Convex Order and Moment Persuasion

arXiv:2609.13696 · econ.TH · Submitted 2026-09-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Convex Order and Moment Persuasion".

Paper summary: Ruth: So, looking at this paper "Convex Order and Moment Persuasion," what’s the big picture takeaway for us outside of just the math? The authors are really giving us a way to visualize and characterize these mean-preserving contractions.

Warren: I think the core contribution is that they provide this complete geometric characterization, moving beyond just knowing that one measure is in convex order of another. They’ve shown how to decompose the state space into these specific convex cells where things concentrate.

Iris: And the real utility for us is seeing how those extreme points behave under this structure. It tells us exactly what kind of information structure would be optimal in a moment persuasion problem, and it links that optimality back to the geometry of the prior itself.

Ruth: It connects Bayesian theory, which is where we start with posterior beliefs being contractions of the prior, to concrete optimization problems by providing these geometric tools. The paper shows that understanding this geometry is essential for characterizing optimal information structures in multidimensional persuasion problems.

Warren: So, if you distill it down, they’re showing that the structure of the prior dictates a concentration pattern on the state space, and extremality happens when your support lines up perfectly with that partition.

Iris: And they flag a limitation: for this sharp characterization using Assumption one to hold cleanly, we need those boundaries between cells to be negligible under the prior <ref:2609.13696#pg2>. If that assumption fails, the clean separation breaks down because boundary points can introduce extra degrees of freedom that complicate finding global extremality.

Ruth: So the paper gives us a precise geometric map for these contractions and extreme points in multidimensional persuasion, but we have to be careful about those boundary conditions when applying it to real-world problems.

Warren: It’s a useful framework for identifying the right types of signals you can send, based on how they fit into these concentration regions.

Iris: Exactly. It moves the discussion from just knowing that something is in convex order to understanding *why* and *how* it happens geometrically across different dimensions.

Conclusion: Ruth: So, we're wrapping up this look at "Convex Order and Moment Persuasion" by Kleiner et al.

Warren: It basically boils down to using geometry—this idea of convex order—to map out exactly how different probability measures relate to each other in a multidimensional setting.

Iris: They show that if one measure is dominated by another in this convex order, the state space breaks down into specific geometric regions where things have to concentrate.

Ruth: That concentration structure is what lets them characterize the extreme points of these contractions, which are basically the most efficient or important measures we can have in this system.

Warren: And those extreme points directly correspond to what an optimal information structure would look like when you're trying to persuade someone about some moment.

Iris: The real implication here is that understanding the shape of the prior itself dictates where you can actually get the best results from your signal or your data.

Ruth: It moves us from just looking at abstract probability to figuring out how the underlying structure of our beliefs limits what we can actually achieve in practice.

Warren: And they spend a lot of time dealing with these boundary conditions, figuring out when that geometric map breaks down because the rules get messy at the edges.

Iris: That boundary stuff is important because it tells us exactly when their neat characterization fails, which is crucial for knowing what assumptions we need to make in real-world modeling.

Ruth: We’ll be diving deeper into those specific conditions and how they apply to persuasion problems right after this break.

Western University

econ.TH

Submitted: 2026-09-12

Updated: 2026-11-17

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 82/100

The gist: If a probability measure is a mean-preserving contraction of another, the two measures are said to be in convex order <ref:2609.13696#pg2>.

Key concepts

Mean-Preserving Contraction
This describes a relationship between two probability measures where one is obtained from the other through a process that preserves the mean. In simple terms, it means one measure can be transformed into another while keeping their average values consistent, often related to how information or beliefs are updated.
Concentration Structure
This central feature states that every state in the system can only be mapped to posterior means located within its own specific irreducible component. This partitions the state space geometrically, ensuring that mass remains localized within these defined regions during the transformation process.
Extreme Points
These are the most fundamental or 'boundary' solutions within a set of possible outcomes (the mean-preserving contraction set). The paper provides a complete characterization showing that for a measure to be an extreme point, its support must satisfy specific affine independence conditions within the geometric cells.

Terminology

Summary

If a probability measure is a mean-preserving contraction of another, the two measures are said to be in convex order <ref:2609. p. 013696#pg2>. This paper provides a complete characterization of mean-preserving contractions and their extreme points in the multidimensional setting, extending the results of Kleiner et al. p. 0 (2024) to incorporate fully revealing regions and lower-dimensional pooling regions <ref:2609. p. 013696#pg2>. A central feature of our characterization is a “concentration” structure: each state can be mapped only to posterior means lying within its own irreducible component <ref:2609. p. 013696#pg2>. We apply these results to the moment persuasion problem <ref:2609. p. 013696#pg2>.

