PSCI 282/482: Making Public Policy

Part 1: Analytical Foundations

Utility, Strategic-Form Games, and Nash Equilibrium
Lecture 2

Prof. Sergio Montero

University of Rochester

Fall 2024
(please do not circulate)
Today

1. Strategic-form games

2. Nash equilibrium

1 / 14
Recap: Games in Strategic Form
I Most of the time, external circumstances relevant for
decision-making are determined by others’ choices
• In other words, these are situations of strategic interdependence:
outcome depends on behavior of multiple decision-makers
I Game theory is the mathematical study of such situations
I We begin by focusing on games in strategic form, which have
three components:
(i) A set of players N = {1, 2, . . . , n}
(ii) For each player i ∈ N , a set of strategies Si
(iii) For each i ∈ N , a utility function ui over potential outcomes:
I i’s utility function assigns a number ui (si ; s−i ) to each choice
si ∈ Si available to i given any potential profile of choices
s−i = (s1 , . . . , si−1 , si+1 , . . . , sn ) by other players,
where each sj ∈ Sj
2 / 14
Date Night
I Player 1 likes the movies, player 2 likes the theater,
but they would both rather be together than alone:
Player 2
Movies Theater
Movies 4,3 1,1
Player 1
Theater 0,0 3,4

I Payoff matrices conveniently summarize two-player games:

• Each row (column) corresponds to an available strategy for the
row (column) player
• The first (second) number listed in each entry is the row
(column) player’s utility from the corresponding strategy profile
(i.e., combination of row and column strategy choices)
• E.g., u1 (Movies; Movies) = 4 and u2 (Theater; Theater) = 4
3 / 14
A Comment on Payoff Notation
I So far, we have used ui (a; s) to denote i’s utility from choosing
alternative a ∈ A given external circumstances s ∈ S
I In the case of games, i chooses a strategy si ∈ Si , and external
circumstances are determined by other players’ strategy choices
s−i = (s1 , . . . , si−1 , si+1 , . . . , sn ), where each sj ∈ Sj
i’s choice • Accordingly, we write i’s utility as ui (si ; s−i ) juxtaposing i with full set
• E.g., Date Night: u1 (Movies; Theater) = 1 but
u2 (Movies; Theater) = 0
I Often, however, it is more convenient to simply list strategies
according to player labels/numbers
full list
• Thus, we may write i’s utility as ui (s1 , . . . , si−1 , si , si+1 , . . . , sn )
it’s giving
(note: we list strategies in order using commas—no semicolon)
ordered pair
• Date Night: u1 (Movies, Theater) = 1 = u2 (Movies, Theater)
and u1 (Theater, Movies) = 0 = u2 (Theater, Movies) 4 / 14
Beliefs and Best Responses
I Should Player 1 choose Movies or Theater?
• Recall that we assumed decision-makers are rational,
expected-utility maximizers
• Player 1 may have beliefs regarding Player 2’s likely choice
• Say Player 1 believes the probability that Player 2 chooses
Movies is pM :
EU1 (Movies) = pM u1 (Movies; Movies) + (1 − pM )u1 (Movies; Theater)
= pM (4) + (1 − pM )(1) = 3pM + 1,
EU1 (Theater) = pM u1 (Theater; Movies) + (1 − pM )u1 (Theater; Theater)
= pM (0) + (1 − pM )(3) = −3pM + 3

I Note that choosing Movies is optimal or a best response for

1
Player 1 (i.e., EU1 (Movies) ≥ EU1 (Theater)) only if pM ≥ 3
5 / 14
Best Response Correspondence
I We’ll typically focus just on players’ best responses to each
other’s strategies (in other words, we’ll only consider beliefs such
that players are certain about others’ behavior)
I Player i’s best response correspondence, denoted BRi , is a rule
that identifies, for each possible profile
s−i = (s1 , . . . , si−1 , si+1 , . . . , sn ) of strategies by the other
players, the set of strategies BRi (s−i ) that maximize i’s utility
assuming other players play s−i absolute certainty
• In other words, if i were certain that others would play s−i (i.e.,
if i believed so with probability 1), then i’s optimal behavior
would be to choose a strategy in BRi (s−i )
• Date Night: BR1 (Movies) = {Movies} = BR2 (Movies) and
solve by: figuring out one response,
BR1 (Theater) = {Theater} = BR2 (Theater) focus on that line/column
6 / 14
Nash Equilibrium
I By assuming rationality and expected-utility maximization, our
model yields useful predictions of behavior conditional on
decision-makers’ beliefs about external circumstances
I In games, one player’s choice affects others’ external
circumstances, so we may naturally want players’ beliefs and
choices to be mutually consistent
I A Nash equilibrium is a strategy profile (s∗1 , . . . , s∗n ) such that,
for every player i ∈ N , s∗i is a best response to s∗−i
• In other words, player i believes that others will play s∗−i , and i
can’t do any better than to respond by playing s∗i :

ui (s∗i ; s∗−i ) ≥ ui (si ; s∗−i )

