📈 economics

Consistent Beliefs without Common Prior

This paper extends Morris's characterization of the common prior with full support from finite to infinite type spaces, demonstrating that this result holds regardless of whether beliefs are countably or purely additive, and suggesting that the concept of a real common prior may not be fundamentally meaningful.

Original authors: Ziv Hellman, Miklós Pintér

Published 2026-02-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ziv Hellman, Miklós Pintér

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a group of friends trying to make a decision, like betting on the outcome of a sports game. In economics and game theory, there is a famous idea called the "Common Prior."

Think of the Common Prior as a shared starting point. It's the idea that before anyone looked at their private clues (like a player's injury report or a team's recent form), everyone was looking at the same "master map" of probabilities. If they all start with the same map, they shouldn't be able to make a bet where everyone thinks they are going to win. If such a bet exists, it means their maps were different to begin with, or they are confused.

For a long time, economists assumed this "master map" covered every single possibility (a concept called full support). They thought: "If everyone agrees on the map, and the map includes every possible outcome, then no one can trick the group into a bad deal."

The Paper's Big Discovery: The Map vs. The Feeling

This paper by Hellman and Pintér says: "Hold on. When we move from small, simple groups to huge, complex, or infinite worlds, that old 'master map' idea breaks down."

Here is the breakdown of their findings using simple analogies:

1. The Finite vs. Infinite Problem

Imagine you are playing a board game with a finite number of squares (like a chessboard). You can easily draw a single map that shows the probability of landing on every square. If everyone agrees on this map, you can prove mathematically that no "win-win" scam exists.

But now, imagine the game is played on an infinite track that never ends.

The Old View: We still try to draw one single "master map" (a probability distribution) that covers the whole infinite track.
The New View: The authors show that in these infinite worlds, you often cannot draw that single map, even if the players' beliefs are perfectly consistent. The "map" might not exist as a single object, even though the players aren't making any logical mistakes.

2. The "Acceptable Bet" Test

The paper uses a test called an "Acceptable Bet" to see if beliefs are consistent.

The Scenario: Imagine a group of people are offered a series of bets.
The Rule: A bet is "acceptable" if every person thinks they will win (or at least not lose) in every possible scenario, but at least one person thinks they will win strictly more than zero.
The Logic: If such a bet exists, the group is inconsistent. They are all "hallucinating" a profit that isn't there. If no such bet exists, their beliefs are Strongly Consistent.

The authors prove that in infinite worlds, you can have a group where no such "win-win" scam exists (their beliefs are consistent), yet you cannot point to a single shared probability map that explains why.

3. The Two Identical Twins (Examples 8 & 9)

This is the most mind-bending part of the paper. The authors create two different scenarios (Example 8 and Example 9) that look almost identical on the surface:

Scenario A: Two players, Anne and Ben. They share the exact same "prior" (a uniform distribution). Their beliefs are Strongly Consistent. No scam is possible.
Scenario B: Two players, Anne and Ben. They share the exact same "prior" as in Scenario A. But here, their beliefs are NOT Strongly Consistent. A scam (an acceptable bet) is possible.

The Metaphor:
Imagine two identical twins.

In Case 1, they both look at the same photo album. They agree on everything. You can't trick them into a bad deal.
In Case 2, they look at the exact same photo album. But somehow, you can trick them into a bad deal.

How is this possible? The paper argues that in infinite worlds, consistency isn't about the photo album (the probability distribution); it's about the geometry of how their beliefs fit together.

It's like two puzzle pieces. In a small box (finite world), if the pieces fit, you can see the picture (the common prior). In a giant, infinite box, the pieces might fit perfectly (no scam possible) without ever forming a single, complete picture you can hold in your hand.

4. The Conclusion: Stop Looking for the "Ghost"

The paper concludes that in complex, infinite worlds, trying to find a "Common Prior" (a single probability distribution) is like looking for a ghost. It might not be there, even if the players are behaving perfectly rationally.

Instead of saying "They have a Common Prior," the authors suggest we should say "They have Strong Consistency of Beliefs."

Old Way: "They agree because they share a master map."
New Way: "They agree because their individual maps are arranged in a way that prevents any logical scams, even if no single master map exists to tie them together."

Summary

The paper tells us that in the real world (which is often complex and infinite), consistency does not require a shared starting point. You can have a group of people who never make a "win-win" mistake, even if they don't share a single, unified probability distribution. The "Common Prior" is a useful fiction for simple worlds, but in complex ones, it's better to focus on the consistency of the beliefs themselves rather than the non-existent "prior" that supposedly caused them.

Technical Summary: Consistent Beliefs without Common Prior

Problem Statement
The literature on the common prior assumption (CPA) generally follows two strands. The first treats the common prior as a probability distribution consistent with all players' posteriors (Aumann, 1976; Samet, 1998). The second, more restrictive strand assumes the common prior has full support, meaning every type of every player is assigned positive probability (Morris, 1991, 1994).

