Simple Communication

arXiv:2509.09621 · econ.TH · Submitted 2025-09-11 · 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: "Simple Communication".

Paper summary: Ruth: To wrap up, "Simple Communication" looks at how communication costs when you have to simplify complex advice, and they show that equilibrium scores are usually linear or coarsely linear in the state <ref:2509.09621#pg1>.

Warren: The authors highlight that these equilibrium payoffs are generally lower than what you'd expect if both players were perfectly committed to a score <ref:2509.09621#pg1>.

Iris: They establish that for normally distributed states, the best equilibrium scores are linear and correspond to the ex-ante best and worst linear scores <ref:2509.09621#pg8>.

Ruth: The overall implication is that even with aligned preferences, forcing simple communication creates strategic frictions because it limits the information flow from a two-dimensional state <ref:2509.09621#pg1>.

Warren: They show that while you can have commitment to a score, you might be better off deviating into strategies that aren't scores when the problem is high-dimensional <ref:2509.09621#pg1>.

Iris: The paper points out that this whole structure depends on how the state is distributed and whether it's one-dimensional or two-dimensional <ref:2509.09621#pg8>.

Conclusion: Ruth: So, we’re wrapping up on "Simple Communication." Basically, this paper looks at what happens when an expert has to send advice using simple language instead of all the technical details they know about a complex situation.

Warren: Yeah, it boils down to how that simplicity creates friction even when both sides are trying to be nice and share the same goals. The core idea is that just because you want the same payoff doesn't mean communicating simply is as good as communicating perfectly.

Iris: From an econometric standpoint, what they nailed here is characterizing those equilibrium scores. They show that any score that actually works in this setup has to be either linear or a coarse version of a linear score. That’s pretty restrictive.

Ruth: Right, so the payoff structure itself forces the strategy to be simple in some way if it's going to hold up under strategic pressure. But the big point isn't just that scores are linear; it's what that linearity means for the information being shared.

Warren: It means you’re losing information compared to just saying everything, which is a real cost when you’re trying to maximize your expected outcome. That difference between what they can actually do and what they *should* be able to do is where the strategic cost lives.

Iris: And that two-dimensional nature of the state—having two different things that matter—is what makes it interesting. If it were just one thing, you could actually get the perfect outcome with a linear score.

Ruth: Exactly, because in a one-dimensional world, full revelation is possible and optimal. But because this is two dimensions, there are ways for players to deviate from those simple scores that aren't possible in the simpler case.

Warren: So what they’re suggesting is that you can get better results by being clever about *how* you communicate your state, not just by revealing it all. That’s a different kind of strategy than just picking the best possible score.

Iris: It sets up a big question for us: how do we model this in the real world when the state is actually high-dimensional? That leads right into what happens when things aren't normal distributions, which we’ll be looking at next.

Oslo Metropolitan University · University of Oslo

econ.TH

Submitted: 2025-09-11

Updated: 2026-11-17

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 81/100

The gist: We study multidimensional cheap talk with simple language and aligned preferences An expert communicates with a decision-maker using a score that aggregates a multidimensional state into a…

Key concepts

Two-Dimensional State ($ heta$)
This represents the complex information an expert knows, defined by two variables $ heta = ( heta_1, heta_2)$. The expert observes this state with full support and finite variance. This state dictates the underlying reality that needs to be communicated.
Equilibrium Score ($s$)
A score is a measurable function that represents the sender's strategy. It must satisfy an Intermediate Value Property (IVP), meaning it prevents injective mappings from the state space to the message space. This structure ensures that even with simple messages, the sender cannot perfectly reveal the full state.
Strategic Frictions
These are barriers to effective communication arising from using simple language instead of revealing every detail. Although players have aligned preferences, this simplicity creates frictions. Linearity in equilibrium implies linear dependence between actions, leading to lower equilibrium payoffs compared to full commitment.

