted criterion that produces a parametric family of rules indexed by a regularization parameter, and it traces the resulting selection paths to show transitions between stability-driven and cost-driven regimes. The ETF portfolio example computes profiles for six strategies under regime uncertainty and illustrates how the choices shift with different belief specifications and the regularization weight. This is a clean extension of standard robust decision theory tools. The technical steps rest on LP duality and monotonicity, both of which are standard and hold without gaps in the finite setting. The application uses historical ETF returns to discretize regimes and gives concrete numbers, which helps make the abstract concepts usable. The soft spots are limited and proportionate. Everything stays inside finite actions and states, so the framework does not claim generality to continuous problems. The regularization parameter is treated as a free knob for exploring the paths, but the paper offers little guidance on defaults or sensitivity to its value. The ETF results are illustrative rather than a full out-of-sample test, and the effect of regime discretization is not probed in depth. Readers working in robust Bayesian decision theory or portfolio choice under model uncertainty will find the methods directly applicable and the examples helpful. The paper shows clear thinking on its own terms and deserves a serious referee because the derivations are grounded and the computational approach is practical. I would send it for peer review.

Referee Report

0 major / 3 minor

Summary. The paper develops a quantitative framework to assess the robustness of Bayes-optimal decisions in finite decision problems under model uncertainty. It introduces the robustness radius and contamination need, characterized via linear programming and computed using bisection methods. It proposes a cost-adjusted stability criterion with a regularization parameter leading to selection paths, and applies it to portfolio choice with ETF data.

Significance. The framework extends robust Bayes methods with efficient computational tools and parametric selection rules. The LP formulations and monotonicity exploitation are standard but well-applied here, providing practical measures for decision stability. The portfolio application illustrates how robustness refines expected utility under prior misspecification. This could be significant for decision theory in statistics and economics if the derivations are tight.

minor comments (3)

[Introduction] The motivation for the two complementary measures could be expanded with a simple example early on.
[§4] Clarify how the bisection method exploits the monotonicity property with a proof sketch or reference.
[Application] Provide more details on the discretization of the ETF returns data and the choice of the six strategies.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our work, the recognition of its potential significance for decision theory, and the recommendation of minor revision. The referee's description of the robustness radius, contamination need, LP characterizations, bisection methods, cost-adjusted selection paths, and the ETF portfolio application aligns closely with the manuscript.

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper introduces robustness radius and contamination need as new stability notions explicitly characterized by linear programming formulations that are constructed independently of the target quantities. These are solved via bisection on monotonic value functions, which relies on standard LP duality and monotonicity properties external to the paper's definitions. The cost-adjusted stability criterion, parametric decision rules, and selection paths are then derived directly from these LP-based measures without any reduction to fitted parameters, self-definitional loops, or load-bearing self-citations. The portfolio application simply instantiates the finite-act framework on discretized data. All load-bearing steps rest on verifiable external technical ingredients rather than internal redefinitions or renamings, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on standard mathematical assumptions for linear programming and monotonicity in finite decision problems; the regularization parameter is a free parameter controlling the cost-stability tradeoff.

free parameters (1)

regularization parameter
Parametric family of decision rules indexed by this parameter that trades off robustness and selection costs.

axioms (2)

domain assumption Finite decision problems allow characterization via linear programming
Invoked to enable efficient computation of stability measures.
domain assumption Monotonicity properties hold for the stability measures under prior perturbations
Used to justify bisection methods for computation.

pith-pipeline@v0.9.0 · 5525 in / 1336 out tokens · 51964 ms · 2026-05-12T04:58:53.545637+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
rob(a,π0) and con(a,π0) characterized via linear programs with perturbation constraints and bisection exploiting monotonicity of R(ε)
IndisputableMonolith/Foundation/BranchSelection.lean branch_selection unclear
cost-adjusted stability score Sλ,π0(a) trading robustness radius against selection costs

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

[1]

Augustin, F

T. Augustin, F. Coolen, G. De Cooman, and M. Troffaes, eds.Introduction to Imprecise Probabilities. Wiley, 2014

work page 2014
[2]

Imprecise probabilities in engi- neering analyses

M. Beer, S. Ferson, and V. Kreinovich. “Imprecise probabilities in engi- neering analyses”. In:Mechanical Systems and Signal Processing37 (2013), pp. 4–29

work page 2013
[3]

Comparing ma- chine learning algorithms by union-free generic depth

H. Blocher, G. Schollmeyer, M. Nalenz, and C. Jansen. “Comparing ma- chine learning algorithms by union-free generic depth”. In:International Journal of Approximate Reasoning169 (2024), p. 109166

work page 2024
[4]

