📈 economics

Sequential Equilibria in a Class of Infinite Extensive Form Games

This paper defines a natural notion of sequential equilibrium for a class of infinite extensive-form games with continuous information, proving that such equilibria exist, refine Nash equilibria, and coincide with traditional sequential equilibria in finite games.

Original authors: Michael Greinecker, Martin Meier, Konrad Podczeck

Published 2026-04-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Michael Greinecker, Martin Meier, Konrad Podczeck

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 you are playing a complex board game with friends. In the classic version of this game (which game theorists call "finite extensive form games"), everyone takes turns, and at every turn, you have a limited list of moves to choose from—like moving a pawn to square A, B, or C.

In these standard games, there is a famous rule for finding the "perfect" way to play, called Sequential Equilibrium. It's like a super-smart referee who says: "If you find yourself in a situation you didn't expect to happen, you should still play the best possible move for that specific moment, assuming everyone else is also playing their best." This rule relies on the idea that you can imagine every possible move happening, even the unlikely ones, to test if your strategy holds up.

The Problem: The Infinite Menu
The trouble starts when the game changes. Imagine instead of choosing between squares A, B, or C, you can choose any number between 0 and 1. You could pick 0.5, or 0.500001, or 0.5000000001. There are infinitely many choices (a "continuum").

In this infinite world, the old referee's rule breaks down. Why? Because to test "every possible move," you would have to assign a tiny bit of probability to every single number between 0 and 1. But mathematically, you can't spread a finite amount of probability (100%) across an infinite number of options so that every single one gets a non-zero share. The old method simply cannot start.

Because of this, economists and game theorists have been stuck. They can't use the "perfect play" rule for real-world scenarios like setting prices (which can be any number) or choosing quantities, so they've had to invent "ad-hoc" (made-up) rules that don't have a solid theoretical foundation.

The Solution: A New Way to Look at the Game
This paper, by Greinecker, Meier, and Podczeck, introduces a new way to define "perfect play" for these infinite games. They don't try to force the old rules to work; instead, they build a new framework based on continuity.

Think of it like this: In the old method, you tried to check every single grain of sand on a beach to see if the beach is stable. In the new method, the authors say, "If the beach is smooth and continuous, we don't need to check every grain. We just need to check a dense, representative sample of grains. If the beach holds up there, it holds up everywhere."

Here is how they do it, using a few key concepts:

The "Strategic Measure" (The Map, Not the Step):
Instead of tracking exactly what a player does at every single moment (which is impossible with infinite choices), they track the overall pattern of play. Imagine a player's strategy not as a list of specific steps, but as a "map" showing the probability of ending up in different parts of the game. They call this a "Strategic Measure." It's like looking at the weather forecast (the overall pattern of rain) rather than tracking every single raindrop.
The "Smooth Signal" Assumption:
The paper assumes that information in the game flows smoothly. If you change your action slightly, the information the other players receive changes slightly, not drastically.
- The Bad Example: Imagine a game where if you pick the number 0.5, your opponent sees "Blue," but if you pick 0.500001, they suddenly see "Red." This "jump" breaks the game.
- The Good Example: If you pick 0.5, they see "Light Blue." If you pick 0.500001, they see "Slightly Lighter Blue." This smoothness allows the new math to work.
The "What-If" Test (Sequential Rationality):
The authors define a "Sequential Equilibrium" as a strategy that is the limit of a series of "almost perfect" strategies.
- Imagine you are testing a strategy by pretending that every possible move has a tiny, non-zero chance of happening (even if it's 0.0000001%).
- You check: "If I were to play this unlikely move, would I still be making the best decision given what I know?"
- If your strategy passes this test for every possible "what-if" scenario (specifically, for every "strategically relevant" set of information), then it is a Sequential Equilibrium.

Why This Matters
The authors prove two big things:

Existence: In this new class of smooth, infinite games, a "perfect play" solution always exists. You don't have to settle for "maybe" or "ad-hoc" rules.
Consistency: If you apply their new definition to a standard, finite game (the kind with just A, B, and C), it gives you the exact same answer as the old, famous definition. It's a generalization, not a replacement.

A Real-World Example: The Noisy Duopoly
To show it works, they apply it to a "Duopoly" (two companies competing).

Scenario: Company A sets a price. Company B sees the price, but only through a "noisy" window (they see a slightly blurry version of the price).
Old Problem: In previous models, if the noise was too specific, the companies might get stuck in a loop where no stable price exists.
New Result: Using their new definition, they show that if the "noise" is smooth enough, a stable equilibrium exists. Furthermore, they prove that the first company (the leader) still has an advantage, even with the noise, provided the noise isn't too wild.

In Summary
This paper solves a long-standing headache in game theory. It says: "We can't check every single infinite number, but if the game is smooth and continuous, we can check the 'shape' of the strategy instead. By doing this, we can find the perfect way to play games where players have infinite choices, just like we do in games with finite choices."

They didn't invent a new game; they just built a better pair of glasses so we can finally see the winning moves in the infinite ones.

1. Problem Statement

The paper addresses a fundamental gap in game theory: the lack of a well-defined Sequential Equilibrium for extensive-form games where players choose from a continuum of actions (e.g., continuous prices or quantities).

