Pith. sign in

REVIEW 2 major objections 6 minor 30 references

Markov Information Processes

T0 review · 2 major / 6 minor · reviewed 2026-07-11 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Far-sighted agents who move a Markov state obey recommendations only if the designer prices the future; better models of the dynamics earn rent that grows only logarithmically under excitation.

desk verdict Solid controlled-Markov extension of BCE with recursive design and clean LQG forms; the log-rent theorem is conditional on the paper’s own Assumption 2, which it already flags as Conjecture 11.4. read the letter →

arxiv 2607.04308 v1 pith:PBF7UZAW submitted 2026-07-05 math.OC econ.TH

Furkan Sezer This is my paper · ORCID
classification math.OCecon.TH MSC 91A1591A2690C4093E20
keywords informationdesigncontrolledMarkovprocessesdynamicgamesBayescorrelatedequilibriumlearninginsystemidentificationlinear-quadratic-Gaussianpromisedutilities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper extends static Bayes correlated equilibrium to a setting where strategically interacting agents are far-sighted and their actions drive a persistent controlled Markov state. The central object is the Markov Bayes correlated equilibrium, defined by a dynamic obedience condition that adds a continuation-value differential to the familiar static obedience constraint; the differential vanishes exactly when actions cannot move the state or the horizon is one-shot. Recommending actions remains without loss, and the designer’s problem is recursive in the agents’ promised continuation utilities, solved by a set-valued backward-induction algorithm whose optimum exists and sits between the no-disclosure and first-best values. In the linear-quadratic-Gaussian class the same condition becomes a covariance restriction with a modified interaction matrix and collapses, in the stationary case, to a discounted algebraic Riccati fixed point. When agents instead observe the state and learn the transition, any agent that holds a sharper model of the dynamics than the designer anticipated extracts a non-negative rent that is zero at the known-dynamics benchmark and can be deterred only by building slack into obedience; under persistent excitation the cumulative rent grows only as the square of the log of time.

What carries the argument

The dynamic obedience condition (Definition 6.1): the static stage-payoff comparison plus a discounted continuation-value differential Δ that prices how a deviation today changes the law of tomorrow’s state; this is the object that couples constraints across time, reduces to ordinary BCE when the differential vanishes, and becomes a covariance condition with modified interaction matrix in the LQG case.

What would settle it

Simulate or solve a small LQG instance in which excitation is purely endogenous, agents learn from their own possibly off-path deviations, and each tracks the others’ estimators; if the cumulative rent grows faster than any log-squared envelope, the rate claim fails outside the paper’s maintained assumptions.

Watch

Extended reading notes

Core claim

Markov Bayes correlated equilibrium is the controlled-Markov generalisation of Bayes correlated equilibrium for far-sighted agents. It is characterised by a dynamic obedience condition that augments the static Bergemann–Morris condition with an explicit continuation-value differential; recommending actions is without loss, the designer’s recursive problem in promised utilities attains an optimum between no-disclosure and first-best, and an agent’s rent from a superior model of the dynamics is non-negative, zero at the known-dynamics benchmark, and ˜O(log T) cumulative under persistent excitation.

Load-bearing premise

The logarithmic cumulative-rent bound assumes agents estimate only from on-path obedient data, the designer injects exogenous persistent excitation, and agents treat the common environment as fixed without tracking one another’s evolving models.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper develops information design for far-sighted, strategically interacting agents whose actions control a persistent Markov state. It introduces the Markov Bayes correlated equilibrium (Markov BCE), characterised by a dynamic obedience condition (Definition 6.1, eq. 5) that augments Bergemann–Morris static obedience with a continuation-value differential Δ^i_h and reduces to the static condition when H=1, B_h=0, or δ=0 (Proposition 6.2). A dynamic revelation principle (Proposition 7.1) justifies action recommendations under unmonitored within-stage deviations. The designer’s problem is cast recursively in promised continuation utilities and solved by an APS-style set-valued backward induction (Algorithm 1); existence and value bounds between no-disclosure and first-best are proved (Propositions 7.2–7.4). In the LQG class, dynamic obedience becomes a covariance condition with a modified interaction matrix Φ̃_h (Theorem 8.2), and the stationary case is a discounted algebraic Riccati fixed point (Proposition 8.3). Part II defines an agent’s rent from a superior model of the transition (non-negative, zero at the known-dynamics benchmark), gives a closed-form LQG expression, and proves a ˜O(log T) cumulative-rent bound under Assumption 2 (Theorem 11.3), leaving the unconditional case as Conjecture 11.4. Two scalar LQG examples (evacuation, power coordination) and a numerical study illustrate the theory.

