📈 economics

Information Design and Full Implementation in Nonatomic Games

This paper establishes that in symmetric nonatomic games with negative externalities and strictly concave potentials, every Bayes correlated equilibrium outcome can be fully implemented via a direct information structure, while weakly concave potentials ensure uniform expected total payoffs across all equilibria and allow for approximate implementation of all outcomes.

Original authors: Frederic Koessler, Marco Scarsini, Tristan Tomala

Published 2026-02-25

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Frederic Koessler, Marco Scarsini, Tristan Tomala

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 massive city with millions of commuters. Every morning, they have to choose between two routes to get to work: the Highway or the Scenic Route.

If everyone takes the Highway, it's a parking lot (congestion).
If everyone takes the Scenic Route, it's also crowded.
Ideally, we want exactly half the people on each route so traffic flows smoothly.

The problem? Everyone is selfish. They only care about their own commute time. If they see the Highway is empty, they all rush there, clogging it up. This is a classic "traffic jam" scenario, but in economics, it's called a Congestion Game.

Now, imagine a Traffic Controller (the "Designer") who knows everything: the weather, road construction, and exactly how many people are on the road. This controller wants to minimize total traffic for the whole city, not just for one person.

This paper asks: How can this Controller give advice to millions of people so that they all end up making the perfect choice, even if they are acting in their own self-interest?

The Core Idea: The "Magic Recommendation App"

The authors propose a system where the Controller doesn't just broadcast a public announcement (like "Everyone, take the Scenic Route!"). Instead, they use a Direct Information Structure.

Think of this as a Magic App installed on every commuter's phone.

The App knows the current state of the world (e.g., "It's raining").
It sends a private, personalized message to each person: "You, take the Highway."
Crucially, the App is designed so that if you follow the advice, you are actually better off than if you ignored it and did something else.

The paper proves that if the game has certain "nice" mathematical properties (specifically, if the total happiness of the group goes up when people spread out nicely), this Magic App can force everyone to follow the advice, no matter how they think.

The Two Big Discoveries

The paper splits its findings into two scenarios, like two different types of traffic systems:

1. The "Strictly Perfect" System (Strictly Concave Potential)

Imagine a system where the more people crowd a road, the dramatically worse it gets for everyone. There is a single, clear "sweet spot" for traffic.

The Result: The Magic App can achieve Full Implementation.
What that means: There is only one way the game will play out. Every single person, acting in their own self-interest, will follow the App's private recommendation. There are no "bad" equilibria where everyone ignores the advice and gets stuck in traffic. The Controller gets exactly the perfect outcome they wanted, every single time.
Analogy: It's like a conductor leading an orchestra. Because the music (the math) is structured perfectly, every musician must play the right note to make the song sound good. There is no room for improvisation that ruins the song.

2. The "Good Enough" System (Weakly Concave Potential)

Now imagine a system where the traffic gets worse as people crowd, but not dramatically worse. There might be a few different ways to arrange the traffic that are all "okay," but not necessarily the perfect one.

The Result: The Controller might not get the exact perfect traffic flow, but they can guarantee the Total Payoff (the total happiness of the city) is the same as the best possible outcome.
What that means: Even if people choose slightly different routes than the App suggested, the total time saved by the city is identical to the best-case scenario.
Analogy: Think of a potluck dinner. The Chef (Controller) wants a specific mix of dishes. Even if guests bring slightly different dishes than requested, as long as the total variety and quality of the meal are the same, the party is a success. The specific arrangement doesn't matter as much as the final result.

Why This Matters: The "Revelation Principle" for Crowds

In the past, economists thought you needed complex, indirect ways to influence large groups. You might have to set up a complicated voting system or a strange market mechanism.

This paper shows that for large crowds (nonatomic games), you can use Direct Recommendations.

The Revelation Principle: This is a fancy way of saying, "If you can get people to do what you want through a complicated scheme, you can also get them to do it by just telling them what to do privately."
The paper proves that for traffic jams, Cournot competition (firms deciding how much to produce), and similar games, a simple "Do X" or "Do Y" private message is enough to solve the problem.

Real-World Examples Used in the Paper

Traffic & Platforms: Imagine a social media platform. If too many people join, it gets slow (congestion). If too few join, it's boring (network effects). The platform wants to recommend who joins to maximize engagement. The paper shows how the platform can privately nudge users to join at the perfect rate.
Cournot Competition (Firms): Imagine two companies deciding how many widgets to make. If they both make too many, prices crash. The paper shows how a regulator (or a market mechanism) can give private signals to these firms to ensure they produce the exact amount that maximizes total profit, even if the firms are trying to outsmart each other.

The "Approximation" Safety Net

What if the math isn't perfect? What if the number of people is huge but not infinite, or the numbers are messy?
The authors show that even then, the system works almost perfectly. You can get within a tiny fraction (epsilon) of the perfect outcome. It's like saying, "We might not hit the bullseye exactly, but we'll hit the target so close that nobody can tell the difference."

Summary

This paper is about taming the chaos of crowds.

It tells us that if a system has a natural tendency to balance itself (like traffic or markets), a smart designer doesn't need to force people with laws or complex rules. Instead, they can simply use private, personalized advice.

If the system is "strictly" balanced: The advice will be followed perfectly by everyone, leading to a unique, perfect outcome.
If the system is "loosely" balanced: The advice might be interpreted differently by different people, but the total result for society will still be the best possible one.

It's a reassuring message for anyone trying to design systems for large groups: Simple, private nudges can be just as powerful as complex, public mandates.

1. Problem Statement

The paper addresses the problem of Information Design (or mechanism design with information) in symmetric nonatomic Bayesian games.