Geometric Characterization of Contractions

Theorem 1 provides a complete geometric characterization of mean-preserving contractions in multidimensional settings <ref:2609.13696#pg4>. We show that whenever a measure μ is dominated by μ0 in convex order, the state space admits a canonical partition into convex, relatively open cells such that the restriction of μ to each cell remains dominated by the corresponding restriction of μ0 <ref:2609. p. 213696#pg4>. Within each cell, any feasible martingale transport must map states into the closure of that same cell, a concentration property that parallels the irreducible-component structure identified in the martingale transport literature <ref:2609. p. 213696#pg5>. This decomposition generalizes the one-dimensional partition into intervals and singleton points and provides the geometric backbone for our analysis of extreme points <ref:2609. p. 213696#pg5>.

Characterization of Extreme Points

Theorem 2 provides a complete characterization of the extreme points of the meanpreserving contraction set <ref:2609.13696#pg5>. It shows that extremality requires that the support of the dominated measure be affinely independent within each convex cell of the convex partition <ref:2609. p. 213696#pg5>. This result generalizes the one-dimensional characterizations of extreme points obtained by Kleiner et al. p. 2 (2021) and Arieli et al. p. 1 (2023) <ref:2609. p. 013696#pg5>. Furthermore, it accommodates full-revelation regions and lower-dimensional pooling regions that naturally arise from the convex partition in Theorem 1 <ref:2609. p. 313696#pg5>.

Role of Regularity Conditions

To obtain a sharp and tractable characterization as Theorem 1, we impose a mild regularity condition—Assumption 1—which requires that the relative boundaries of the convex cells be negligible under the prior <ref:2609.13696#pg5>. This assumption eliminates the possibility that a point lies simultaneously on the boundary of multiple convex cells in a way that affects feasible martingale transports <ref:2609. p. 913696#pg11>. When Assumption 1 fails, boundary points introduce additional degrees of freedom that can break global extremality even when each local component is well-behaved <ref:2609. p. 313696#pg5>.

Application to Moment Persuasion

In Section 6, we apply our characterization of the extreme points of the mean-preserving contraction set to moment persuasion, a special case of the Bayesian persuasion problem <ref:2609.13696#pg6>. Under these assumptions, the optimal information structure corresponds precisely to the extreme points of the mean-preserving contraction set characterized in this paper <ref:2609. p. 413696#pg6>. The two conditions in Theorem 3 jointly characterize one of the optimal signals in moment persuasion <ref:2609. p. 413696#pg6>. Fully revealing regions correspond to singleton cells, and pooling regions correspond to higher-dimensional cells <ref:2609. p. 013696#pg6>.

Implementability via Martingale Transports

Section 5 establishes that if there exists only one full dimensional concentration region and Assumption 1 holds, the set of martingale transports between μ and μ0 is remarkably flexible <ref:2609.13696#pg6>. This result implies that for any dominated measure concentrated on a convex cell, there exists a martingale transport whose conditional distributions span the entire support of the prior <ref:2609. p. 413696#pg6>.

Failure Modes and Subtleties

When Assumption 1 fails, the clean separation may break down because boundary points may simultaneously belong to the closures of multiple cells, and martingale transports may exploit these boundary interactions to mix mass across otherwise distinct concentration regions <ref:2609.13696#pg17>. This additional flexibility can destroy extremality even when each local component satisfies affine independence <ref:2609. p. 313696#pg18>. Theorem 3 provides necessary conditions for extremality even when Assumption 1 fails, but these conditions are no longer sufficient <ref:2609. p. 313696#pg18>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive signal selection for moment persuasion systems can be optimized by identifying extreme points of mean-preserving contractions, specifically those with affine independent supports within each irreducible component of a canonical convex partition. This allows systems to choose signals that either reveal states exactly in fully revealing regions or pool states in higher dimensional cells while maintaining the necessary geometric constraints.

  2. Robustness against prior measure irregularities can be enhanced by incorporating Assumption 1 checks, which ensures that the relative boundaries of non-singleton concentration regions have zero prior mass. This prevents martingale transports from exploiting boundary interactions to mix mass across distinct concentration regions, leading to more predictable and stable optimal information structures in higher dimensions.

  3. Moment persuasion algorithms can be designed to leverage the implementational richness guaranteed by Theorem 4, which states that for any point in the support of a measure, there exists a martingale transport whose conditional distribution spans the entire support of the prior under Assumption 1. This ensures that even complex posterior mean distributions can be realized through feasible information structures.

Sources

Related papers