Lecture 18: Practice: Strategic Moves and Sequential

Games of Imperfect Information

Vera Sharunova

1. Mover Advantages and Strategic Moves

(Problem 6S2 and 8S10 from Dixit, Skeath, McAdams)

A Rebel Force of guerilla fighters seeks to inflict damage on a Conventional Army of the
government, while the Conventional Army would like to destroy the Rebel Force. The
two sides play a game in which each must decide whether to locate their forces in the
Hills or in the Valley. The Rebels can inflict the most damage from the Valley, but if
the Conventional Army is also in the Valley, it will be able to force the Rebels into open
combat, where they are most vulnerable. By contrast, the Rebel Force is safer in the
Hills, although it is also less capable of inflicting damage from that location. The payoff
matrix for this game is shown below.

Conventional Army
Valley Hills
Valley 1, 4 4, 1
Rebel Force
Hills 3, 2 2, 3

(a) Is this a zero-sum game? Explain your answer.

(b) Identify all pure-strategy Nash equilibria of this game (or explain why no such
equilibrium exists) in the case when the Rebel Force and Conventional Army move si-
multaneously.

(c) Draw the game tree for the case when the Rebel Force moves first. What is the
SPNE of this game?

(d) Draw the game tree for the case when the Conventional Army moves first. What is
the SPNE of this game?

(e) Does the order of moves matter in this game? If so, does the game have a first-mover
advantage, a second-mover advantage, or do both players prefer the same ordering of

1
ECON3308.01: Game Theory in Economics Summer 2021

moves?

(f) In the 1961 military handbook Guerilla Warfare, Marxist revolutionary Che Guevara
wrote, “The fundamental characteristic of a guerilla band is mobility.” Discuss how the
mobility, and therefore the nimbleness, of guerilla forces might affect the possible ways
in which moves could be ordered in this game. In particular, consider whether all the
move orders analyzed in parts (b), (c), and (d) would actually be possible with a very
mobile Rebel Force.

(g) Now assume that the high mobility of the Rebel Force makes it impossible for the
Conventional Army to move last. Is it possible for the Rebel Force to achieve an out-
come that is better than the SPNE outcome by credibly declaring a strategic move in
the pregame? If not, why not? And if so, what sort of strategic move would it use, and
how would it phrase its declaration?

2. A Sequential Game of Imperfect Information

(0, 2)
A (−6, −6)
E
P1
C P2
B F (−1, 1)
P1
E (1, −1)
D P2

F
(−3, −3)

Consider the game tree above:

(a) How many subgames does this game have?

(b) Represent the proper subgame of this game in normal form.

(c) Find all subgame-perfect Nash equilibria of this game.

3. Information and Nuclear Safety

(Problem 2 from PS8 of Open Yale Course in Game Theory)

2
ECON3308.01: Game Theory in Economics Summer 2021

Fire (−1, −4)

A Fire (−4, −1)

1
2
Wait R
N
Fire (−4, −1) Wait (0, 0)
1
2
R Fire (−1, −4)

Wait A

Wait (0, 0)

Consider the following game involving two real players and a chance move by Nature
(N). America (A) and Russia (R) have the nuclear capability to destroy each other. Na-
ture tosses a fair coin so that with probability 12 America moves first and Russia moves
second, and with probability 12 Russia moves first and America moves second. For now,
assume that both countries observe Nature’s choice so they know whether they are first
or second. The country who moves first decides whether to fire its missiles or to wait. If
it fires, the game ends. The country who fired gets a payoff of -1, and the other country
gets -4. If the first country waits, then the second country gets to move. It too must
decide to fire or to wait. If it fires, then the game ends; it gets -1 and the other country
gets -4. If it waits, then both countries get 0. Assume that each country maximizes its
expected payoff.

Treat this as one game, rather than two different games. The extensive form of this
game is given above. The first payoff refers to America and the second payoff refers
to Russia. Nature is a fictitious player that represents external uncertainty and has no
payoffs.