Terminology

Summary

We study multidimensional cheap talk with simple language and aligned preferences An expert communicates with a decision-maker using a score that aggregates a multidimensional state into a one-dimensional message Even though the expert and the decision-maker share the same payoffs, the use of simple language introduces strategic frictions. p. 0

Where the paper says it

An expert communicates with a decision-maker using a score that aggregates a multidimensional state into a one-dimensional message. Even though the expert and the decision-maker share the same payoffs, the use of simple language introduces strategic frictions.

Model Setup

The model involves two players, a sender and a receiver, who share identical payoff functions based on their actions and a two-dimensional state of the world. The sender observes a two-dimensional state θ = (θ1, θ2) with full support and finite variance. p. 5

Where the paper says it

There are two players: a sender and a receiver. The sender has private information about a two-dimensional state of the world, θ = (θ1, θ2) ∈ Θ = R 2 , whose distribution has full support and admits a density function f.

The receiver takes two actions represented by a = (a1, a2) ∈ R squared. p. 5
Where the paper says it

When there is risk of ambiguity, we write ˜θ for the random variable and θ for its realization. The receiver takes two actions represented by a = (a1, a2) ∈ R 2 .

p. 5
Where the paper says it

When there is risk of ambiguity, we write ˜θ for the random variable and θ for its realization. The receiver takes two actions represented by a = (a1, a2) ∈ R 2 .

Payoffs are defined as u(a, θ):= −ϕ(a1 − θ1)2 − (a2 − θ2)2 with ϕ > 0. p. 5
Where the paper says it

There are two players: a sender and a receiver. The sender has private information about a two-dimensional state of the world, θ = (θ1, θ2) ∈ Θ = R 2 , whose distribution has full support and admits a density function f.

Equilibrium Score Characterization

A score s is defined as a non-constant measurable function s: Θ → R that satisfies the Intermediate Value Property (IVP). An equilibrium score requires that for all θ, m, m′ ∈ s(Θ), s(θ) = m ⇒ −ϕ(α1(m)−θ1)2−(α2(m)−θ2)2 ≥ −ϕ(α1(m′)−θ1)2−(α2(m′)−θ2)2. p. 6

Where the paper says it

Definition 1. A score s is an equilibrium score if there is a receiver’s strategy α such that • for all θ ∈ Θ, m, m′ ∈ s(Θ), s(θ) = m ⇒ −ϕ(α1(m)−θ1) 2−(α2(m)−θ2) 2 ≥ −ϕ(α1(m′ )−θ1) 2−(α2(m′ )−θ2) 2 , • for all m ∈ s(Θ), α(m) = E[ ˜θ|s(θ) = m].

The main result is that any equilibrium score must be either linear in the state or a discrete coarsening of a linear score up to a relabeling of messages. p. 2
Where the paper says it

Our main result is a characterization of equilibrium scores, that is, scores that constitute the sender’s strategy in a Perfect Bayesian Equilibrium. For any prior distribution, equilibrium scores exist (Proposition 1) and must be either linear in the state or a discrete coarsening of a linear score, up to a relabeling of messages (Proposition 2).

Strategic Frictions and Payoffs

Communicating via linear scores or their coarsening conveys less information than revealing the state. The key barrier to effective communication is strategic frictions, despite the players’ aligned preferences. p. 2

Where the paper says it

Moreover, there are scores that yield higher payoffs for both the sender and receiver, yet these scores are not equilibrium strategies. The key barrier to effective communication is strategic frictions, despite the players’ aligned preferences.

A consequence of linearity is that in equilibrium the two actions are linearly dependent (Proposition 3 and Proposition 4) which in particular implies a monotone relationship between them. p. 2
Where the paper says it

In Section 3, we illustrate the tension between informativeness and equilibrium behavior in a simplified setting. A consequence of the linearity of scores is that in equilibrium the two actions are linearly dependent (Proposition 3 and Proposition 4), which in particular implies a monotone relationship between them.