Significance. If the results hold, the paper supplies a clean controlled-Markov specialisation of dynamic Bayes correlated equilibrium that is missing from both the static BCE literature and the myopic Markov-persuasion stream. The dynamic obedience condition, the recursive promised-utility formulation, and the LQG covariance/Riccati characterisation are genuine contributions that make multi-agent, far-sighted information design tractable in a standard control setting. The rent analysis in Part II is a novel object—an agent’s informational advantage over a committed designer’s model of the dynamics—and the conditional logarithmic bound is carefully derived from self-normalised least squares plus Riccati Lipschitz continuity. Strengths include explicit reduction to the static case, existence under standard compactness/Feller assumptions, a complexity statement for the LQG recursion (Proposition 8.4), and transparent isolation of the open unconditional learning problem as a conjecture. The framework is directly usable for congestion and resource-coordination applications of the type sketched in Section 12.

major comments (2)
  1. [Abstract; §3 Contributions; Theorem 11.3 / Assumption 2] Theorem 11.3 and the corresponding contributions bullet / abstract sentence state a ˜O(log T) cumulative-rent bound. The proof relies on Assumption 2 (L2)–(L4): on-path (counterfactual) estimation, exogenous persistent excitation λ_min(Λ_t) ≥ λ_0 t, and non-entangled estimators of a common fixed environment. The paper correctly isolates the unconditional case as Conjecture 11.4, but the abstract and the fifth contributions bullet present the logarithmic rate without an explicit qualifier that it holds only under those three modelling restrictions. Because the design tension noted after the theorem (excitation both sustains obedience and accelerates learning) is precisely what makes (L3) non-innocuous, the abstract and contributions list should state the conditioning assumptions in the same sentence as the rate claim so that the result is not over-read.
  2. [Proposition 7.1 / Remark 1; §12] Proposition 7.1 (dynamic revelation principle) and Remark 1 correctly restrict the principle to the class in which the designer cannot monitor or contract on within-stage actions; deviations propagate only through the realised next state. The two worked examples (evacuation staggering, power reserve coordination) are drawn from settings in which a planner often can observe egress rates or reserve provision. The manuscript does not discuss how much of the implementable set would change if the designer could condition continuation policy on observed within-stage actions, nor whether the Markov BCE recommendations remain approximately optimal under partial monitoring. A short paragraph in §7 or §12 clarifying the scope for the motivating applications would prevent misapplication of the recommendation-policy formulation.
minor comments (6)
  1. [Abstract; throughout] Abstract and several body paragraphs contain irregular word-internal spaces (e.g., “charact erised”, “de signer’s”, “s et-valued”, “covaria nce”). These appear to be line-break artefacts and should be cleaned for the camera-ready version.
  2. [§12.1 / Figure 1] Figure 1 (right panel) is described as tracking the c log² t envelope of Theorem 11.3, but the plotted object is cumulative squared estimation error ∑∥θ̂_k−θ∥² rather than cumulative rent. A one-sentence clarification that the estimation error is the driver of the rent bound (Rent_t ≤ C∥G_t∥²) would make the panel self-contained.
  3. [§8, eq. (10)] In the LQG payoff (10), the term d_i(a_{-i},γ) is said not to affect best responses; this is correct for pure-strategy best responses but should be noted as potentially affecting the designer’s objective when the designer cares about opponents’ payoffs or when public randomisation is used.
  4. [Proposition 8.3 / 8.4] Proposition 8.3’s contraction condition δ∥A∥² + δL < 1 is used both for existence and for the geometric rate in Proposition 8.4(ii). A brief remark on how L (the Lipschitz constant of the policy-dependent map Π ↦ R_i) can be bounded a priori from the spectrahedron of second moments would help implementers.
  5. [§12.1 / Table 1] The numerical study imposes E[a_i²] ≤ 4 to restore compactness (consistent with Proposition 7.4). Reporting sensitivity of Table 1 to this bound (or to the asymmetric γ-variances) would strengthen the illustration.
  6. [§2.5; References] References [24] and [25] are the author’s own continuous-time Stackelberg papers; the positioning in §2.5 is appropriate, but the arXiv identifiers and dates should be double-checked for consistency with the July 2026 manuscript date.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: dynamic obedience, LQG covariance/Riccati, and rent bounds are derived from stated primitives and FOCs, not fitted or self-defined into their conclusions.

