Sequential Rationality and Subgame Perfection

This Version: Mar 27, 2025

0/15
Example 1: Entry Game

Two Nash equilibria in pure strategies: (I , A) and (O, P).

Do you think (O, P) is “plausible”?
I What if Player 2’s information set is reached for some reason (e.g., Player 1 may
make an unintended move)? In that case, A will be dominated by P.
I In the Nash equilibrium (O, P), (Player 1 believes that) Player 2 fails to play a
"rational" strategy if Player 2’s information set is actually reached.

1/15
Sequential Rationality

Sequential rationality: Players must demonstrate rationality whenever they are

called upon to make decisions.
I An optimal strategy for a player should maximize their expected payoff, con-
ditional on every information set at which they must move.
I That is, player i’s strategy should specify an optimal action at each of player
i’s information sets, even information sets that player i does not expect to be
reached in the game.

2/15
Backward Induction

Backward Induction: Identify an optimal action for each info set by working
backward in game trees. The simplest way to represent sequential rationality.
I Start from the decision nodes that are immediate predecessors of terminal
nodes. At this stage, determining the optimal action does not require consid-
ering the behavior of other players. Any “dominated" action can be eliminated.
I Move to nodes whose immediate successors are either terminal nodes or nodes
that have already been evaluated. With the optimal actions identified in the
previous step, the payoff consequences of each action become clear. Any
“dominated" action can be eliminated.
I Repeat this process until you reach the initial node.

Backward induction identifies a Nash equilibrium in which all players’ strategies

are sequentially rational.
3/15
Backward Induction

4/15
Subgame

Given an extensive-form game, a node x in the tree is said to initiate a subgame

if neither x nor any of its successors belong to an information set that contains
nodes that are not successors of x. A subgame is the tree structure defined by
such a node x and its successors.

More succinctly, a subgame must satisfy all Tree Rules 1–5. Importantly, any
subgame must have a single initial node.

Any subgame must satisfy the following properties:

I It contains a single initial node;
I It includes all successors of this initial node (including terminal nodes);
I If any node from an information set appears in the subgame, all nodes from
that information set must be included (preserving the same information struc-
ture as the original extensive-form game).
5/15
Subgame

Question: How many subgames are there in this game tree?

6/15
Subgame Perfect Nash Equilibrium

Subgame Perfect Nash Equilibrium (SPE): A strategy profile is called a

subgame perfect Nash equilibrium (SPE) if it specifies a Nash equilibrium in
every subgame of the original game.

SPE ≈ NE + sequential rationality.

For any extensive-form game, the set of SPEs is a subset of the set of NEs (i.e.,
every SPE is a NE). Why? In this sense, SPE is often considered a refinement
of NE.

For extensive-form games with perfect information, the backward induction pro-
cedure identifies an SPE.

7/15
Subgame Perfect Nash Equilibrium: Example 1

There are three pure-strategy Nash equilibria: (UA, X ), (DA, Y ), and (DB, Y ).

There is only one SPE: (UA, X ).

The SPE concept requires that all players follow NE strategies in all subgames,
regardless of whether a given equilibrium actually reaches the subgame in ques-
tion. The NEs (DA, Y ) and (DB, Y ) fail to meet this requirement.

8/15
Subgame Perfect Nash Equilibrium: Example 2

Consider the following variant of the Cournot game, called Stackelberg duopoly.
I Firm 1 first selects a quantity q1 ∈ [0, 12], which is observed by Firm 2.
I Then, Firm 2 selects its quantity q2 ∈ [0, 12].
I The payoff functions: ui (qi , q−i ) = (12 − qi − q−i )qi for both i = 1, 2.
We can solve for the SPE using backward induction.
I Fix an arbitrary number x ∈ [0, 12] and consider the subgame that follows
Player 1 choosing q1 = x.
I Note that there are infinitely many such subgames.
12−x
I Show that, in each subgame, it is optimal for Player 2 to choose q2 = 2 .
I Given this, Player 1 must choose q1 that maximizes
maxq1 ∈[0,12] 12 − q1 − 12−q 1


2 q1 .

9/15
Subgame Perfect Nash Equilibrium: Example 2

The SPE strategy profile:

s1 = 6 and s2 (q1 ) = (12 − q1 )/2 for all q1 ∈ [0, 12].

Note that Player 2’s equilibrium strategy is a function that specifies the optimal
action at every one of Firm 2’s information sets.

The SPE outcome: (q1∗ , q2∗ ) = (6, 3) w/ the equilibrium market price given by
12 − q1∗ − q2∗ = 3.

The SPE payoffs: u1 (6, 3) = 18 and u2 (6, 3) = 9.

First-mover advantage

10/15
Subgame Perfect Nash Equilibrium: Example 3 (Ultimatum Game)

Player i makes a take-it-or-leave-it offer. If player j rejects player i’s offer, the
game ends with ui = uj = 0.

Player i can commit to the initial offer and never revise it.

11/15
Subgame Perfect Nash Equilibrium: Example 3 (Ultimatum Game)

SPE:
I Player i (the proposer) chooses m = 0.
I In response to any m ∈ [0, 1], player j (the responder) always chooses Y .
I That is, si∗ = 0 and sj∗ (m) = Y for any m ∈ [0, 1].

Is there any other SPE?

I If player i offers m > 0, player j will certainly accept (Y ).
I If player i offers m = 0, player j is indifferent between choosing Y and N. However,
in SPE, player j MUST choose Y in response to m = 0; otherwise, player i could
slightly increase m to ensure player j accepts.

12/15
Subgame Perfect Nash Equilibrium: Example 4 (Two-Period Bargaining Game)

Period 1:
I Player 1 (P1) makes an offer m1 ∈ [0, 1]. After observing the offer, Player 2 (P2)
decides whether to accept or reject it.
I If P1’s offer is accepted, the game ends with P1 receiving 1 − m1 and P2 obtaining
m1 . If P2 rejects the offer, the interaction continues into the second period.

Period 2:
I P2 makes a counteroffer m2 ∈ [0, 1], which P1 either accepts or rejects.
I If P1 accepts P2’s offer, P1 receives δ1 m2 , and P2 receives δ2 (1 − m2 ) as their final
payoffs. If P1 rejects P2’s offer, the game ends, and both players receive 0.

δ1 ∈ (0, 1) and δ2 ∈ (0, 1): Each player’s discount factor.

13/15
Subgame Perfect Nash Equilibrium: Example 4 (Two-Period Bargaining Game)

Suppose that Player 2 rejects Player 1’s offer in the first period.
I Then, the game played in the second period is simply the ultimatum game.
I In equilibrium, Player 2 will offer m2 = 0, and Player 1 will accept this offer.

In the first period, Player 2 will accept any m1 ≥ δ2 and reject any m1 < δ2 .
Given this, Player 1 will offer m1 = δ2 .

In SPE: u1 = 1 − δ2 and u2 = δ2 . Patience =⇒ bargaining power.

14/15
Discussion

Game-theoretical analysis of bargaining is useful but has some limitations:

I For example, in all games discussed here, fairness concerns are disregarded. In
reality, we often reject even highly lucrative offers if we perceive them as “unfair.”
I Experimental evidence: In the ultimatum game (played in the lab), proposers typ-
ically offer about 40% of the pie to the other player. Moreover, offers are rejected
about 15-20% of the time.

15/15