Equilibrium payoffs are lower than those achievable under commitment to a score. p. 0
Where the paper says it

Under quadratic-loss utility, any equilibrium score must be linear in the state or a coarsening of a linear score. As a result, equilibrium payoffs are lower than those achievable under commitment to a score.

Best Equilibrium Scores under Normal Distribution

When the state is normally distributed, the best equilibrium score is linear and coincides with the ex-ante optimal linear score. This best equilibrium linear score is an eigenvector of ΦΣ, which corresponds to the ex-ante best and worst linear scores. p. 0

Where the paper says it

As a result, equilibrium payoffs are lower than those achievable under commitment to a score. Finally, for normally distributed states, we characterize the equilibrium linear scores and show they correspond to the ex-ante best and worst linear scores.

The payoffs associated with these scores are given by v+:= −ϕσ21 − σ22 + p(ϕσ21 − σ22)2 + 4ϕρ2σ21σ22 / 3, and v−:= −ϕσ21 − σ22 - p(ϕσ21 − σ22)2 + 4ϕρ2σ21σ22 / 3.

Conclusion on Multidimensionality

The multidimensionality of the model plays a central role because if the state were one-dimensional, a best equilibrium score would yield the commitment payoffs. In a one-dimensional setting, if M = R, full revelation is possible and the best equilibrium yields the commitment payoffs. p. 2

Where the paper says it

A score that yields high payoffs ex-ante might therefore not be an equilibrium score if the sender has a profitable deviation in some states. In Section 3, we illustrate the tension between informativeness and equilibrium behavior in a simplified setting.

The two-dimensional case allows for profitable deviations to strategies that are not scores, which is not possible in the one-dimensional setting where any profitable deviation must be a score. p. 2
Where the paper says it

A score that yields high payoffs ex-ante might therefore not be an equilibrium score if the sender has a profitable deviation in some states. In Section 3, we illustrate the tension between informativeness and equilibrium behavior in a simplified setting.

References

Aybas, Y. C. and Turkel, E. p. 4

Where the paper says it

When considering the commitment problem, we impose a restriction directly on the shape of the information structure, by limiting the sender to select among scores. In this way we are closer to Le Treust and Tomala (2019) and Aybas and Turkel (2024), who consider exogenous constraints on the capacity or cardinality of the message space.

(2024), ‘Persuasion with coarse communication’, arXiv preprint arXiv:1910. p. 0
Where the paper says it

We also thank Katinka Holtsmark for early discussions on this project. 1 arXiv:2509.

13547

Ball, I. (2025), ‘Scoring strategic agents’, American Economic Journal: Microeconomics 17(1), 97–129

Battaglini, M. (2002), ‘Multiple referrals and multidimensional cheap talk’, Econometrica 70(4), 1379–1401

Bloedel, A. W. and Segal, I. p. 4

Where the paper says it

, communication with commitment, where the amount of information transmitted is limited. In Gentzkow and Kamenica (2014), the limitation comes from the cost of designing the experiment, while in Bloedel and Segal (2021) it comes from the information-processing cost faced by the receiver.

R. (2021), ‘Persuasion with rational inattention’, Available at SSRN 3164033

Blume, A. (2018), ‘Failure of common knowledge of language in common-interest communication games’, Games and Economic Behavior 109, 132–155

Blume, A. and Board, O. p. 3

Where the paper says it

They characterize the best equilibria and study their stability. In Blume and Board 2This literature, like us, looks at the consequences of language limitations, not its causes.

(2013), ‘Language barriers’, Econometrica 81(2), 781–812

Bonatti, A. and Cisternas, G. p. 4

Where the paper says it

3Other papers consider scores that aggregate a multidimensional variable in different settings. For example, Ball (2025) and Bonatti and Cisternas (2020) study linear scores where sender and receiver have different payoffs and the sender can manipulate the score input.