full rationale

The paper’s load-bearing claims are definitional extensions or direct derivations, not circular. Dynamic obedience (Def. 6.1 / eq. 5) is the static Bergemann–Morris condition plus an explicit continuation differential Δ^i_h; Prop. 6.2 shows reduction to the static case by direct cancellation when H=1, B_h=0 or δ=0. The dynamic revelation principle (Prop. 7.1) is a standard one-shot-deviation + pooling argument under the model’s monitoring structure. The designer’s problem is the APS recursion (Alg. 1) in promised utilities; existence and sandwich bounds (Props. 7.2–7.4) follow from compactness/Feller continuity and feasibility of Bayes–Nash recommendations, not from self-reference. In LQG, Thm. 8.2 obtains the covariance condition by differentiating the quadratic stage-plus-continuation payoff and taking second moments under joint normality; the modified matrix Φ̃_h and the Riccati recursion (Lem. 8.1, Prop. 8.3) are the ordinary controlled-LQ objects, collapsing to the static covariance condition when B=0 or δ=0. Part II’s rent (21) is defined as perceived one-shot deviation gain; Props. 10.1–10.2 and 11.1 show non-negativity and vanishing exactly when the agent’s model matches the designer’s commitment—by the definition of true-model obedience, not by fitting. The ˜O(log T) cumulative-rent theorem (Thm. 11.3) is a standard self-normalised LS + Riccati-Lipschitz argument under the paper’s own Assumption 2; the unconditional case is correctly left as Conjecture 11.4. Self-citations [24,25] supply only the strategic structure of the worked examples and are explicitly not used for any theorem. No fitted parameter is renamed a prediction, no uniqueness theorem is imported from the author, and no ansatz is smuggled via citation. The derivation chain is therefore self-contained against its stated primitives.

Assumptions & free parameters 2 free parameters · 6 assumptions · 3 invented entities

Central claims rest on standard dynamic-game and control tools (one-shot deviation, APS promised utilities, Riccati/Lyapunov maps, self-normalized least squares) plus domain modeling choices (designer commitment, unmonitored within-stage actions, controlled Markov state, far-sighted discounted agents). Part II’s rate theorem adds strong regularity and decoupling assumptions (L2–L4). Invented entities are definitional solution concepts and the rent object, not physical postulates. Numerical free parameters affect only the illustration, not the theorems.

free parameters (2)
  • Numerical LQG primitives (h0, h1, A, B, C, δ, σ²_ε, σ²_γ, action bound E[a_i²]≤4) = h0=1, h1=0.4, A=0.4, B=0.25, C=0.3, δ=0.9, σ²_ε=0.05, σ²_γ1=1.5, σ²_γ2=0.5, E[a_i²]≤4
    Chosen by hand for the congestion instance in §12.1 to illustrate utilitarian vs maximin designs and value iteration; not fitted to external data and not used in the theorems.
  • Ridge regularizers λ^{(i)} and prior means for heterogeneous estimators = λ=1 in the numerical learning panel
    Enter the constants C_1^{(i)} in the self-normalized LS bound of Theorem 11.3; illustration uses λ=1. Affect constants, not the log rate under Assumption 2.
assumptions (6)
  • standard math One-shot-deviation principle for finite-horizon discounted payoffs with unmonitored within-stage deviations that propagate only through the next state
    Used to equate incentive compatibility with dynamic obedience (Def. 6.1); standard under bounded payoffs and finite H.
  • domain assumption Designer commits ex ante to a recommendation policy and cannot condition continuation on unobserved within-stage actions (Remark 1)
    Load-bearing for the dynamic revelation principle (Prop. 7.1); if monitoring were available, richer history-dependent schemes could enlarge the implementable set.
  • standard math Abreu–Pearce–Stacchetti promised-utility recursion and set-valued backward induction
    Algorithm 1 and the recursive formulation (8)–(9) import the APS apparatus from repeated games.
  • domain assumption Φ+Φ^⊤ ≻ 0 (strict concavity of the LQG stage game) and joint Gaussianity of (γ,s,a) under linear-Gaussian recommendations
    Needed for unique stage best responses and for FOCs to be equivalent to the covariance obedience condition (Thm. 8.2).
  • ad hoc to paper Assumption 2 (L1)–(L4): stationary contractive primitives; on-path estimation; exogenous persistent excitation λ_min(Λ_t)≥λ_0 t; common fixed environment with heterogeneous ridge estimators and sub-Gaussian noise
    Required for Theorem 11.3’s ˜O(log T) cumulative rent; (L2)–(L4) deliberately shut down self-perturbed data, endogenous excitation, and mutual learning, which the paper leaves as Conjecture 11.4.
  • standard math Self-normalized least-squares concentration (Abbasi-Yadkori et al.) and local Lipschitz continuity of the Riccati map θ↦Π(θ)
    Steps 1–2 of the proof of Theorem 11.3; standard identification and implicit-function tools under δ∥A∥²+δL<1.
invented entities (3)
  • Markov information process (MKIP)
    purpose: Name the horizon-H controlled-Markov information-design environment with designer commitment and far-sighted agents.
    Definitional packaging of primitives (S, Γ, P_h, σ_h, far-sighted payoffs); no independent physical content.
  • Markov Bayes correlated equilibrium (Markov BCE)
    purpose: Solution concept: policies satisfying dynamic obedience (5) against their own induced continuation values.
    Controlled-Markov specialization of BCE / sequential BCE; characterized by the paper’s dynamic obedience condition rather than an external empirical handle.
  • Rent from superior knowledge of the dynamics (Renth,i)
    purpose: Measure an agent’s perceived one-shot deviation gain under a misspecified transition model relative to the designer’s committed policy.
    Defined in (21) and specialized to LQG closed form (27); falsifiable in principle via observed deviations under known model gaps, but introduced here as a theoretical object.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Markov Information Processes." pith.science (2026). https://pith.science/paper/PBF7UZAW

