← The collection

Correlated Equilibrium.

Follow or ignore a private recommendation in a collision-avoidance payoff game.

Interactive experimentintuitiveField note ·
INTERACTIVE EXPERIMENT / P41

Correlated Equilibrium

Follow or ignore a private recommendation in a collision-avoidance payoff game.

PLAY & EXPLAIN

Two players choose Go or Wait. Go/Go gives −5 each; Go/Wait gives 3 and 0; Wait/Wait gives 1 each. A public device privately recommends opposite actions with equal probability, and the other player obeys.

Your policy
Recommendation and action history
Your recommendationOther actionYour actionYour payoff
Rounds
0
Your total payoff
0
Expected payoff under this policy
1.5
Collision probability if both independently go half the time
25%

Given a Go recommendation, obeying yields 3 instead of 1. Given Wait, obeying yields 0 instead of −5. Those conditional incentives support this correlated equilibrium; the device does not improve every game.

The partner obeys and all payoffs are stipulated. This distribution need not improve another game, and the activity is not traffic-control guidance.

THE SHORT VERSION

Correlated Equilibrium, explained.

A correlated equilibrium is a joint recommendation distribution under which no player benefits by deviating from their recommendation given the information it supplies.

01 / THE MECHANISM

Why it happens

A device equally recommends Go/Wait or Wait/Go in a declared two-action game. Against an obedient partner, following yields expected payoff 1.5; always Go yields −1 and always Wait 0.5. Conditional obedience inequalities are shown.

A correlated equilibrium is a joint recommendation distribution under which no player benefits by deviating from their recommendation given the information it supplies.

Read the result

Use conditional incentives after each recommendation, not only the unconditional average payoff. The partner’s obedience is a supplied part of this game.

02 / FOLLOW IT THROUGH

A worked example

Compare following a device with ignoring it

  1. Play recommendations while following the device and inspect the Go/Wait histories.

  2. Compare expected payoff 1.5 with always-Go payoff minus one and always-Wait payoff one-half.

  3. Check that deviating after either recommendation lowers payoff against the implied partner action.

OPTIONAL DEEPER DETAILGo deeper: inside the model

Inside this model

A device equally recommends Go/Wait or Wait/Go in a declared two-action game. Against an obedient partner, following yields expected payoff 1.5; always Go yields −1 and always Wait 0.5. Conditional obedience inequalities are shown.

03 / BEYOND THE EXPERIMENT

Where this idea is useful

A practical use

Ask how a shared coordination device changes conditional incentives rather than simply adding another independent random choice.

CHECK YOUR INTUITION

A common misconception

THE TEMPTING CONCLUSION

“Correlated equilibrium is just independent random play by both players.”

THE MORE USEFUL DISTINCTION

The device supplies a joint distribution and information through recommendations. Here opposite recommendations avoid the collisions caused by independent half-probability Go choices.

What this explanation leaves out

  • The partner obeys and all payoffs are stipulated. This distribution need not improve another game, and the activity is not traffic-control guidance.
ONE MORE QUESTION

Does this device solve every coordination problem?

No. The incentive inequalities depend on the declared game and information; another payoff matrix can invalidate obedience.

TAKE THE IDEA WITH YOU

What does your recommendation reveal about the other participant’s likely action?

Associated thinkers

Associated withRobert Aumann ↗

Associations marked provisional are awaiting source review.

Further reading

Explore the original research or the teaching reference behind this experiment.