(2020), ‘Consumer scores and price discrimination’, The Review of Economic Studies 87(2), 750–791

Chakraborty, A. and Harbaugh, R. p. 4

Where the paper says it

This literature has shown that multiple dimensions can be useful for information revelation, e. , Battaglini (2002), Chakraborty and Harbaugh (2007) and Chakraborty and Harbaugh (2010).

p. 4
Where the paper says it

This literature has shown that multiple dimensions can be useful for information revelation, e. , Battaglini (2002), Chakraborty and Harbaugh (2007) and Chakraborty and Harbaugh (2010).

(2007), ‘Comparative cheap talk’, Journal of Economic Theory 132(1), 70–94

Chakraborty, A. and Harbaugh, R. (2010), ‘Persuasion by cheap talk’, American Economic Review 100(5), 2361–2382

Gentzkow, M. and Kamenica, E. p. 4

Where the paper says it

, communication with commitment, where the amount of information transmitted is limited. In Gentzkow and Kamenica (2014), the limitation comes from the cost of designing the experiment, while in Bloedel and Segal (2021) it comes from the information-processing cost faced by the receiver.

(2014), ‘Costly persuasion’, American Economic Review 104(5), 457–462

Hurwicz, L. (1972), ‘On informationally decentralized systems’, Decision and organization: A volume in Honor of J. p. 6

Where the paper says it

a full technical report. Second, a single message may need to serve multiple audiences, as when a public health organization gives public advice that cannot be tailored to individual circumstances.

Marschak

Jager, G., Metzger, L. P. and Riedel, F. (2011), ‘Voronoi languages: Equilibria in cheap-talk games with high-dimensional types and few signals’, Games and Economic Behavior 73(2), 517–537

Kallenberg, O. (2021), Foundations of modern probability, Probability Theory and Stochastic Modelling, 99, third edition. p. 8

Where the paper says it

The analysis in this section is self-contained. Abusing notation slightly, let Θ = {0, 1} 2 and f(θ) denote the probability of state θ ∈ {0, 1} 2 .

edn, Springer, Cham, Switzerland

Lang, R. (1986), ‘A note on the measurability of convex sets’, Archiv der Mathematik 47, 90–92

Le Treust, M. and Tomala, T. p. 4

Where the paper says it

When considering the commitment problem, we impose a restriction directly on the shape of the information structure, by limiting the sender to select among scores. In this way we are closer to Le Treust and Tomala (2019) and Aybas and Turkel (2024), who consider exogenous constraints on the capacity or cardinality of the message space.

(2019), ‘Persuasion with limited communication capacity’, Journal of Economic Theory 184, 104940

Levy, G. and Razin, R. p. 4

Where the paper says it

, Battaglini (2002), Chakraborty and Harbaugh (2007) and Chakraborty and Harbaugh (2010). In this strand of the literature, the contribution closest to ours is Levy and Razin (2007), who show that correlation across dimensions can limit communication by creating informational spillovers across dimensions.

(2007), ‘On the limits of communication in multidimensional cheap talk: a comment’, Econometrica 75(3), 885–893

Lipman, B. L. (2003), Language and economics, in ‘Cognitive Processes and Economic Behaviour’, Routledge, pp. p. 3

Where the paper says it

In Blume and Board 2This literature, like us, looks at the consequences of language limitations, not its causes. On the latter topic see Lipman (2003) and Lipman (2025).

75–93

Lipman, B. L. (2025), ‘Why is language vague? p. 3

Where the paper says it

In Blume and Board 2This literature, like us, looks at the consequences of language limitations, not its causes. On the latter topic see Lipman (2003) and Lipman (2025).

’, International Journal of Game Theory 54(1), 8

Mount, K. and Reiter, S. p. 7

Where the paper says it

This literature imposes continuity-like assumptions on mechanisms to avoid bijections between sets of different dimensions. See for example, Hurwicz (1972), Mount and Reiter (1974) and Reichelstein (1984).