Morris (1991, 1994) established an epistemological–behavioral duality for finite type spaces: a finite type space admits a common prior with full support if and only if it contains no acceptable bet (a bet where all players expect non-negative payoffs, with at least one strictly positive expectation). However, the concept of "full support" is problematic in infinite state spaces. Furthermore, existing characterizations of common priors often fail in infinite settings unless strong additional assumptions (e.g., compactness or countable state spaces) are imposed, particularly when modeling beliefs as $\sigma$ -additive probability measures.

This paper addresses the gap in characterizing the "common prior with full support" in general infinite type spaces and investigates whether such a prior can be meaningfully represented as a probability distribution.

Methodology
The authors employ an epistemic–behavioral dual characterization approach, extending Morris's results to infinite state spaces. The methodology involves:

Generalizing Definitions: Extending the notion of an "acceptable bet" to infinite state spaces.
Convex Separation: Utilizing the separation of convex sets in topological vector spaces. The paper distinguishes between strong separation (corresponding to the common prior–agreeable bet duality) and proper separation (corresponding to the strong consistency of beliefs–acceptable bet duality).
Model Robustness: Analyzing the results under two distinct modeling frameworks for beliefs:
- Probability Measures: $\sigma$ -additive probabilities (standard measure theory).
- Probability Charges: Finitely additive probabilities (charges), which allow for a broader class of infinite models.
Counter-Examples: Constructing specific infinite type spaces (Examples 8 and 9 in the main text; Examples 4–6 in the supplement) where players share identical prior sets (intersections) but differ in whether their beliefs are "strongly consistent."

Key Contributions
The paper makes three primary contributions:

Extension to Infinite Type Spaces: The authors generalize Morris's (1991, 1994) characterization of the common prior with full support to infinite type spaces. They introduce the concept of strong consistency of beliefs (Definition 5) as the appropriate infinite-dimensional analogue.
Robustness to Probability Models: The characterization of strong consistency of beliefs via the absence of acceptable bets is shown to be robust. It holds true whether beliefs are modeled as $\sigma$ -additive probability measures (Theorem 7) or as finitely additive probability charges (Supplementary Material). This contrasts with the "common prior–agreeable bet" characterization, which fails in the measure model without significant additional assumptions.
Re-evaluation of the Common Prior as a Distribution: The paper demonstrates that in infinite state spaces, the notion of a common prior with full support cannot generally be represented by a single probability distribution or a set of distributions. Through Examples 8 and 9, the authors show that two players can have identical prior sets (with the same intersection) yet possess beliefs that are strongly consistent in one case but not in the other. This implies that proper separation of convex sets (the mathematical condition for strong consistency) is not always detectable by the intersection of the sets.

Key Results

Theorem 7 (Main Result): In a type space $T$ , the players' beliefs are strongly consistent if and only if there exists no acceptable bet within the type space. This establishes the epistemic–behavioral duality for infinite spaces.
Lemma 6: In finite type spaces, strong consistency is equivalent to the non-empty intersection of the relative interiors of the players' prior sets ( $\cap \text{int} \Pi_i \neq \emptyset$ ). This "witness" (a specific probability distribution) disappears in infinite spaces.
Examples 8 and 9: These examples illustrate that even when $\Pi_{Anne} \cap \Pi_{Ben}$ is identical (containing only a uniform distribution $\nu$ ), the beliefs may be strongly consistent in one scenario (no proper separation possible) and not in the other (proper separation possible).
Implication for Representation: Because proper separation cannot be guaranteed by the intersection of prior sets in infinite dimensions, the "common prior with full support" cannot be universally defined as a probability distribution.

Significance and Claims
The paper argues that the traditional view of the common prior with full support as a concrete probability distribution (or a set of distributions) serving as a witness to belief consistency is not sustainable in infinite state spaces.

Terminological Shift: The authors propose adopting the term "strong consistency of beliefs" instead of "common prior with full support." This shift moves the focus from a potentially non-existent distribution to the structural relationship between players' priors.
Nature of the Ex-Ante Stage: The findings suggest that the ex-ante stage should be viewed as a non-primitive element of the model. In infinite settings, the absence of arbitrage (no acceptable bet) should be interpreted as the (strong) consistency of beliefs rather than the existence of a real common prior distribution.
Contribution to Debate: This result contributes to the debate initiated by Aumann (1998) and Gul (1998) regarding the necessity and interpretation of common priors. The paper concludes that while the "common prior" intuition is useful in finite settings, in infinite settings, the concept is best understood as a consistency condition on beliefs rather than a specific probabilistic object.

The paper does not propose new experimental applications or future research directions beyond the theoretical implications for the interpretation of type spaces and priors in game theory and economics.