* whatever the profile ascribes
for all si ∈ Si
7 / 14
Virtues of Nash Equilibrium
I Is Nash equilibrium an appealing solution concept for strategic
games? Are its predictions reasonable?
• Since every player is best responding to what everybody else is
doing, there are no regrets: after the fact, there isn’t anything
anybody could have done to make themselves better off
• Beliefs are self-fulfilling: when every player believes others will
play Nash equilibrium strategies, the best thing they can do for
themselves is to fulfill others’ expectations (no player has an
incentive to deviate from what’s expected of them)
• All players behave rationally given beliefs that are not only
plausible but, as it turns out, correct
“we are all mutually assuming each other to be rational”
8 / 14
Weaknesses of Nash Equilibrium
I When we require players to be certain about each other’s
behavior, Nash equilibria may not always exist
• This can be addressed by considering mixed strategies

I At the same time, Nash equilibrium may not deliver precise

predictions: a game may have multiple Nash equilibria, and we
may not be able to determine which is the “right” equilibrium
• In some cases, this can be addressed by considering games in
extensive form

I While a Nash equilibrium is robust to unilateral deviations (no

single player can do better for themselves), it may not be robust
to joint deviations by multiple players
9 / 14
Verifying Versus Finding Nash Equilibria

I As a practical matter, checking whether s∗ = (s∗1 , . . . , s∗n ) is a

Nash equilibrium is very straightforward:
• If there exists a player i ∈ N with a strategy s0i ∈ Si such that,
holding other players’ strategies fixed at s∗−i , we have
ui (s0i ; s∗−i ) > ui (s∗i ; s∗−i ), then s∗ is not a NE
• In other words, a strategy profile is not a NE if (at least) one
player has a profitable unilateral deviation

I In contrast, finding Nash equilibria is more involved:

• We need to check whether players’ best response
correspondences “cross”
have you underlined both numbers? Nash equilibria. BOOM. EXPLOSION NOISES.

10 / 14
Date Night Revisited
Player 2
Movies Theater
Movies 4,3 1,1
Player 1
Theater 0,0 3,4
I Is (Movies, Theater) a NE? No, Player 1 would strictly prefer to
go to the theater with Player 2
• Incidentally, Player 2 would also strictly regret not going to the
movies with Player 1, but Player 1’s profitable deviation is
already enough to rule out (Movies, Theater) as a NE
I Recall that BR1 (Movies) = {Movies} = BR2 (Movies) and
BR1 (Theater) = {Theater} = BR2 (Theater)
I So (Movies, Movies) and (Theater, Theater) are both NE:
players are mutually best responding to each other’s strategies
11 / 14
Matching Pennies
Player 2
Heads Tails
Heads 1,-1 -1,1
Player 1
Tails -1,1 1,-1

I Is (Heads, Heads) a NE? No, when Player 1 plays Heads, Player 2

strictly prefers playing Tails

I Is (Heads, Tails) a NE? No, Player 1 strictly prefers Tails

I Is (Tails, Heads) a NE? No, Player 1 strictly prefers Heads

I Is (Tails, Tails) a NE? No, Player 2 strictly prefers Heads

I So this game has no (pure strategy) Nash equilibria

12 / 14
Choosing a Number

I Two players have $10 to split, and they each submit a bid
si ∈ Si = [0, 10]
I Player i’s payoff: 
s
i if s1 + s2 ≤ 10,
ui (si ; s−i ) =
0 if s1 + s2 > 10

I What is i’s best response to s−i ? Well, they’d like to make their
bid as large as possible without

overshooting, so
10 − s if s−i < 10,
−i
BRi (s−i ) =
[0, 10] if s−i = 10
• Note that, if s−i = 10, then i is getting 0 regardless, so any bid
is optimal

13 / 14
Choosing a Number (cont’d)

I Any (s∗1 , s∗2 ) such that s∗1 + s∗2 = 10 is a NE

I (10, 10) is also a NE

14 / 14