Boyd and L

S. Boyd and L. Vandenberghe.Convex Optimization. Cambridge Univer- sity Press, 2004

work page 2004
[5]

Semi-supervised Learning Guided by the Generalized Bayes Rule Under Soft Revision

S. Dietrich, J. Rodemann, and C. Jansen. “Semi-supervised Learning Guided by the Generalized Bayes Rule Under Soft Revision”. In:Combining, Mod- elling and Analyzing Imprecision, Randomness and Dependence. Springer, 2024, pp. 110–117

work page 2024
[6]

The notion of breakdown point

D. Donoho and P. Huber. “The notion of breakdown point”. In:A festschrift for Erich L. Lehmann157184 (1983), p. 1004

work page 1983
[7]

Kenneth R. French Data Library

K. French. “Kenneth R. French Data Library”. In:Dartmouth College (2024).url: https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ data library.html

work page 2024
[8]

Gordienko, C

P. Gordienko, C. Jansen, J. Rodemann, and G. Schollmeyer.Beyond Ar- row: From Impossibility to Possibilities in Multi-Criteria Benchmarking

work page
[9]

arXiv: 2602.07593[cs.LG].url: https://arxiv.org/abs/2602.07593

work page arXiv
[10]

A General Qualitative Definition of Robustness

F. Hampel. “A General Qualitative Definition of Robustness”. In:The Annals of Mathematical Statistics42.6 (1971), pp. 1887–1896.doi: 10. 1214/aoms/1177693054.url: https://doi.org/10.1214/aoms/1177693054

work page doi:10.1214/aoms/1177693054 1971
[11]

Efficiency in normal samples and tolerance of extreme val- ues for some estimates of location

J. Hodges Jr. “Efficiency in normal samples and tolerance of extreme val- ues for some estimates of location”. In:Proceedings of the fifth Berkeley symposium on mathematical statistics and probability. Vol. 1. Univ of Cal- ifornia Press. 1967, pp. 163–186. 12 Christoph Jansen & Georg Schollmeyer

work page 1967
[12]

The use of previous experience in reaching statistical decisions

J. Hodges Jr and E. Lehmann. “The use of previous experience in reaching statistical decisions”. In:The Annals of Mathematical Statistics(1952), pp. 396–407

work page 1952
[13]

Huber.Robust Statistics

P. Huber.Robust Statistics. Wiley & Sons, 1980

work page 1980
[14]

Contributions to the decision theoretic foundations of machine learning and robust statistics under weakly structured information

C. Jansen.Contributions to the Decision Theoretic Foundations of Ma- chine Learning and Robust Statistics under Weakly Structured Informa- tion. Habilitationsschrift, Ludwig-Maximilians-Universit¨ at M¨ unchen: https: //arxiv.org/abs/2501.10195. 2025

work page arXiv 2025
[15]

Some Contributions to Decision Making in Complex Informa- tion Settings with Imprecise Probabilities and Incomplete Preferences

C. Jansen. “Some Contributions to Decision Making in Complex Informa- tion Settings with Imprecise Probabilities and Incomplete Preferences”. PhD thesis. Ludwig-Maximilians-Universit¨ at M¨ unchen: https://edoc.ub. uni-muenchen.de/22653/1/Jansen Christoph.pdf, 2018

work page 2018
[16]

Decision Making with State-Dependent Pref- erence Systems

C. Jansen and T. Augustin. “Decision Making with State-Dependent Pref- erence Systems”. In:Information Processing and Management of Uncer- tainty in Knowledge-Based Systems. Springer, 2022, pp. 729–742

work page 2022
[17]

Decision Theory Meets Lin- ear Optimization Beyond Computation

C. Jansen, T. Augustin, and G. Schollmeyer. “Decision Theory Meets Lin- ear Optimization Beyond Computation”. In:Symbolic and Quantitative Approaches to Reasoning with Uncertainty. Springer, 2017, pp. 329–339

work page 2017
[18]

Information ef- ficient learning of complexly structured preferences: Elicitation procedures and their application to decision making under uncertainty

C. Jansen, H. Blocher, T. Augustin, and G. Schollmeyer. “Information ef- ficient learning of complexly structured preferences: Elicitation procedures and their application to decision making under uncertainty”. In:Interna- tional Journal of Approximate Reasoning144 (2022), pp. 69–91

work page 2022
[19]

Statistical Com- parisons of Classifiers by Generalized Stochastic Dominance

C. Jansen, M. Nalenz, G. Schollmeyer, and T. Augustin. “Statistical Com- parisons of Classifiers by Generalized Stochastic Dominance”. In:Journal of Machine Learning Research24.231 (2023), pp. 1–37

work page 2023
[20]