(1974), ‘The informational size of message spaces’, Journal of Economic Theory 8(2), 161–192

Parlett, B. N. (1998), The symmetric eigenvalue problem, SIAM

Reichelstein, S. (1984), ‘Incentive compatibility and informational requirements’, Journal of Economic Theory 34(1), 32–51

West, D. B. (2001), Introduction to graph theory, Vol. p. 1

Where the paper says it

Give me a one-handed economist. Harry Truman 1 Introduction Decision-makers often seek advice on complex matters.

2, Prentice hall Upper Saddle River

--- Page 1 ---

Simple Communication

Jacopo Bizzotto Nathan Hancart

--- Page 2 ---

Give me a one-handed economist.

Harry Truman

--- Page 3 ---

Introduction

Decision-makers often seek advice on complex matters. In principle, experts can write volumes on the potential impacts of a new policy, and dietitians can craft personalized eating plans. p. 1

Where the paper says it

Policymakers consult experts about the impact of new policies; individuals seek professional guidance across diverse domains, including nutritional practices, exercise regimens and financial investments. In principle, experts can write volumes on the potential impacts of a new policy, and dietitians can craft personalized eating plans.

In practice, however, advice must be simple. p. 1
Where the paper says it

In principle, experts can write volumes on the potential impacts of a new policy, and dietitians can craft personalized eating plans. In practice, however, advice must be simple.

Most people rely on general dietary guidelines, such as the “5 a day” rule, or on standard nutrition labels that cannot be tailored to their specific nutritional needs. p. 1
Where the paper says it

Experts present their findings in short executive summaries. Most people rely on general dietary guidelines, such as the “5 a day” rule, or on standard nutrition labels that cannot be tailored to their specific nutritional needs.

In this study, we examine the strategic incentives that arise when advice on complex issues must be simple. p. 1
Where the paper says it

Most people rely on general dietary guidelines, such as the “5 a day” rule, or on standard nutrition labels that cannot be tailored to their specific nutritional needs. In this study, we examine the strategic incentives that arise when advice on complex issues must be simple.

--- Page 4 ---

To isolate the effect of simple language, we focus on settings in which the expert and the decision-maker share identical preferences. If the expert could use a language as rich as the object described, revealing every relevant aspect of the object under consideration would be both optimal and an equilibrium strategy. p. 1

Where the paper says it

To isolate the effect of simple language, we focus on settings in which the expert and the decision-maker share identical preferences. If the expert could use a language as rich as the object described, revealing every relevant aspect of the object under consideration would be both optimal and an equilibrium strategy.

When instead experts communicate using simple language, the nature of communication is not immediately clear. p. 1
Where the paper says it

If the expert could use a language as rich as the object described, revealing every relevant aspect of the object under consideration would be both optimal and an equilibrium strategy. When instead experts communicate using simple language, the nature of communication is not immediately clear.

--- Page 5 ---

We explore communication via simple language in a multidimensional cheap talk game. A sender observes a two-dimensional state of the world and sends a cheap-talk message to a decision-maker. p. 1

Where the paper says it

We explore communication via simple language in a multidimensional cheap talk game. A sender observes a two-dimensional state of the world and sends a cheap-talk message to a decision-maker.

The receiver takes a two-dimensional action to minimize a quadratic loss function. p. 1
Where the paper says it

A sender observes a two-dimensional state of the world and sends a cheap-talk message to a decision-maker. The receiver takes a two-dimensional action to minimize a quadratic loss function.

Sender and receiver share the same payoffs. p. 1
Where the paper says it

The receiver takes a two-dimensional action to minimize a quadratic loss function. Sender and receiver share the same payoffs.

The model is also equivalent to one with two receivers, each with a one-dimensional action. p. 1
Where the paper says it

Sender and receiver share the same payoffs. The model is also equivalent to one with two receivers, each with a one-dimensional action.