Context: In games with a continuum of players (nonatomic), traditional equilibrium concepts like Bayes Nash Equilibrium (BNE) are often computationally intractable. The authors utilize Bayesian Wardrop Equilibrium (BWE) and Bayes Correlated Wardrop Equilibrium (BCWE) as the natural solution concepts.
The Challenge: A designer (e.g., a platform or regulator) wishes to design an information structure (signals) to correlate players' actions and achieve a specific outcome (e.g., maximizing total welfare).
- Partial Implementation: Can a desired BCWE outcome be induced by some equilibrium of the extended game?
- Full Implementation: Can a desired BCWE outcome be induced by all equilibria of the extended game? This is crucial for robustness against adversarial equilibrium selection.
Goal: To determine under what conditions a designer can fully implement a desired BCWE outcome using direct information structures (where signals are identical to actions) and obedient strategies (players follow recommendations).

2. Methodology and Model

The authors employ a framework combining potential game theory, convex analysis, and information design.

The Basic Game ( $\Gamma$ ): A symmetric nonatomic game with a finite set of actions $A$ , a finite set of states $\Theta$ , and a probability distribution $p$ over states. Players are distributed as a unit mass.
Flow Profiles: The state of the system is described by a flow profile $y \in \Delta(A)$ , representing the distribution of player mass across actions.
Information Structures: The authors consider structures where the population is partitioned into finite sub-populations, and all players within a sub-population receive the same signal.
- Direct Information Structure: The set of signals equals the set of actions ( $T^k = A$ ).
- Obedient Strategy: Players follow the signal received (e.g., if recommended action $a$ , they choose $a$ ).
Equilibrium Concepts:
- BCWE: A state-dependent distribution over flows satisfying incentive constraints (no player has an incentive to deviate from the recommended flow distribution).
- BWE: An equilibrium of the game extended with an information structure, where players maximize expected utility given their signals.
Potential Games: A central tool in the analysis is the potential function $\Phi_\theta$ $Φ_{θ}$ . A game is a potential game if the marginal change in a player's payoff equals the marginal change in a global potential function.
- Strictly Concave Potential: Implies a unique equilibrium flow profile.
- Weakly Concave Potential: Implies a convex set of equilibria, but all equilibria yield the same payoff profile.

3. Key Contributions and Results

A. Partial Implementation (Revelation Principle for Nonatomic Games)

Theorem 4.1 establishes a version of the Revelation Principle for symmetric nonatomic Bayesian games.

Result: Any BCWE with finite support and rational flow distributions can be partially implemented as an obedient BWE using a direct information structure.
Approximation: For BCWEs with infinite support or irrational flows, they can be approximately implemented (within $\epsilon$ ) as a Bayesian Wardrop $\epsilon$ -equilibrium.
Significance: This confirms that the set of BCWEs characterizes all outcomes achievable by some information structure, justifying the use of BCWE as the solution concept for information design.

B. Full Implementation in Strictly Concave Potential Games

Theorem 5.5 is the paper's main contribution regarding robustness.

Condition: The basic game admits a strictly concave potential in every state (e.g., congestion games with strictly decreasing payoffs, Cournot games with binary entry/exit).
Result: For any BCWE with finite support and rational flows, there exists a direct information structure such that the BCWE is fully implemented.
- Specifically, the obedient outcome is the unique BWE of the extended game.
- Consequently, all equilibria of the extended game yield the same outcome, eliminating the problem of equilibrium selection.
Mechanism: The strict concavity of the potential function ensures that the potential maximizer is unique. Since BWEs correspond to potential maximizers in the auxiliary game, the equilibrium is unique.

C. Full Payoff Implementation in Weakly Concave Potential Games

Theorem 5.9 addresses cases where the potential is only weakly concave (e.g., standard Cournot games with continuous quantities).

Result: While the flow profile (outcome) may not be unique (multiple equilibria may exist), the expected total payoff is unique across all equilibria.
Implication: The designer can fully implement the expected total payoff of any desired BCWE. Even if players coordinate on a different flow distribution than intended, the aggregate welfare remains the same.
Application: This applies to Cournot games where inverse demand is strictly decreasing.

D. Optimality of Deterministic Flows

Proposition 6.1 identifies a class of games (binary actions, convex costs/concave payoffs) where the designer can restrict attention to deterministic BCWEs (where a single flow profile is chosen per state) without loss of optimality. This simplifies the optimization problem for the designer.

4. Significance and Applications

Theoretical Advancement: The paper bridges the gap between finite-player information design (where full implementation often requires complex, indirect mechanisms with infinite type spaces) and nonatomic games. It demonstrates that in nonatomic settings, simple direct information structures are sufficient for full implementation under natural conditions (concave potentials).
Robustness: The results provide strong guarantees for designers. In strictly concave potential games, the designer does not need to worry about which equilibrium players will select; the mechanism works regardless.
Practical Applications:
- Congestion Games: Traffic routing, network access, and platform design where user density affects utility. The paper shows how to design signals to achieve socially optimal traffic flows or platform participation levels.
- Cournot Oligopolies: Markets with uncertain costs or demand. The results suggest how regulators or market makers can design information to maximize total industry profit or welfare, even when firms have private information.
Approximation: The paper acknowledges that real-world flows might be irrational or infinite. The $\epsilon$ -implementation results ensure that the theoretical findings are robust and applicable to practical scenarios where exact rationality or finite support cannot be guaranteed.

5. Conclusion

Koessler, Scarsini, and Tomala demonstrate that nonatomic games with concave potentials possess a unique structural property that allows for full implementation of desired outcomes via simple direct information structures. This contrasts sharply with finite-player games, where full implementation is often difficult or impossible without complex mechanisms. The paper establishes that for a wide class of economically relevant games (congestion, Cournot), information design can be both effective and robust.