Subgame Perfection

Notes

Quick quiz

A few questions to warm up and check the basics, before the exam-type questions below. They are reshuffled each time, so you can keep retrying.

Example questions

The following are exam-type questions in the style of the examination paper, with marks at the rates used in the papers. A question totalling fewer than 25 marks would, in the examination, be combined with further parts, often one of the examinable proofs, to make a full 25-mark question. Attempt them in full before reading the worked solutions.

Question 1 (based on the in-class activity)

In the Traitors debrief we saw that the Faithful's threat to banish a deviator works only when they can be sure who deviated. Once players sometimes break Vote-Left for fun, a flagged suspect might be the colluding Traitor or merely a Faithful having fun, and the threat can lose its bite. We model this with a two-stage game. Let \(q\) be the Faithful's belief, the probability they assign, that the suspect they are poised to banish is the Traitor rather than someone breaking Vote-Left for fun. The Traitor chooses to comply with Vote-Left or deviate (keep colluding); the Faithful then choose to banish the suspect or carry on. Complying gives payoffs (Traitor, Faithful) of \((1, 2)\), and carrying on after a deviation gives \((3, 0)\). Banishing the suspect is a gamble: it yields \((-5, 3)\) if the suspect really is the Traitor, which happens with probability \(q\), and \((3, -5)\) if the suspect is an innocent Faithful, with probability \(1 - q\). Taking \(q = \tfrac{1}{2}\), the expected payoff to banishing is therefore \(\tfrac{1}{2}(-5, 3) + \tfrac{1}{2}(3, -5) = (-1, -1)\).

(a) Represent the game in extensive form (draw the tree). [3]

(b) Use backward induction to find the subgame perfect equilibrium and its payoffs. [5]

(c) Show that (comply, banish) is a Nash equilibrium of the game. [5]

(d) Explain why (comply, banish) is not subgame perfect, and how this mirrors the Traitors threat to banish a suspect that does not survive contact with the actual decision once the Faithful are unsure who deviated. [6]

(e) More generally, let the suspect be the Traitor with probability \(q\), so that banishing gives expected payoffs \((3 - 8q,\ 8q - 5)\). Determine the range of \(q\) for which the threat deters the Traitor, the range for which it is credible, and hence the range for which (comply, banish) is a Nash equilibrium that is not subgame perfect. [4]

Question 2

(a) Provide definitions for the following terms:

(b) An entrant must decide whether to enter a market or stay out. If the entrant enters, the incumbent must decide whether to fight or accommodate. The payoffs (entrant, incumbent) are: stay out \((0, 2)\); enter then fight \((-1, -1)\); enter then accommodate \((1, 1)\).

(i) Draw the game in extensive form. [2]

(ii) Use backward induction to find the subgame perfect equilibrium and the resulting payoffs. [3]

(iii) Find a Nash equilibrium of the game that is not subgame perfect, and explain why it is not. [5]

(iv) Write the game in normal form and confirm the pure Nash equilibria. [3]

(c) Explain the difference between a Nash equilibrium and a subgame perfect equilibrium. [3]

(d) State the theorem on the existence of a subgame perfect equilibrium in finite games of perfect information. [4]

Question 3

(a) Define sequential rationality. [2]

(b) Player 1 chooses \(L\) or \(R\). After \(L\), player 2 chooses \(\ell\) (payoffs \((3, 1)\)) or \(r\) (payoffs \((0, 0)\)). After \(R\), player 2 chooses \(\ell\) (payoffs \((2, 2)\)) or \(r\) (payoffs \((1, 3)\)).

(i) Use backward induction to find the subgame perfect equilibrium and its payoffs. [3]

(ii) Write the game in normal form. [4]

(iii) Obtain all pure Nash equilibria. [4]

(iv) State which Nash equilibrium is subgame perfect, and for each of the others identify the non-credible threat that makes it fail. [4]

(c) Consider the centipede game: player 1 may take, giving \((2, 0)\), or pass; then player 2 may take, giving \((1, 3)\), or pass; then player 1 may take, giving \((4, 2)\), or pass, giving the leaf \((3, 5)\).

(i) Solve the game by backward induction. [5]

(ii) State the subgame perfect outcome and payoffs, and comment on why the prediction is striking given that both players could do better. [3]

Question 4 (hard)

Two firms compete by choosing quantities, one after the other. Inverse demand is \(P(Q) = a - Q\), where \(Q = q_1 + q_2\) is total output and \(a > 0\); marginal costs are zero, so firm \(i\) earns profit \(q_i P(Q)\). Firm 1 (the leader) chooses \(q_1\) first; firm 2 (the follower) observes \(q_1\) and then chooses \(q_2\).

(a) Explain what a subgame perfect equilibrium is for a game with continuous action sets. [3]

(b) By backward induction, find the follower's best response \(q_2(q_1)\). [6]

(c) Hence find the leader's subgame perfect quantity, the follower's quantity, and the resulting profits. [8]

(d) Compare with the simultaneous-move (Cournot) equilibrium, in which \(q_1 = q_2 = a/3\), and explain the first-mover advantage in terms of commitment: why the leader gains by moving first even though the follower observes the choice. [8]

Marking exercises (optional)

These exercises are entirely optional: they are extra practice, not extra content. The best way to use an AI assistant is to be its examiner, not its student: checking work is exactly the skill the examination rewards, and the skill you will need to use these tools well. Each exercise gives a transcript in the style of a widely used free AI chatbot answering one of the example questions above. Every transcript contains at least one error. Mark it as an examiner would: find each error, explain what goes wrong, and award a mark out of the marks of the original question. A marker's report for each transcript is on the solutions page.

Marking exercise 1 (on Question 2(b)(ii) and (iii), worth [3] and [5])

A student pastes Question 2(b) into a free AI chatbot, which replies:

Backward induction: if the entrant enters, the incumbent compares
fighting (-1) with accommodating (1) and accommodates. Anticipating
this, the entrant compares staying out (0) with entering (1) and
enters. The subgame perfect equilibrium is (enter, accommodate), with
payoffs (1, 1). ✅

A Nash equilibrium that is not subgame perfect: consider (stay out,
accommodate): the entrant stays out, and the incumbent would
accommodate if entered. This is a Nash equilibrium, but it is not
subgame perfect, because the entrant's choice to stay out is not
optimal in every subgame.

Marking exercise 2 (on Question 4(b) and (c), worth [6] and [8])

A student pastes Question 4 into a free AI chatbot, which replies:

Each firm maximises q_i (a - q1 - q2). The first-order conditions are

    a - 2 q1 - q2 = 0,    a - q1 - 2 q2 = 0

Solving simultaneously gives q1 = q2 = a/3, with profits a²/9 each.
Firm 1 moves first, so it simply commits to a/3, and firm 2 follows
with a/3. Moving first makes no difference: both firms play the same
equilibrium either way. ♟️

Optional further reading

You do not need this to follow the topic, but the Repeated Games chapter of the textbook may help if you would like more background on subgame perfect equilibria over a long horizon.

Worked solutions to the example questions →

Download the full quiz as a PDF (all questions, no solutions) →

Facilitator notes: Subgame Perfection