--- Page 7 ---

We view simplicity as a notion that must hold in equilibrium. In our interpretation, the sender is an expert who, once the state is realized, is free to send any message. p. 2

Where the paper says it

1 We view simplicity as a notion that must hold in equilibrium. In our interpretation, the sender is an expert who, once the state is realized, is free to send any message.

Our main result is a characterization of equilibrium scores, that is, scores that constitute the sender’s strategy in a Perfect Bayesian Equilibrium. p. 2
Where the paper says it

In our interpretation, the sender is an expert who, once the state is realized, is free to send any message. Our main result is a characterization of equilibrium scores, that is, scores that constitute the sender’s strategy in a Perfect Bayesian Equilibrium.

--- Page 8 ---

2.1 Setting

There are two players: a sender and a receiver. The sender has private information about a two-dimensional state of the world, θ = (θ1, θ2) ∈ Θ = R squared, whose distribution has full support and admits a density function f. p. 5

Where the paper says it

There are two players: a sender and a receiver. The sender has private information about a two-dimensional state of the world, θ = (θ1, θ2) ∈ Θ = R 2 , whose distribution has full support and admits a density function f.

Without loss of generality, we assume that E[θ] = 0. p. 5
Where the paper says it

The sender has private information about a two-dimensional state of the world, θ = (θ1, θ2) ∈ Θ = R 2 , whose distribution has full support and admits a density function f. 4 Without loss of generality, we assume that E[θ] = 0.

We also assume that the variance of θ is finite. p. 5
Where the paper says it

4 Without loss of generality, we assume that E[θ] = 0. We also assume that the variance of θ is finite.

The receiver takes two actions represented by a = (a1, a2) ∈ R squared.

--- Page 9 ---

Payoffs. Sender and receiver share the payoff function u(a, θ):= −ϕ(a1 − θ1)2 − (a2 − θ2)2. p. 5

Where the paper says it

Before the receiver acts, the sender sends a cheap-talk message m ∈ M for some measurable space M ⊇ R. Sender and receiver share the payoff function u(a, θ) := −ϕ(a1 − θ1) 2 − (a2 − θ2) 2 , with ϕ > 0.

Both players want each action to match the corresponding component of the state. p. 5
Where the paper says it

Sender and receiver share the payoff function u(a, θ) := −ϕ(a1 − θ1) 2 − (a2 − θ2) 2 , with ϕ > 0. Both players want each action to match the corresponding component of the state.

The parameter ϕ governs the relative cost of mismatch along each dimension. p. 5
Where the paper says it

Both players want each action to match the corresponding component of the state. The parameter ϕ governs the relative cost of mismatch along each dimension.

--- Page 10 ---

Equilibrium Selection

A score s is a non-constant measurable function s: Θ → R that satisfies the following property:5 Intermediate Value Property (IVP): for any θ, θ′ ∈ Θ such that s(θ) > s(θ′), any m ∈ [s(θ′), s(Θ)] ∩ s(Θ) and any continuous function γ: [0, 1] → Θ with γ(0) = θ′ and γ(1) = θ, there is a t ∈ [0, 1] such that s(γ(t)) = m. We denote the collection of all scores by S. p. 5

Where the paper says it

A score s is a non-constant measurable function s : Θ → R that satisfies the following property:5 Intermediate Value Property (IVP): for any θ, θ′ ∈ Θ such that s(θ) > s(θ ′ ), any m ∈ [s(θ ′ ), s(θ)] ∩ s(Θ) and any continuous function γ : [0, 1] → Θ with γ(0) = θ ′ and γ(1) = θ, there is a t ∈ [0, 1] such that s(γ(t)) = m. We denote the collection of all scores by S.

--- Page 13 ---

2.2 Discussion of Scores

Scores can be continuous or discrete mappings from R squared to R. Scores prevent injective mappings. p. 7

Where the paper says it

2 Discussion of Scores Scores capture several desirable properties of simple communication: they impose natural constraints, they are flexible, and they prevent injective mappings. Scores can be continuous or discrete mappings from R 2 to R.

