Correlated equilibrium¶

A joint distribution over players' action recommendations such that, conditional on each private recommendation, no player gains by unilaterally remapping their action.

Version: v1 · 2026-09-28 · History
Domain-specific #: 8738
Domain group: Social Sciences
Origin domain: Economics & Finance
Subdomain: Game Theory → Economics & Finance

Core Idea¶

A correlated equilibrium is a joint distribution of private action recommendations under which no player can improve conditional expected utility by unilaterally remapping any recommendation. A correlated equilibrium is a joint distribution of privately observed action recommendations such that no player benefits by unilaterally remapping any recommendation while others obey. A mediator is conceptual and need not communicate after play begins. Mixed Nash equilibria are included, but correlation permits additional distributions. Utilities, information, and timing must be fixed; contracts or later communication change the game. No-regret learning can approach the set under specified assumptions.

How would you explain it like I'm…

Follow the Secret Whisper

Imagine a helper who secretly whispers to each player in a game what to do. It's a correlated equilibrium if every player is happy to follow the whisper, as long as everyone else follows theirs too. Nobody could do better by ignoring their own whisper.

Secret Suggestions Everyone Follows

In game theory, a game is any situation where people's choices affect each other. A correlated equilibrium works like a fair helper who picks a plan at random and then secretly tells each player only their own part: 'you do this.' It counts as an equilibrium if, whatever you are told, following the suggestion is your best move as long as the others follow theirs. Because the helper can link everyone's suggestions, players can line up their actions in ways they couldn't if each one just flipped their own coin. The helper doesn't even have to be real — it's a way of describing the pattern — and the players don't talk to each other after getting their suggestions.

Obedience-Stable Joint Recommendations

A correlated equilibrium is a probability distribution over combinations of players' actions, used to give each player a private recommendation. It is an equilibrium if, for every player and every recommendation that player might receive, following the recommendation is a best response given that the others follow theirs. The recommendations can be correlated, so players can coordinate in ways independent random strategies can't. Every Nash equilibrium, including mixed ones, counts as a correlated equilibrium, since independent randomizing is a special case — but correlated equilibria can include additional outcomes. The 'mediator' who draws the recommendations is just a modeling device: players don't have to talk after receiving them, and the test checks every possible way a player might deviate depending on the signal they got.

A correlated equilibrium is a joint probability distribution over action profiles, interpreted as private action recommendations from a conceptual mediator, such that each player's obedience is optimal given that all others obey. Formally, for each player and each recommended action with positive probability, the conditional expected payoff from following it is at least that of any alternative action, conditional on that recommendation. The criterion thus tests all signal-dependent unilateral deviations, meaning every rule of the form 'when told a, play a-prime'. Because signals can be correlated across players, the concept permits coordination unavailable to independent mixed strategies. The set of correlated equilibria contains every distribution induced by a Nash equilibrium and may contain additional outcomes. The mediator is only a device: no post-recommendation communication among players is assumed.

Scope of Application¶

The concept applies in game theory and related work when its identity and evidence are explicit. Use it with game, utilities, signal distribution, player information, and deviation timing explicit; distinguish arbitrary correlation, Nash, coarse correlation, communication, and enforcement.

Game theory. Generalizes equilibrium.
Mechanism design. Uses information and incentives.
Learning in games. Connects to regret.
Auctions/markets. Models correlated recommendations.
Multiagent systems. Coordinates decentralized actions.

Clarity¶

State game, action sets, utilities, signal/recommendation distribution, information observed, and the full conditional deviation inequalities. The closest near miss sets the boundary: Coarse correlated equilibrium is the closest miss because deviations are chosen before seeing recommendations rather than conditioned on them.

Manages Complexity¶

Correlated equilibrium describes a high-dimensional incentive-compatible distribution by linear constraints, expanding coordination possibilities while keeping unilateral rationality testable. The mediator need not be a person and need not reveal the entire signal. Mathematically a joint distribution over action recommendations is a correlated equilibrium when, conditional on each recommendation, no player gains in expectation by applying any unilateral action-remapping rule while others follow theirs. This obedience condition is stronger than checking only one unconditional deviation and naturally handles private recommendations. Every mixed Nash equilibrium induces a correlated equilibrium through independent recommendations, but correlation can support additional distributions and sometimes higher welfare. The concept assumes utilities and information structure are fixed; communication or enforceable contracts after recommendations create different games. Learning procedures such as no-regret dynamics can make empirical play approach the correlated-equilibrium set, but convergence claims require the exact regret notion and observation model. The central coordination–individual autonomy tradeoff is this: Correlation expands joint outcomes while preserving unilateral choice.

Abstract Reasoning¶

Use three linked moves: fix strategic game and information structure; specify the joint recommendation distribution; condition each player on each recommendation. As a collapse test, identity collapses when any player can improve expected utility through an allowed unilateral remapping. A fourth check is to compare obedience with every unilateral remapping.

Knowledge Transfer¶

Obedience-under-shared-signal structure transfers across games, but utilities, signals, and deviation sets must be remapped exactly. No canonical parent prime is currently asserted; broader structural comparisons remain related-prime analogies until separately adjudicated in the DAG. Independent mixed Nash distributions are included.

Relationships to Other Abstractions¶

Current abstraction Correlated equilibrium Domain-specific

Parents (1) — more general patterns this builds on

Correlated equilibrium is a kind of Equilibrium Prime

A correlated equilibrium is a game-theoretic equilibrium satisfying conditional no-profitable-deviation constraints.

Children (1) — more specific cases that build on this

Bayes Correlated Equilibrium Domain-specific is a kind of Correlated equilibrium

Bayes Correlated Equilibrium is a strict kind of Correlated equilibrium: its frozen identity entails the parent's defining structure while adding domain-specific restrictions.

Hierarchy path (1) — routes to 1 parentless root

Correlated equilibrium → Equilibrium → Fixed Point

Neighborhood in Abstraction Space¶

Correlated equilibrium sits in a crowded region of the domain-specific corpus (24^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Strategic Decision Biases & Mechanisms (29 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08