@misc{pith2026260704308,
  author       = {Pith},
  title        = {Pith review of: Markov Information Processes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PBF7UZAW}},
  note         = {Machine review of arXiv:2607.04308}
}
read the original abstract

We study information design when a designer with commitment shapes the information of strategically interacting, far-sighted agents whose actions drive a persistent, controlled Markov state. We introduce the Markov Bayes correlated equilibrium (Markov BCE), the controlled-Markov generalisation of the BCE of Bergemann and Morris (2016), characterised by a dynamic obedience condition that adds a continuation-value term to the static one and reduces to it when actions cannot move the state. Recommending actions is without loss; the designer's problem is recursive in the agents' promised continuation utilities and is solved by a set-valued backward-induction algorithm whose optimum exists and lies between the no-disclosure and first-best values. For linear-quadratic-Gaussian payoffs the obedience condition becomes a covariance condition with a modified interaction matrix, and the stationary case reduces to an algebraic Riccati equation. When agents instead learn the transition, we identify the rent an agent earns from a model of the dynamics sharper than the designer anticipates: it is non-negative, zero at the known-dynamics benchmark, and deterred only by building slack into obedience. Under persistent excitation the cumulative rent grows logarithmically as heterogeneous agents' estimates converge. Two worked examples, in congestion and resource coordination, together with a numerical study illustrate the theory.

Figures

Figures reproduced from arXiv: 2607.04308 by the authors.

Figure 1
Figure 1. Left: geometric convergence of value iteration on [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 9 canonical work pages

  1. [1]

    Bayesian persuasio n

    Emir Kamenica and Matthew Gentzkow. Bayesian persuasio n. American Economic Review , 101(6):2590–2615, October

  2. [2]

    URL https://www.aeaweb.org/articles?id=10.1257/aer.101.6.2590

    doi: 10.1257/aer.101.6.2590. URL https://www.aeaweb.org/articles?id=10.1257/aer.101.6.2590. 16 Markov Information Processes

    doi:10.1257/aer.101.6.2590

  3. [3]

    Information design: A unified perspective

    Dirk Bergemann and Stephen Morris. Information design: A unified perspective. Journal of Economic Literature, 57(1):44– 95, March 2019. doi: 10.1257/jel.20181489. URL https://www.aeaweb.org/articles?id=10.1257/jel.20181489

    doi:10.1257/jel.20181489 2019

  4. [4]

    Bayes correlated equ ilibrium and the comparison of informa- tion structures in games

    Dirk Bergemann and Stephen Morris. Bayes correlated equ ilibrium and the comparison of informa- tion structures in games. Theoretical Economics , 11(2):487–522, 2016. doi: 10.3982/TE1808. URL https://onlinelibrary.wiley.com/doi/abs/10.3982/TE1808

    doi:10.3982/te1808 2016

  5. [5]