By preventing injections from Θ into the message space, scores prevent the sender from fully revealing the state. p. 7
Where the paper says it

Scores prevent injective mappings. By preventing injections from Θ into the message space, scores prevent the sender from fully revealing the state.

--- Page 14 ---

  1. Illustration in a 2x2 Setting

We call a score optimal if it maximizes the payoffs over all scores, subject to the receiver best-replying (BR). We now show that commitment has value, i. p. 9

Where the paper says it

α(m) = E[θ|m], ∀m ∈ s(Θ). (BR) We now show that commitment has value, i.

e., it can be the case that none of the optimal scores are equilibrium strategies. p. 9
Where the paper says it

(BR) We now show that commitment has value, i. , it can be the case that none of the optimal scores are equilibrium strategies.

--- Page 23 ---

  1. Analysis

We first establish that an equilibrium score exists. We first establish existence of an equilibrium score by showing that there always exists a PBE with two messages in the support of the sender’s strategy. p. 4

Where the paper says it

As a consequence, they find that ex-ante optimal strategies are equilibrium strategies. Similarly, Lipman (2025) uses the fact that optimal strategies are equilibrium strategies to show that there is always an equilibrium in pure strategy when preferences are aligned in cheap-talk games with a possibly constrained set of messages.

Since any strategy of the sender with two messages trivially satisfies the properties of a score, an equilibrium score exists. p. 2
Where the paper says it

Simply requiring one-dimensional messages would not, however, capture a meaningful notion of simple language: since injections from R 2 to R exist, the sender could fully reveal the state. To prevent injections while imposing natural restrictions on communication, we additionally require that the sender’s equilibrium strategy satisfies a weak form of continuity.

--- Page 24 ---

Proof

If s(Θ) ≤ 2, then s cannot be injective. Otherwise, take some messages m, m1, m2 ∈ s(Θ) with m1 < m < m2, and states θ, θ1, θ2 such that s(θ) = m, s(θ1) = m1 and s(θ2) = m2. p. 7

Where the paper says it

Examples of scores include: • s(θ) = β1θ1 + β2θ2; • s(θ) =    1 if β1θ1 + β2θ2 ≥ c, 0 otherwise. ; • s(θ) = p (θ1 − c1) 2 + (θ2 − c2) 2 .

--- Page 30 ---

Proof of Remark 3

We now construct a sequence of scores indexed by ε such that the sequence of associated payoffs converges to v∗ = 0. As an intermediate step, we build a sequence of correspondences. p. 2

Where the paper says it

To prevent injections while imposing natural restrictions on communication, we additionally require that the sender’s equilibrium strategy satisfies a weak form of continuity. Specifically, we focus on mappings from the state space to the real numbers that satisfy a property we dub the Intermediate Value Property (IVP).

Figure 4 illustrates a correspondence in the sequence for K = 3. p. 8
Where the paper says it

Viewing the square {0, 1} 2 as a simple graph where each point is a vertex and the edges of the graph are the edges of the square, we can rewrite the IVP as follows: for any θ, θ′ ∈ Θ with s(θ) > s(θ ′ ), any m ∈ [s(θ ′ ), s(θ)] ∩ s(Θ) and any path p from θ ′ to θ, there is some θ ′′ on the path such that s(θ ′′) = m. 7 Figure 1 illustrates four scores in this environment.

The score sϵ as defined is indeed a score since for any fixed θk, sϵ(θ−k, θk) increases by at most one increment as θ−k increases. p. 3
Where the paper says it

The optimality of each score depends on the correlation structure: when the dimensions are positively correlated, the ex-ante best linear score correlates the actions and vice versa. Accordingly, an increase in the correlation is good for welfare in the best equilibrium but also increases the cost of coordinating on the wrong equilibrium.

