Elias Tsakas

# Abstract

In this note, we extend Aumann’s agreement theorem to a framework where beliefs are modelled by conditional probability systems à la Battigalli, P., and M. Siniscalchi. 1999. “Hierarchies of Conditional Beliefs and Interactive Epistemology in Dynamic Games.” Journal of Economic Theory 88: 188–230. We prove two independent generalizations of the agreement theorem, one where the agents share some common conditioning event, and one where they may not.

## Proof of Proposition 1.

First of all observe that B covers X, since PB is a partition of X. Then, consider an arbitrary collection (αP)PP of positive reals such that PPαP=1. For each AA define

(9)p(A):=PPαPμ(A|P).

Verifying that p is a probability measure in Δ(X,A) is trivial. Moreover, notice that for every AA and every BB with p(B)>0, it is the case that

(10)p(AB)p(B)=PPαPμ(AB|P)PPαPμ(B|P)=μ(A|B)PPαPμ(B|P)PPαPμ(B|P)
=μ(A|B),

with eq. (10) following from (C3) since either ABBP or BP=.

## Proof of Theorem 1.

Define TiB:={tiTi:ti(ω)=tifor some ωCK([qa]B[qb]B)supp(p)[B]}, which is by hypothesis nonempty. Since the event [qi]B is measurable with respect to the partition {[ti]|tiTi}, either [ti][qi]B or [ti][qi]B=. Now, for an arbitrary tiTiB, it is necessarily the case that [ti]CK([qi]B)[ti]Ki([qi]B)[ti][qi]B. Hence, for each tiTiB,

qi=πti(E|B)=p(E|[B][ti]).

The second equality follows from p([B][ti])>0, which is by definition true for all tiTiB. Now, define the set M:=B×TaB×TbB. Hence,

qi=tiTiBqip([B][ti]|M)=tiTiBp(E|[B][ti])p([B][ti]|M)=p(E|M).

Finally, since p(E|M) does not depend on iI, it is the case that qa=qb.

## Proof of Theorem 2.

Step 1. Define Ti:={tiTi:ti(ω)=tifor some ωCK([qa][qb])supp(p)[B]}, which is by hypothesis nonempty. Then, define the set M:=B×Ta×Tb. Then, following Tsakas and Voorneveld (2011), Proof of Lemma, we define Qi as the coarsest partition of B that generates Bi, i.e., formally for every θB, the element of Qi that contains θ is defined by Qi(θ):=BBi:θBB. Thus, we obtain

(11)p(E|M)=QQip(E|Q×Ta×Tb)p(Q×Ta×Tb|M)=QQi(BBi:BQλB)p(E|Q×Ta×Tb)p(Q×Ta×Tb|M)
(12)=BBiλBQQi:QBp(E|Q×Ta×Tb)p(Q×Ta×Tb|M)=BBiλBQQi:QBp(E|Q×Ta×Tb)p(Q×Ta×Tb|B×Ta×Tb)p(B×Ta×Tb|M)=BBiλBp(B×Ta×Tb|M)p(E|B×Ta×Tb),

with eq. (11) following from the fact that Bi is balanced. Before moving forward, let us point out that p(Q×Ta×Tb) or p(B×Ta×Tb) might in principle be equal to 0 for some QQi or BBi respectively. In such cases, the conditional probabilities given these null events take arbitrary values.

Step 2. Now, similarly to the proof of Theorem 1, it is the case that [ti][qi] for all tiTi, thus implying πti(E|B)=qi for all BBi and all tiTi.

Step 3. Now take an arbitrary BBi such that p(E|B×Ta×Tb)>0. Then, there exists some tiTi such that p([B][ti])>0, and therefore since p is a common prior, we obtain p(E|B×Ta×Tb)=qi for all BBi with p(E|B×Ta×Tb)>0. Hence, by eq. (12) it follows that

(13)p(E|M)=qiBBiλBp(B×Ta×Tb|M)=qiBBiλBQQip(B×Ta×Tb|Q×Ta×Tb)p(Q×Ta×Tb|M)=qiQQip(Q×Ta×Tb|M)BBiλBp(B×Ta×Tb|Q×Ta×Tb)=qiQQip(Q×Ta×Tb|M)
=qi,

with eq. (13) following again from the fact that Bi is balanced. Indeed, the conditional probability can be seen as an indicator function, with p(B×Ta×Tb|Q×Ta×Tb)=1 if QB and p(B×Ta×Tb|Q×Ta×Tb)=0 otherwise. Finally, notice that p(E|M) does not depend on iI, thus completing the proof.

## Proof of Corollary 1.

Recall from Remark 1 that every treelike Bi is also balanced. Moreover, by (T2) it follows that Ba=Bb=Θ for all iI. Hence, CK([qa][qb])supp(p) directly implies CK([qa][qb])supp(p)[B]. Thus, all the conditions of Theorem 2 are satisfied, and therefore qa=qb.

