Consistent Beliefs without Common Prior

Ziv Hellman, Miklós Pintér · 2025

Read Consistent Beliefs without Common Prior by Ziv Hellman, Miklós Pintér (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-09835.txt.

Introduction

Ziv Hellman and Miklós Pintér revisit in 2025 a foundation of game theory: the common prior, the idea that all players' beliefs spring from one shared probability view. A classic result ties such a prior, when it gives every type some chance, to the absence of a bet all sides would accept. That result covered only finite type spaces.

The pair extend the characterization to infinite type spaces. Their extension holds whether beliefs are built with standard countable probability or with the weaker purely additive kind. A deeper lesson follows: a real common prior, read as one distribution or even a set of them, may not always be a meaningful notion. The paper is a work of economic theory, aimed at the logic of belief itself rather than at any market data.

Catalog record

CREATORZiv Hellman, Miklós Pintér
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-09835

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Consistent Beliefs without Common Prior Thanks: We thank Jack Stecher and participants at SAET 2022, SOR 2023, the 14th Annual Financial Market Liquidity Conference, EWET 2024, RUD 2025, and SAET 2025 for their helpful comments and suggestions. Hellman gratefully acknowledges the support by Israel Science Foundation grants 448/22 and 458/25. Pintér acknowledges the support by the Hungarian Scientific Research Fund under the project K 146649 and by the HUN-REN under the project TKCS-2024/34. Ziv Hellman, Miklós Pintér Abstract In a strand of the literature, it is assumed that the common prior has full support; that is, every type of every player is assigned positive probability. Morris (1991) ; Morris (1994) established an epistemological–behavioral duality characterisation of the common prior with full support, showing that a finite type space admits such a prior if and only if it contains no acceptable bet. This result forms the basis of the present paper. The paper makes three contributions: (1) The characterisation of Morris (1991) ; Morris (1994) is extended to infinite type spaces. (2) The extension is robust: it does not depend on whether the infinite model applies countably additive or purely additive probabilities as beliefs. (3) The analysis implies that the notion of a real common prior—understood as a single probability distribution or a set of probability distributions—is not necessarily meaningful. Keywords: Type space, Information structure, Common prior, Consistencies of Beliefs, Prior, Bet 1 Introduction There are two strands in the literature concerning the notion of a common prior. The first treats the common prior as a probability distribution over the state space that is consistent with, or serves as a prior for, the posteriors of all players (see Aumann (1976) ; Samet (1998) ; Feinberg (2000) , among others). The second strand adopts the same notion of the common prior as does the one in the previous paragraph, but further assumes that the prior has full support (see Morris (1991) ; Morris (1994) ; Morris (2020) ; Frick et al (2023) ; Morris et al (2024) , among others). Here, full support means that every type of every player is assigned positive probability under the common prior. Henceforth, we refer to this as the common prior with full support assumption. At first glance, the distinction between a common prior and a common prior with full support may appear to be minor, but we will show here that this is not the case. In the epistemic–behavioral dual characterisation ( Samet, 1998 ; Morris, 1994 ) 1 1 1 The epistemic pillar is the notion of a common prior, and the behavioural pillar is the notion of a bet (or trade). Characterisation means the pillars are equivalent. , the behavioural counterpart of a common prior is an agreeable bet ( Samet, 1998 ; Feinberg, 2000 ) , where every player at every state expects a positive payoff. In contrast, the behavioural counterpart of a common prior with full support is an acceptable bet ( Morris, 1994 ) , where all expected payoffs are non-negative and some are strictly positive. While the concept of full support is meaningful in finite state spaces, it is less so in infinite ones. Our first result generalises the notion of common prior with full support to infinite state spaces via the epistemic–behavioral dual characterisation: we extend acceptable bets to infinite state spaces and prove a corresponding dual characterisation theorem (Theorem 7 ). The resulting generalised notion of a common prior we term strong consistency of beliefs (Definition 5 ). When generalising the notion of a common prior to infinite state spaces, a key question is which model to adopt. One approach treats beliefs and (common) priors as probability measures ( \sigma -additive probabilities) ( Mertens and Zamir, 1985 ; Brandenburger and Dekel, 1993 ; Heifetz, 1993 ; Morris, 1994 ; Heifetz and Samet, 1998 ; Feinberg, 2000 ; Lehrer and Samet, 2014 ) . There are other approaches that models beliefs and priors as probability charges (additive probabilities) ( Hellman and Pintér, 2022 ) . The common prior–agreeable bet dual characterisation fails in the probability measures model ( Feinberg, 2000 ) , unless one adds significant additional assumptions such as compactness or a countable state space ( Feinberg, 2000 ; Lehrer and Samet, 2014 ) . However, but it generalises straightforwardly in the probability charges model ( Hellman and Pintér, 2022 ) . The second result of this paper is that the generalisation of strong consistency of beliefs (common prior with full support) — the acceptable bet epistemic–behavioral dual characterisation — is robust to the choice of model. That is, it holds both for probability measures (Theorem 7 ) and for probability charges (Supplementary Material). It follows that the strong consistency of beliefs characterisation is less fragile than the common prior–agreeable bet dual characterisation – a point in favour of the former. With regards to the concept of type spaces, these admit two interpretations. One views them as models of asymmetric information, with a concrete ex-ante stage from which players update their beliefs. The other treats them as tools for incomplete information, representing hierarchies of beliefs while the ex-ante stage is merely an abstraction. In the asymmetric information case, priors are elicited via ex-ante bets and have clear meaning. In the incomplete information case, priors are purely analytical, and the ex-ante stage is at most a fiction. Regardless of whether the ex-ante stage is regarded as concrete or merely fictional, the common prior—as a probability distribution—serves as a witness that players’ beliefs are consistent in the type space. Mathematically, an epistemic–behavioural dual characterisation relies on the separation of convex sets: strong separation corresponds to the common prior–agreeable bet characterisation, while proper separation corresponds to the strong consistency of beliefs (common prior with full support)–acceptable bet characterisation. Hence, a common prior, interpreted as a probability distribution or a set of distributions, can indicate the presence or absence of strong or proper separation. The third result of this paper is illustrated in Examples 8 and 9 . These examples show that proper separation of convex sets may not be detected by the intersection of the sets. In finite-dimensional spaces, two convex sets are properly separated if and only if the intersection of their relative interiors is nonempty; an element of this intersection serves as a witness to proper separation. Each example features two players and two closed convex sets in the space of probability distributions with identical intersections. In one example, the sets are properly separable; in the other, they are not. These examples imply that, in infinite state spaces, the common prior with full support cannot generally be represented by a probability distribution or a set of distributions. Consequently, not only may a concrete ex-ante stage be meaningless, but even a probability distribution as a convenient abstraction may fail to capture the notion. Building on this observation, we propose adopting the term (strong) consistency of beliefs instead of common prior (with full support). This result contributes to the debate initiated by Aumann (1998) and Gul (1998) . In summary, given the insights gained from infinite models, we prefer the term strong consistency of beliefs for common prior with full support, and more generally advocate the use of consistency of

Read the complete text