--- Page 31 ---

Proof of Proposition 2

The proof of Proposition 2. All equilibrium scores are equivalent to linear or coarsely linear scores. p. 0

Where the paper says it

As a result, equilibrium payoffs are lower than those achievable under commitment to a score. Finally, for normally distributed states, we characterize the equilibrium linear scores and show they correspond to the ex-ante best and worst linear scores.

The payoffs from any score β with receiver strategy α(m) = β1/β2 1+ϕβ2/β2 1+ϕβ2/β2 1+ϕβ2/β2 1+ϕβ2/θ, where m ∈ R. p. 6
Where the paper says it

6 We call an equilibrium score a best equilibrium score if it maximizes the players’ payoffs among all equilibrium scores. We select equilibria in which the sender’s strategy is a score.

The payoffs from any score β with receiver strategy α(m) = β1/β2 1+ϕβ2/θ, where m ∈ R. p. 6
Where the paper says it

6 We call an equilibrium score a best equilibrium score if it maximizes the players’ payoffs among all equilibrium scores. We select equilibria in which the sender’s strategy is a score.

--- Page 35 ---

Proof of Proposition 4

The proof of Proposition 4 follows the same lines as the proof of Proposition 3 replacing the sets s−1β(m) with the sets of types indifferent between m and m − 1.

--- Page 38 ---

Proof of Proposition 5

For any strategy s(θ) = β'θ, we have the unconditional distribution over messages m induced by the score s, m ∼ N(0, σ2s) where σ2s = β21σ21 +β22σ22 + 2β1β2ρσ1σ2 = β'Σβ. The payoff of the sender can be rewritten, up to a constant, as −a'Φa + 4a'Φθ.

--- Page 39 ---

G Value of Ignorance

If the sender does not observe θ2, then score s1 is an equilibrium strategy. We show next that any equilibrium score in the standard setting is associated with a larger loss than s1. Similar arguments apply for any score such that s(θ) = s(θ′) for any two high-probability states. p. 8

Where the paper says it

Abusing notation slightly, let Θ = {0, 1} 2 and f(θ) denote the probability of state θ ∈ {0, 1} 2 . Our definition of scores does not apply directly to this finite environment but it can be adapted by requiring that the score does not “jump” messages when moving through points that differ by at most one coordinate.

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. Strategic Friction Modeling for Simple Communication: The AI system can model communication where an expert uses a score that aggregates a multidimensional state into a one-dimensional message, specifically showing how strategic frictions arise even when experts and decision-makers share identical payoffs under quadratic-loss utility.

  2. Equilibrium Score Characterization: The system can determine that in equilibrium, any equilibrium score must be either linear in the state or a discrete coarsening of a linear score, which provides a structural constraint on the communication strategy space.

  3. Optimal Linear Score Identification: For normally distributed states, the system can identify the ex-ante best and worst linear scores as the best and worst equilibrium scores, allowing for pre-emptive selection of an optimal simple message based on state distribution parameters like correlation and variance.

  4. Welfare Comparison via Covariance Structure: The AI system can calculate a specific loss function derived from the covariance structure of state dimensions to quantify the difference between equilibrium payoffs and commitment payoffs, indicating how increases in correlation affect welfare.

  5. Information Revelation Limits: The system can characterize conditions under which revealing only one dimension is an equilibrium score, demonstrating that this is only possible when the two dimensions are uncorrelated, thus providing a constraint on how much information can be conveyed simply.

Abstract

We study simple communication in a multidimensional cheap talk model with aligned preferences. An expert communicates with a decision-maker using a score, i.e., an aggregator of a multidimensional state into a one-dimensional message. Even though the expert and the decision-maker share the same payoffs, equilibrium payoffs are lower than those achievable under commitment to a score. Under quadratic-loss utility, any equilibrium score must be linear in the state or a coarsening of a linear score. Finally, for normally distributed states, we characterize the equilibrium linear scores and show they include the ex-ante best and worst linear scores.

Sources

Related papers