The Limitation of Existing Theory: The standard definition of sequential equilibrium (Kreps and Wilson, 1982) relies on the concept of "completely mixed strategies" (trembling hand perfection), where every action at an information set is played with positive probability. In games with uncountable action spaces, it is impossible to assign positive probability to every single action. Consequently, the standard definition fails, forcing researchers to rely on ad-hoc refinements that lack general theoretical backing.
Pathologies in Infinite Games: The authors highlight that without specific continuity assumptions, infinite extensive-form games often suffer from severe pathologies, including:
- Non-existence of Nash equilibria (e.g., Example 1 in the paper).
- Non-existence of sequentially rational equilibria (e.g., Example 2, a signaling game).
- The space of feasible distributions over plays failing to be closed under weak convergence, leading to "strategic entanglement" where future players correlate behavior on vanishingly small variations that disappear in the limit.

2. Methodology and Framework

The authors propose a new framework based on Strategic Measures and Informational Continuity.

A. Informational Continuity Assumption

To resolve the pathologies, the authors impose a strict continuity condition on the signal functions (the mechanism by which players observe past actions).

Assumption: Signal functions are given by an integrably bounded family of conditional densities that vary continuously with respect to past actions.
Implication: This excludes games of perfect information with continuous actions (where a specific action reveals the exact state) but includes games with "noisy observability." This assumption ensures that small changes in actions lead to small changes in the distribution of signals, preventing the "strategic entanglement" that causes non-existence.

B. Strategic Measures

Instead of working directly with behavior strategies (which are difficult to handle topologically in infinite spaces), the authors utilize Strategic Measures.

Definition: A strategic measure is a Borel probability measure on the space of signals and actions for a specific player. It represents the joint distribution of a player's actions and the signals they receive, independent of the specific strategy function used to generate them.
Properties:
- The set of strategic measures is compact, convex, and metrizable in the weak topology.
- Expected payoffs are continuous and multilinear functions of the profile of strategic measures.
- This allows the use of fixed-point theorems (Kakutani–Fan–Glicksberg) to prove existence.

C. Definition of Sequential Equilibrium

The authors redefine sequential equilibrium by adapting the characterization of Myerson and Reny (2020) for finite games.

Strategically Relevant Sets: Since one cannot reach every measurable information set with positive probability in a continuum setting, the authors define a set of information sets as "strategically relevant" if they are reached with positive probability and possess open action-sections (i.e., if an action is in the set, a small neighborhood of that action is also in the set).
The Equilibrium Concept: A profile of strategic measures is a sequential equilibrium if it is the limit of a sequence of profiles where:
1. Every strategically relevant set is reached with positive probability.
2. Players are $\epsilon$ -optimal conditional on reaching these strategically relevant sets.
Key Distinction: Unlike standard definitions, this approach does not require beliefs to be defined at every information set, only at those that are strategically relevant and reachable with positive probability in the approximating sequence.

3. Key Contributions

Existence Theorem: The paper proves that Sequential Equilibria exist in the defined class of infinite extensive-form games. This is a major theoretical breakthrough, as existence was previously unknown or required finitely additive measures (which are non-constructive) in similar contexts.
Refinement of Nash Equilibrium: The proposed sequential equilibrium refines the Nash equilibrium concept by ensuring that strategies are optimal not just on the equilibrium path, but also off-equilibrium at all strategically relevant contingencies.
Consistency with Finite Games: The authors demonstrate that in standard finite extensive-form games, their definition selects the exact same strategy profiles as the traditional Kreps-Wilson sequential equilibrium.
Constructive Solutions: By relying on countably additive measures (standard probability) rather than finitely additive measures, the solutions are "explicitly usable" and avoid the non-constructive nature of previous approaches (e.g., Myerson and Reny's use of nets and finitely additive distributions).

4. Key Results

Theorem 1 (Existence): A sequential equilibrium exists for the class of games satisfying the informational continuity assumption.
Proposition 2 (Compactness): The set of sequential equilibria is compact in the topology of strategic measures.
Application to Duopoly (Proposition 3): The authors apply the concept to a noisy sequential duopoly (extending Maggi, 1999). They show that if the noise in observing the leader's quantity is sufficiently small, a sequential equilibrium exists where the leader retains a first-mover advantage.
- Contrast: In models with perfect observability, the leader has an advantage. In models with fixed support noise (Bagwell, 1995), the advantage often vanishes. The authors show that with continuous noise (varying support), the advantage persists, resolving a tension in the literature.

5. Significance

Theoretical Unification: The paper bridges the gap between finite game theory (where sequential equilibrium is standard) and continuous action games (where it was undefined). It provides a rigorous foundation for analyzing dynamic games with continuous variables, which are ubiquitous in economics (e.g., pricing, investment, R&D).
Robustness of Results: By establishing that sequential equilibria exist and are compact, the paper validates the use of backward induction logic in continuous settings, provided informational continuity holds.
Methodological Innovation: The shift from behavior strategies to strategic measures as the primary object of analysis simplifies the topological arguments required for existence proofs in infinite-dimensional spaces. This approach offers a template for analyzing other complex dynamic games where standard strategy spaces are ill-behaved.
Practical Application: The resolution of the sequential duopoly problem demonstrates that the "first-mover advantage" is robust even under noisy observation, provided the noise structure is continuous, offering new insights for industrial organization and market design.

In summary, Greinecker, Meier, and Podczeck successfully extend the most important refinement of Nash equilibrium to a broad class of infinite games, overcoming the mathematical hurdles of continuous action spaces through the novel use of strategic measures and informational continuity.