### References

Alós-Ferrer, C., and Ritzberger K. 2016. The Theory of Extensive Form Games. Springer.Search in Google Scholar

Aumann, R.J. 1976. “Agreeing to Disagree.” Annals of Statistics 4: 1236–1239.10.1214/aos/1176343654Search in Google Scholar

Aumann, R.J., and Brandenburger A. 1995. “Epistemic Conditions for Nash equilibrium.” Econometrica 63: 1161–1180.10.2307/2171725Search in Google Scholar

Bach, C.W., and Perea A. 2013. “Agreeing to Disagree with Lexicographic Prior Beliefs.” Mathematical Social Sciences 66: 129–133.10.1016/j.mathsocsci.2013.03.004Search in Google Scholar

Bonanno, G., and Nehring K. 1997. “Agreeing to Disagree: A Survey.” Working Paper, UC Davis.Search in Google Scholar

Bonanno, G., and Nehring K. 1999. “How to Make Sense of the Common Prior Assumption Under Incomplete Information.” International Journal of Game Theory 28: 409–434.10.1007/s001820050117Search in Google Scholar

Battigalli, P., and Siniscalchi M. 1999. “Hierarchies of Conditional Beliefs and Interactive Epistemology in Dynamic Games.” Journal of Economic Theory 88: 188–230.10.1006/jeth.1999.2555Search in Google Scholar

Battigalli, P., and Siniscalchi M. 2002. “Strong Belief and Forward Induction Reasoning.” Journal of Economic Theory 106: 356–391.10.1006/jeth.2001.2942Search in Google Scholar

Brandenburger, A., Friedenberg A., and Keissler H.J. 2007. “Notes on the Relationship Between Strong Belief and Assumption.” Unpublished manuscript.Search in Google Scholar

Brandenburger, A., Friedenberg A., and Keissler H.J. 2008. “Admissibility in Games.” Econometrica 76: 307–352.10.1111/j.1468-0262.2008.00835.xSearch in Google Scholar

Feinberg, Y. 2000. “Characterizing Common Priors in the form of Posteriors.” Journal of Economic Theory 91: 127–179.10.1006/jeth.1999.2592Search in Google Scholar

Geanakoplos, J. 1989. “Game Theory Without Partitions, and Applications to Speculation and Consensus.” Cowles Foundation Discussion Paper 914.Search in Google Scholar

Halpern, J.Y. 2002. “Characterizing the Common Prior Assumption.” Journal of Economic Theory 106: 316–355.10.1006/jeth.2002.2925Search in Google Scholar

Halpern, J.Y. 2010. “Lexicographic Probability, Conditional Probability, and Nonstandard Probability.” Games and Economic Behavior 68: 155–179.10.1016/j.geb.2009.03.013Search in Google Scholar

Harsanyi, J. 1967-68. “Games with Incomplete Information Played by Bayesian Players, I-III.” Management Science 14: 159–182, 320–334, 486–502.Search in Google Scholar

Milgrom, P., and Stokey N. 1982. “Information, Trade and Common Knowledge.” Journal of Economic Theory 26: 17–27.10.1016/0022-0531(82)90046-1Search in Google Scholar

Monderer, D., and Samet D. 1989. “Approximating Common Knowledge with Common Beliefs.” Games and Economic Behavior 1: 170–190.10.1016/0899-8256(89)90017-1Search in Google Scholar

Rényi, A. 1955. “On a New Axiomatic Theory of Probability.” Acta Mathetica Academiae Scientiarum Hungaricae 6: 285–335.10.1007/BF02024393Search in Google Scholar

Sebenius, J., and Geanakoplos J. 1983. “Don’t Bet on It: Contingent Agreements with Asymmetric Information.” Journal of the American Statistical Association 78: 424–426.10.1080/01621459.1983.10477988Search in Google Scholar

Tsakas, E. 2014. “Epistemic Equivalence of Extended Belief Hierarchies.” Games and Economic Behavior 86: 126–144.10.1016/j.geb.2014.03.008Search in Google Scholar

Tsakas, E., and Voorneveld M. 2011. “On Consensus Through Communication Without a Commonly Known Protocol.” Journal of Mathematical Economics 47: 733–739.10.1016/j.jmateco.2011.10.003Search in Google Scholar

## Note

I am indebted to three anonymous referees and the associate editor for their valuable comments. This note supersedes two previous papers by the same author, titled “Strong belief and agreeing to disagree” and “Hierarchies of conditional beliefs derived from commonly known priors” respectively.

Published Online: 2018-05-04

© 2018 Walter de Gruyter GmbH, Berlin/Boston