(a) What makes this a game of perfect information? Write down the definition of a
strategy in an extensive form game, and identify the possible strategies for America and
for Russia in this game.

(b) Find all pure-strategy subgame-perfect Nash equilibria (SPNE). Are there any Nash
equilibria which are not SPNE?

(c) Now suppose that neither Russia nor America observes the move by Nature, or each
other’s move. That is, should a country be called upon to move, it does not know
whether it is the first mover or whether it is the second mover and the other country
chose to wait. Again, treat this as one game. Draw a game tree for this new game.
Indicate clearly which nodes are in the same information sets.

3
ECON3308.01: Game Theory in Economics Summer 2021

(d) Identify the possible strategies for America and for Russia in the game from part
(c). Find and explain carefully two ‘symmetric’ pure-strategy SPNE in this game that
have very different outcomes.

(e) Now suppose that America can observe the move by Nature and also (when it is the
second mover) Russia’s move. Russia knows what America can observe, but, as before,
Russia can observe neither Nature’s nor America’s move. Draw the game tree for this
game. Argue whether you think the world is a safer place or a more dangerous place
now that America is better informed than Russia. That is, compare the SPNE of this
game with the SPNE of the game of parts (b) and (d).

4
ECON3308.01: Game Theory in Economics Summer 2021

Solutions

1. Mover Advantages and Strategic Moves

(a) Yes, this is a zero-sum game. The opponents’ payoffs in each cell sum up to five, so
we can renormalize the payoffs to sum up to zero.

(b) Pure-strategy Nash equilibria do not exist in zero-sum games.

(c) The game tree for the case when Rebels move first is shown below. In the SPNE, as
last-mover, Army will go wherever Rebels do; anticipating that as first-mover, Rebels
compare the outcomes (Valley, Valley) and (Hills, Hills) and choose Hills leading to an
equilibrium outcome of (Hills, Hills). The SPNE strategies are (Hills, Valley Hills), giv-
ing Rebels a payoff of 2 and Army a payoff of 3.

(d) The game tree for the case when Army moves first is shown below. In the SPNE,
as last-mover, Rebels will go wherever Army does not; anticipating that as first-mover,
Army compares the outcomes (Valley, Hills) and (Hills, Valley) and chooses Valley lead-
ing to a rollback equilibrium outcome of (Valley, Hills). The SPNE strategies are (Valley,
Hills Valley), giving Army a payoff of 2 and Rebels a payoff of 3.

(e) Yes, the order of moves matters. Moreover, both players prefer to move last, each
getting a payoff 3 when moving last versus a payoff 2 when moving first; this game has
a second-mover advantage.

(f) Mobility or nimbleness allows the Rebel Force to change its position on a moment’s
notice, compared to the Army which takes much more time in making any move. In
other words, the Army’s decision where to go becomes irreversible before the Rebel’s
decision becomes irreversible. Recall that, by definition, a player is the “first mover” if
its decision is irreversible and observable to the other player before the other player’s

5
ECON3308.01: Game Theory in Economics Summer 2021

move is irreversible. The fact that Army’s move must be made first (from a chronological
point of view) therefore means that Army cannot be the last mover (from a strategic
point of view). This leaves two possibilities. First, if Army’s decision is observable to
Rebel ahead of time (e.g., any movement by Army creates a cloud of dust), then Army
is the first mover as in part (d). On the other hand, if Army’s movements are unob-
servable until it is too late for the Rebels to escape (e.g., Army troops fly in on attack
helicopters), then the game has simultaneous moves as in part (b).