    Robust predictions i n games with incomplete information

    Dirk Bergemann and Stephen Morris. Robust predictions i n games with incomplete information. Econometrica, 81(4):1251– 1308, 2013. doi: 10.3982/ECTA11105. URL https://onlinelibrary.wiley.com/doi/abs/10.3982/ECTA11105

    work page doi:10.3982/ecta11105 2013

  6. [6]

    Jordan, and Haifeng Xu

    Jibang Wu, Zixuan Zhang, Zhe Feng, Zhaoran Wang, Zhuoran Yang, Michael I. Jordan, and Haifeng Xu. Se- quential information design: Markov persuasion process an d its efficient reinforcement learning, 2022. URL https://arxiv.org/abs/2202.10678

    arXiv 2022

  7. [7]

    Markov persuasion processes: Learning to persuade f rom scratch

    Francesco Bacchiocchi, Francesco Emanuele Stradi, Mat teo Castiglioni, Alberto Marchesi, and Nicola Gatti. Markov persuasion processes: Learning to persuade f rom scratch. In D. Belgrave, C. Zhang, H. Lin, R. Pascanu, P . Koniusz, M. Ghassemi, and N. Chen, edit ors, Advances in Neural Infor- mation Processing Systems , volume 38, pages 70123–70160. Curran...

    2025

  8. [8]

    Markov per suasion processes with endogenous agent be- liefs

    Krishnamurthy Iyer, Haifeng Xu Xu, and Y ou Zu. Markov per suasion processes with endogenous agent be- liefs. Proceedings of The 19th Conference On W eb And InterNet Econo mics (WINE 2023) , 2023. URL https://par.nsf.gov/servlets/purl/10532237

    arXiv 2023

Show all 30 references
  1. [9]

    Toward a theory of discounted repeated games with imperfect monito ring

    David Pearce, Dilip Abreu, and Ennio Stacchetti. Toward a theory of discounted repeated games with imperfect monito ring. Econometrica, 58(5):1041–1063, 1990. URL https://www.jstor.org/stable/2938299

    1990

  2. [10]

    A continuous- time version of the princi pal: Agent problem

    Y uliy Sannikov. A continuous- time version of the princi pal: Agent problem. The Review of Economic Studies , 75(3): 957–984, 2008. ISSN 00346527, 1467937X. URL http://www.jstor.org/stable/20185061

    2008

  3. [11]

    Information desig n in multistage games

    Miltiadis Makris and Ludovic Renou. Information desig n in multistage games. Theoretical Economics, 18(4):1475–1509,

  4. [12]

    URL https://onlinelibrary.wiley.com/doi/abs/10.3982/TE4769

    doi: https://doi.org/10.3982/TE4769. URL https://onlinelibrary.wiley.com/doi/abs/10.3982/TE4769

    doi

  5. [13]

    Robert J. Aumann. Correlated equilibrium as an express ion of bayesian rationality. Econometrica, 55(1):1–18, 1987. URL https://www.jstor.org/stable/1911154

    1987

  6. [14]

    On inf ormation design in games

    Laurent Mathevet, Jacopo Perego, and Ina Taneva. On inf ormation design in games. Journal of Political Economy , 128(4): 1370–1404, 2020. doi: 10.1086/705332. URL https://www.journals.uchicago.edu/doi/abs/10.1086/705332

    2020 doi

  7. [15]

    Jeffrey C. Ely. Beeps. American Economic Review , 107(1):31–53, 2017. doi: 10.1257/aer.20150218. URL https://www.aeaweb.org/articles?id=10.1257/aer.20150218

    2017 doi

  8. [16]

    Op timal dynamic information provision

    J´ erˆ ome Renault, Eilon Solan, and Nicolas Vieille. Op timal dynamic information provision. Games and Eco- nomic Behavior , 104:329–349, 2017. ISSN 0899-8256. doi: https://doi.org /10.1016/j.geb.2017.04.010. URL https://www.sciencedirect.com/science/article/pii/S089982561730074X

    2017 doi

  9. [17]

    Ely and Martin Szydlowski

    Jeffrey C. Ely and Martin Szydlowski. Moving the goalpo sts. Journal of Political Economy , 128(2):468–506, 2020. doi: 10.1086/704387. URL https://www.journals.uchicago.edu/doi/abs/10.1086/704387

    2020 doi

  10. [18]