A probabilistic evaluation framework for preference aggregation reflecting group homogeneity

C. Jansen, G. Schollmeyer, and T. Augustin. “A probabilistic evaluation framework for preference aggregation reflecting group homogeneity”. In: Mathematical Social Sciences96 (2018), pp. 49–62

work page 2018
[21]

Concepts for decision mak- ing under severe uncertainty with partial ordinal and partial cardinal pref- erences

C. Jansen, G. Schollmeyer, and T. Augustin. “Concepts for decision mak- ing under severe uncertainty with partial ordinal and partial cardinal pref- erences”. In:International Journal of Approximate Reasoning98 (2018), pp. 112–131

work page 2018
[22]

Multi-target decision mak- ing under conditions of severe uncertainty

C. Jansen, G. Schollmeyer, and T. Augustin. “Multi-target decision mak- ing under conditions of severe uncertainty”. In:Modeling Decisions for Artificial Intelligence (MDAI). Springer. 2023

work page 2023
[23]

Quantifying degrees of E- admissibility in decision making with imprecise probabilities

C. Jansen, G. Schollmeyer, and T. Augustin. “Quantifying degrees of E- admissibility in decision making with imprecise probabilities”. In:Reflec- tions on the Foundations of Statistics: Essays in Honor of Teddy Seiden- feld. Springer, 2022, pp. 319–346

work page 2022
[24]

Robust statistical comparison of random variables with locally varying scale of measurement

C. Jansen, G. Schollmeyer, H. Blocher, J. Rodemann, and T. Augustin. “Robust statistical comparison of random variables with locally varying scale of measurement”. In:Proceedings of the Thirty-Ninth Conference on Uncertainty in Artificial Intelligence. Vol. 216. Proceedings of Machine Learning Research. PMLR, 2023, pp. 941–952. Robustness of Bayes-Acts 13

work page 2023
[25]

Statistical Multicriteria Benchmarking via the GSD-Front

C. Jansen, G. Schollmeyer, J. Rodemann, H. Blocher, and T. Augustin. “Statistical Multicriteria Benchmarking via the GSD-Front”. In:Advances in Neural Information Processing Systems 37 (NeurIPS 2024). 2024

work page 2024
[26]

Kofler and G

E. Kofler and G. Menges.Entscheidungen bei unvollst¨ andiger Information. Springer, 1976

work page 1976
[27]

Impre- cise probability assessment of tipping points in the climate system

E. Kriegler, J. Hall, H. Held, R. Dawson, and H. Schellnhuber. “Impre- cise probability assessment of tipping points in the climate system”. In: Proceedings of the National Academy of Sciences106 (2009), pp. 5041– 5046

work page 2009
[28]

On indeterminate probabilities

I. Levi. “On indeterminate probabilities”. In:Journal of Philosophy71 (1974), pp. 391–418

work page 1974
[29]

In all likeli- hoods: robust selection of pseudo-labeled data

J. Rodemann, C. Jansen, G. Schollmeyer, and T. Augustin. “In all likeli- hoods: robust selection of pseudo-labeled data”. In:Proceedings of the 13th International Symposium on Imprecise Probability: Theories and Applica- tions. Vol. 215. PMLR, 2023, pp. 412–425

work page 2023
[30]

Ryan and J

J. Ryan and J. Ulrich.quantmod: Quantitative Financial Modelling Frame- work. R package version available on CRAN. 2025.url: https://CRAN.R- project.org/package=quantmod

work page 2025
[31]

Savage.The Foundations of Statistics

L. Savage.The Foundations of Statistics. Wiley, 1954

work page 1954
[32]

Decision making under uncertainty using imprecise proba- bilities

M. Troffaes. “Decision making under uncertainty using imprecise proba- bilities”. In:International Journal of Approximate Reasoning45 (2007), pp. 17–29

work page 2007
[33]

Powerful algorithms for decision making under partial prior information and general ambiguity attitudes

L. Utkin and T. Augustin. “Powerful algorithms for decision making under partial prior information and general ambiguity attitudes”. In:Proceedings of the Fourth International Symposium on Imprecise Probability: Theories and Applications. SIPTA, 2005, pp. 349–358

work page 2005
[34]

Von Neumann and O

J. Von Neumann and O. Morgenstern.Theory of Games and Economic Behavior. Princeton University Press, 1944

work page 1944
[35]

Walley.Statistical Reasoning with Imprecise Probabilities

P. Walley.Statistical Reasoning with Imprecise Probabilities. Chapman & Hall, 1991. [35]Yahoo Finance Historical Price Data. Accessed via thequantmodR pack- age. ETF price data for SPY, EFA, EEM, TLT, GLD, VNQ. 2024

work page 1991