(g) No, there is no way for the Rebels to get a better outcome by credibly declaring a
strategic move. Why not? Note that the SPNE outcome gives the Rebels their second-
best possible outcome (payoff 3 for Rebels). The only better outcome for the Rebels
is (Hills, Valley) for (1, 4); however, this outcome gives the Army their worst possible
outcome (payoff 1 for Army) and no player can ever be induced to choose their worst
possible outcome. To see this point even more explicitly, suppose that the Rebels were to
commit to a strategic move with the intention of inducing the outcome (Hills, Valleys).
The Rebels’ declaration would have to include the phrase “If you choose Hills, we will
choose Valley,” as well as another phrase of the form “And if you choose Valley, we will
choose X,” where “X” is either “Hills” or “Valley.” Such a strategic move forces Army
to choose between the outcomes (Valley, X) or (Hills, Valley). However, no matter what
“X” may be, Army is going to prefer (Valley, X) over (Hills, Valley), because (Hills,
Valley) gives Army a payoff of 1 while (Valley, X) gives Army a payoff of 4 (if X =
Valley) or 2 (if X = Hills).

6
ECON3308.01: Game Theory in Economics Summer 2021

2. A Sequential Game of Imperfect Information

(a) The game has two subgames.

(b) The normal form is represented in the figure below. The fact that P2 cannot observe
P1’s move and cannot choose separate actions for the two decision nodes implies that
P1 and P2 play a simultaneous game.

(0, 2)
A

P1 P2
E F
B C -6, -6 -1, 1+
P1
D 1, -1+ -3, -3

(c) Use backward induction and the definition of SPNE to find the SPNE of this game.
At the terminal subgame, there are two NE – (D, E) and (C, F). Using the outcomes
associated with these NE as continuation payoffs, we find that the root subgame becomes:

(0, 2) (0, 2)
A A

P1 P1

B B
(1, −1) (−1, 1)

So there are two SPNE in this game – (BD, E) and (AC, F).

7
ECON3308.01: Game Theory in Economics Summer 2021

3. Information and Nuclear Safety

(a) All information sets are singleton, so this game is a game of perfect information. A
strategy in an extensive-form game is a complete plan of actions for each information
set. Both America and Russia have four pure strategies that describe their actions in
each decision node they get to move in: Fire Fire (FF), Fire Wait (FW), Wait Fire
(WF), Wait Wait (WW).

(b) Use backward induction to find that there is a unique SPNE – (WW, WW). See the
figure below for best responses.

Fire (−1, −4)

A Fire (−4, −1)

1
2
Wait R
N
Fire (−4, −1) Wait (0, 0)
1
2
R Fire (−1, −4)

Wait A

Wait (0, 0)

In order to see if there are NE that are not subgame-perfect, we need to convert this
game into normal form and find all pure Nash equilibria. See the figure below.

R
Fire Wait
Fire -1, -4+ -1, -4
A
1 Wait -4, -1 0, 0+
2

N R
1 Fire Wait
2 Fire -4, -1+ -1, -4
A
Wait -4, -1 0, 0+

With probability 12 America moves first and leads us to the normal-form representation of
the top subgame. With probability 12 Russia moves first and leads us to the normal-form
representation of the bottom subgame. Since America’s payoffs are the first number in
each outcome, I put it as the row player in both cases. In both subgames, both Amer-
ica and Russia move at a single decision node, therefore their pure strategies are their
actions. We keep Nature’s move as is because the players will observe Nature move and

8
ECON3308.01: Game Theory in Economics Summer 2021

know with certainty which of the two games they are in.

Each of the subgames has two pure Nash equilibria (F, F) and (W, W). Recall that
Nash equilibrium does not require players to play best responses in subgames that are
not visited on the equilibrium path. So as long as players play a Nash equilibrium in
the subgame we assume Nature puts them in, it does not matter what strategies players
choose in subgames off the equilibrium path. For example, assume that Nature picks
America to move first, then there are 4 Nash equilibria associated with each of the two
Nash equilibria from the normal form:

1. (FF, FF): leading to the (F, F) equilibrium outcome in the top subgame and to
the (F, F) outcome (also happens to be an equilibrium) in the bottom subgame.

2. (FW, FF): leading to the (F, F) equilibrium outcome in the top subgame and to
the (W, F) outcome in the bottom subgame. (W, F) is not an equilibrium, but it
does not matter what happens in the bottom subgame if America moves first.