    Laura Doval and Jeffrey C. Ely. Sequential information design. Econometrica, 88(6):2575–2608, 2020. doi: https://doi.org/ 10.3982/ECTA17260. URL https://onlinelibrary.wiley.com/doi/abs/10.3982/ECTA17260

    2020 doi

  11. [19]

    Spear and Sanjay Srivastava

    Stephen E. Spear and Sanjay Srivastava. On repeated mor al hazard with discounting. The Review of Economic Studies , 54 (4):599–617, 10 1987. ISSN 0034-6527. doi: 10.2307/229748 4. URL https://doi.org/10.2307/2297484

    1987 doi

  12. [20]

    Dynami c mechanism design: A myersonian ap- proach

    Alessandro Pavan, Ilya Segal, and Juuso Toikka. Dynami c mechanism design: A myersonian ap- proach. Econometrica, 82(2):601–653, 2014. doi: https://doi.org/10.3982/ECT A10269. URL https://onlinelibrary.wiley.com/doi/abs/10.3982/ECTA10269

    2014 doi

  13. [21]

    Th e value of the information in the moral hazard setting

    Ishak Hajjej, Caroline Hillairet, and Mohamed Mnif. Th e value of the information in the moral hazard setting. Stochastics, 0 (0):1–35, 2025. doi: 10.1080/17442508.2025.2587756. URL https://doi.org/10.1080/17442508.2025.2587756

    2025 doi

  14. [22]

    Improved algorithms for linear stochas- tic bandits

    Yasin Abbasi-yadkori, D´ avid P´ al, and Csaba Szepesv´ ari. Improved algorithms for linear stochas- tic bandits. In J. Shawe-Taylor, R. Zemel, P . Bartlett, F. Pe reira, and K. Weinberger, editors, Ad- vances in Neural Information Processing Systems , volume 24. Curran Associ...

    2011

  15. [23]

    Jor dan, and Benjamin Recht

    Max Simchowitz, Horia Mania, Stephen Tu, Michael I. Jor dan, and Benjamin Recht. Learning without mixing: Towards a sharp analysis of linear system identification. In S´ ebasti en Bubeck, Vianney Perchet, and Philippe Rigollet, editors , Proceed- ings of the 31st Conference On ...

    2018

  16. [24]

    On the sample complexity of the linear quadr atic regulator

    Sarah Dean, Horia Mania, Nikolai Matni, Benjamin Recht , and Stephen Tu. On the sample complexity of the linear quadr atic regulator. F oundations of Computational Mathematics , 20(4):633–679, 2020. doi: 10.1007/s10208-019-09426-y. URL https://link.springer.com/article/10.1007...

    2020 doi

  17. [25]

    Fudenberg and D.K

    D. Fudenberg and D.K. Levine. The Theory of Learning in Games . Economics Learn- ing and Social Evolution Series. MIT Press, 1998. ISBN 97802 62061940. URL https://mitpress.mit.edu/9780262529242/the-theory-of-learning-in-games/ . 17 Markov Information Processes

    1998

  18. [26]

    Distributionally robust joint informat ion and mechanism design for multi-area power system coordi nation,

    Furkan Sezer. Distributionally robust joint informat ion and mechanism design for multi-area power system coordi nation,

  19. [27]

    URL https://arxiv.org/abs/2606.24015

    arXiv

  20. [28]

    Continuous-time information design for hurricane evacuation: Disclosure, congestion, and optima l phasing under model uncertainty, 2026

    Furkan Sezer. Continuous-time information design for hurricane evacuation: Disclosure, congestion, and optima l phasing under model uncertainty, 2026. URL https://arxiv.org/abs/2606.30320

    2026 arXiv

  21. [29]

    Continuous-time persuasion by filter- ing

    Ren´ e A¨ ıd, Ofelia Bonesini, Giorgia Callegaro, and Lu ciano Campi. Continuous-time persuasion by filter- ing. Journal of Economic Dynamics and Control , 176:105100, 2025. doi: 10.1016/j.jedc.2025.105100. URL https://www.sciencedirect.com/science/article/pii/S0165188925000661

    2025 doi

  22. [30]

    Judd, Sevin Yeltekin, and James Conklin

    Kenneth L. Judd, Sevin Yeltekin, and James Conklin. Com puting supergame equilibria. Econo- metrica, 71(4):1239–1254, 2003. doi: https://doi.org/10.1111/1 468-0262.t01-1-00445. URL https://onlinelibrary.wiley.com/doi/abs/10.1111/1468-0262.t01-1-00445 . 18

    2003 doi

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.