Skip to content
Tutorials Center
My learning

Topics

Lesson 13 of 37 ยท Nash Equilibrium

Multiple Equilibria and Selection

Learn what to do when a game has several Nash equilibria: focal points, payoff dominance, risk dominance and basins of attraction, with Python code for each criterion.

Nash equilibrium is a powerful filter, but it often leaves more than one profile standing. The stag hunt has two, battle of the sexes has two, a game of choosing which side of the road to drive on has two, and large games can have dozens. When the theory says "any of these could happen", how do you know which one will? Selecting among equilibria is a whole research area called equilibrium selection. This lesson shows the main criteria and lets you compute them.

You should be comfortable finding equilibria with Pure Nash Equilibrium and judging outcomes with Pareto Efficiency and Social Dilemmas.

Coordination games

A coordination game is one where players do best when they take the same action (or complementary actions) and the alternatives are worse. The payoffs in a pure coordination game are the same for both players and the diagonal cells pay more than the others. Examples: choosing which side of the road to drive on, agreeing on a file format, picking the same meeting place without being able to talk.

Here the problem is not conflict but ambiguity. Both (Left, Left) and (Right, Right) are equilibria. Theory alone cannot choose, yet real people coordinate easily, which tells us that something outside the payoff matrix is at work.

Criterion 1: Focal points

Thomas Schelling observed that people coordinate on options that stand out for cultural, psychological or logical reasons. Asked to meet a stranger in New York with no instructions, many people say "Grand Central Station at noon." Asked to choose a number, many pick 1. These salient choices are focal points.

Focal points are not derived from payoffs; they come from shared context, labels, history or precedent. A designer can create them: a default option on a form, a posted standard, a traffic rule or a leader's announcement. In code we can only represent them as extra information the players share.

Criterion 2: Payoff dominance

An equilibrium payoff dominates another if every player gets a strictly higher payoff in it. If one equilibrium payoff dominates all others, it is natural to expect players to coordinate on it: it is best for everyone, and everyone knows this. In the stag hunt, (Stag, Stag) payoff dominates (Hare, Hare).

Python Payoff-dominant equilibrium

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

In the stag hunt a payoff-dominant equilibrium exists. In battle of the sexes none does: each equilibrium is better for one player and worse for the other, so payoff dominance is silent.

Criterion 3: Risk dominance

Payoff dominance ignores risk. In the stag hunt, going for the stag is wonderful if the other also goes, but disastrous if the other chases a hare. If you are unsure about the partner, hare is the safe choice. John Harsanyi and Reinhard Selten formalized this with risk dominance.

For a symmetric 2x2 coordination game with payoffs [[a, b], [c, d]] for the row player (first action X, second action Y; a > c and d > b so both diagonal cells are equilibria), compare the cost of a mistaken deviation from each equilibrium:

  • At (X, X), deviating to Y loses a - c.
  • At (Y, Y), deviating to X loses d - b.

The equilibrium whose deviation loss product is larger (here, with symmetry, the larger single loss) is risk dominant: it is the one that is better against a 50-50 belief about what the partner does. Equivalently, X is the best reply whenever the chance that the partner plays X exceeds p* = (d - b) / ((a - c) + (d - b)). If p* < 0.5, then X is risk dominant.

Python Risk dominance in a 2x2 coordination game

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

In the stag hunt with payoffs 4, 1, 3, 3, the deviation loss at (Stag, Stag) is 4 - 3 = 1, and at (Hare, Hare) it is 3 - 1 = 2. The larger loss belongs to the Hare equilibrium, so (Hare, Hare) is risk dominant, and the threshold is 2/3: you should hunt stag only if you believe the partner hunts stag with probability above two thirds. The efficient equilibrium is not the safest one, and which one wins depends on how much people trust each other.

Criterion 4: Learning and basins of attraction

A dynamic view asks how play would evolve if players adjust by trial and error. Start with random behavior; at each step a randomly chosen player switches to a best response to the others' current action. Where does the process end? Each equilibrium has a basin of attraction, the set of starting points that lead to it. Larger basins make an equilibrium more likely in practice. Here is a population version of the stag hunt: a fraction of the population hunts stag and each round every individual compares the expected payoff of both actions against the current population and switches to the better one.

Python Basins of attraction in the stag hunt

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The result shows the basin of attraction for the cooperative equilibrium is only a third of the possible starting points, matching the threshold computed above. If most people start out cautious, the group sinks into the safe equilibrium and stays there. Nothing inside the game can pull it out. This is a coordination trap, and it is why leaders, communication, standards and pre-commitments matter: they move people across the threshold.

flowchart LR
    S0["Few trust the partner<br/>share of stag < 2/3"] --> H["Hare equilibrium"]
    S1["Many trust the partner<br/>share of stag > 2/3"] --> T["Stag equilibrium"]
    H -.->|"communication, standards"| S1

The threshold decides which equilibrium a population falls into.

Criterion 5: Refinements

Game theorists have proposed many refinements of Nash equilibrium that rule out "unreasonable" equilibria: subgame perfection (Chapter 5), trembling-hand perfection, and others. Sequential games show how timing rules out equilibria that rely on empty threats. For simultaneous games, the tools in this lesson are the main ones.

Tip: When you design a system with coordination issues (an API standard, a payment method, a meeting protocol), create a focal point on purpose. A clear default beats a perfect equilibrium nobody can find.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. What is a focal point?

    Focal points are salient options, such as a landmark or an obvious default. They come from shared context, not from the payoffs.
  2. 2. In the stag hunt with payoffs [[4, 1], [3, 3]], which equilibrium is payoff dominant and which is risk dominant?

    Both players earn more at Stag-Stag (4 vs 3). But the deviation loss at Hare-Hare is larger, so Hare-Hare is safer against uncertainty about the partner.
  3. 3. The threshold p* in the stag hunt example is 2/3. What does it mean?

    Below 2/3 the expected payoff of the safe Hare exceeds that of Stag, above it Stag is better.
  4. 4. Why does battle of the sexes have no payoff-dominant equilibrium?

    Payoff dominance needs one equilibrium to be better for everyone. Here each equilibrium favors a different player.

Summary

  • Many games have several Nash equilibria; theory alone may not say which will occur.
  • Focal points come from shared context and let players coordinate without talking.
  • Payoff dominance picks the equilibrium best for everyone; risk dominance picks the safest one against uncertainty.
  • In the stag hunt the efficient equilibrium is not risk dominant, so groups can get trapped in the safe one.
  • Basins of attraction describe how likely each equilibrium is under adaptive play; leaders and standards can move a group across a threshold.

Next: Cournot Duopoly: Nash Equilibrium with Continuous Choices