3. (FF, FW): leading to the (F, F) equilibrium outcome in the top subgame and to
the (F, W) outcome in the bottom subgame.

4. (FW, FW): leading to the (F, F) equilibrium outcome in the top subgame and to
the (W, W) outcome (also happens to be an equilibrium) in the bottom subgame.

Similarly for the other three Nash equilibria we found in the normal form representation
of the game, so we have a total of 12 additional NE:

• From (W, W) in the top subgame: (WF, WF), (WW, WF), (WF, WW),
(WW, WW).

• From (F, F) in the bottom subgame: (FF, FF), (FF, WF), (WF, FF), (WF, WF).

• From (W, W) in the bottom subgame: (FW, FW), (FW, WW), (WW, FW),
(WW, WW).

Eliminating repeating equilibria, we get the final list of 12 NE: (FF, FF), (FW, FF),
(FF, FW), (FW, FW), (WF, WF), (WW, WF), (WF, WW), (WW, WW), (FF, WF),
(WF, FF), (FW, WW), (WW, FW).

9
ECON3308.01: Game Theory in Economics Summer 2021

(c) The information sets are drawn as dashed lines in the figure below. They connect
America’s (Russia’s) decision nodes when it is the first and the second player.

Fire (−1, −4)

A Fire (−4, −1)

1
2
Wait R
N
Fire (−4, −1) Wait (0, 0)
1
2
R Fire (−1, −4)

Wait A

Wait (0, 0)

(d) Since a player cannot pick different actions for different nodes in the same information
set, America and Russia have only two pure strategies – Fire and Wait. This corresponds
to the two countries picking the actions simultaneously. See the normal form below.

R
Fire Wait
Fire -2.5, -2.5+ -1, -4
A
Wait -4, -1 0, 0+

Note that if both countries decide to fire we need to weigh each payoff by the proba-
bility that a given country will be the first mover, so the expected payoffs of America
and Russia are given by 12 ·Payoff if first+ 12 ·Payoff if second = 12 ·(−1)+ 21 ·(−4) = −2.5.

There are two NE in this game (F, F) and (W, W). Both are SPNE because they con-
stitute NE in each subgame (the game itself). They are ‘symmetric’ in a sense that
both countries pick the same action. If either of the countries decides to fire, the other
country’s best response is to fire as well. If either of the countries decides to wait, the
other country’s best response is to wait as well. What we have here is a coordination
game, where, unfortunately, miscoordination is possible.

(e) Now America can tell which decision node it is at, unlike Russia. See the exten-
sive form below. Let’s use backward induction in order to find the SPNE. In the last
decision node of America, it is a best response to Wait (0 > −1). Russia knows that
if Nature picks America to be the first mover, with probability 12 it will be in its top
decision node. Russia also knows that if Nature picks Russia to be the first mover, with
probability 12 it will be in its bottom decision node. Knowing that America will choose
to wait in its last decision node, Russia has to decide whether to Fire and get a payoff

10
ECON3308.01: Game Theory in Economics Summer 2021

of 12 · (−4) + 21 · (−4) = −4 or Wait and get a payoff of 12 · 0 + 12 · 0 = 0. So, Russia’s

best response is to Wait. Intuitively, Russia knows that if it gets to move as a second
mover, it must be the case that America chose to Wait. Also, if Russia moves as a first
mover, it has to choose ‘Wait’, knowing that America will respond by waiting. Knowing
that Russia will choose to Wait, America chooses to Wait as well in its first decision
node. Therefore, the only SPNE in this game is (WW, W). The world is as safe as it
was relative to (a), but is much safer relative to case (d), since there is now a single
SPNE leading to the good outcome.

Fire (−1, −4)

A Fire (−4, −1)

1
2
Wait R
N
Fire (−4, −1) Wait (0, 0)
1
2
R Fire (−1, −4)

Wait A

Wait (0, 0)

11