Skip to content
Tutorials Center
My learning

Topics

Lesson 21 of 37 ยท Sequential Games

Subgame-Perfect Equilibrium

Learn subgames and subgame-perfect equilibrium (SPE), why it refines Nash equilibrium, and how to find SPE in the ultimatum game with runnable Python code.

A Nash equilibrium of a game tree can rest on a promise or threat that nobody would carry out. The entry game showed one: the incumbent "fights", the entrant stays out, and nobody gains from changing strategy given the other's plan, yet the plan relies on an action that is irrational if reached. Subgame-perfect equilibrium (SPE) removes such equilibria. It requires players to behave rationally not only along the path of play but in every part of the game that could be reached.

Subgames

A subgame is a part of the tree that:

  • starts at a single decision node,
  • includes everything that follows that node,
  • and does not cut any information set (if a node is in the subgame, so are all nodes it cannot be distinguished from).

The whole game is a subgame of itself. In a game of perfect information, every decision node starts a subgame. With imperfect information, only some nodes do.

A strategy profile is a subgame-perfect equilibrium if it induces a Nash equilibrium in every subgame. For finite games of perfect information, SPE is exactly the set of outcomes of backward induction (Kuhn's theorem, with ties giving extra solutions).

Nash versus subgame-perfect

Nash equilibrium No one gains by deviating
Subgame A tree that starts at one node
SPE Nash in every subgame
Removes empty threats Off-path play must be rational

Every SPE is a Nash equilibrium, not the other way round.

1 / 2

Finding Nash equilibria that are not subgame perfect

The program below lists every pure strategy of each player, finds the Nash equilibria of the normal form, and then tests each one against the backward-induction plan. It reuses the dictionary representation from Extensive-Form Games, and any equilibrium that does not match backward induction at every node is flagged as non-credible.

Python Nash versus subgame-perfect in the entry 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.

Both equilibria appear as Nash, and only the one with "Enter, Accommodate" passes the subgame-perfection test.

The ultimatum game

Player 1 (the proposer) splits a pie of 10 units: offers x to player 2 and keeps 10 - x. Player 2 (the responder) accepts (payoffs 10 - x, x) or rejects (both get 0). Offers are whole numbers.

Backward induction: the responder accepts any x > 0, since 1 beats 0. At x = 0 the responder is indifferent. The proposer, anticipating this, offers the smallest amount the responder accepts. In the simplest solution, the proposer offers 1 and keeps 9.

Python Solving the ultimatum 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.

Notice the tie: at an offer of 0 the responder is indifferent between accepting and rejecting. With the rule "accept when indifferent", the proposer offers 0 and keeps 10. With the rule "reject when indifferent", an offer of 0 is rejected, so the proposer offers 1 and keeps 9. The answer depends on the tie-breaking convention, a reminder that SPE can be a set of equilibria, not a single point.

In real experiments, offers below about 20% of the pie are often rejected, and the typical offer is near 40 to 50%. The theory treats players as maximizers of their own money. People also care about fairness, so rejecting a low offer has a value that the pure money payoff leaves out. This is a modeling choice that behavioral game theory studies, not a failure of backward induction itself.

Many Nash equilibria, one SPE

The ultimatum game has many Nash equilibria. For instance: "the responder accepts only offers of at least 5, and the proposer offers exactly 5". Neither side gains by changing, since a lower offer would be rejected. But the responder's rule "reject offers below 5" is a threat: rejecting a 1 gives 0, accepting gives 1. SPE discards the threat.

Python Nash equilibria of the ultimatum game with threshold rules

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.

Every split from (10, 0) to (0, 10) appears as a Nash outcome with a suitable threshold, including the "fair" 5-5 split, plus a few outcomes where the offer is rejected. Only the extreme offers (0 or 1) survive subgame perfection.

Imperfect information and subgames

If a node is inside an information set with other nodes, it does not start a subgame, because starting there would cut the set. So in a game where the second player cannot see the first move, the whole game is the only subgame and SPE equals Nash. To refine further you need ideas such as beliefs and perfect Bayesian equilibrium, which assign probabilities to nodes in an information set.

Common mistakes

  • Thinking that every Nash equilibrium of a tree is acceptable. Check each one against backward induction.
  • Treating a node inside an information set as a subgame root.
  • Reporting one SPE when ties produce several.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. A strategy profile is subgame perfect if it...

    SPE requires equilibrium behavior in all subgames, including those that are not reached on the equilibrium path.
  2. 2. Which statement is true?

    The whole game is a subgame of itself, so SPE implies Nash; the converse fails when threats are empty.
  3. 3. In the ultimatum game with money payoffs, what does subgame perfection predict about the responder's behavior?

    After a positive offer, accepting gives more than the 0 from rejecting, so a rational responder accepts.
  4. 4. Which node starts a subgame?

    A subgame must start at a node alone in its information set and must not cut any information set.

Summary

  • A subgame is a self-contained piece of the tree starting at one node, without cutting information sets.
  • SPE is a strategy profile that is a Nash equilibrium in every subgame.
  • In finite perfect-information games, SPE coincides with backward induction (ties give extra solutions).
  • Nash equilibria that depend on empty threats, such as the incumbent's fight, are not subgame perfect.
  • The ultimatum game has many Nash equilibria but a sharp SPE prediction; experiments show that fairness matters.

Next: Stackelberg Competition and Commitment