Strictly Competitive Games

Felix Munoz-Garcia

EconS 503 - Washington State University

Reading materials - Strictly competitive games

Osborne, Chapter 11:

Posted on Angel website.
It is only 15 pages long, including examples, …gures,
experiments, etc.
Strictly competitive games

Some strategic situations involve players with completely

opposite interests/incentives.
We analyze those situations with Strictly Competitive Games.
They are a type of simultaneous-move games, as those
described so far...
but with an additional assumption (next slide):
Strictly competitive games

De…nition
A two-player, strictly competitive game is a two-player game with
the property that, for every two strategy pro…les s and s’,

u1 (s ) > u1 (s 0 ) and u2 (s ) < u2 (s 0 )

Intuition: Hence, players have exactly opposite rankings over

the outcomes resulting from the strategy pro…le s and s’.
Alternatively: if my payo¤ increases if we play s = (s1 , s2 )
rather than s 0 = (s10 , s20 ), then your payo¤ must decrease.
An implication is that, in a strictly competitive game,
if u1 (s ) = u1 (s 0 ), then u2 (s ) = u2 (s 0 ).
Alternatively, to check if a game is not strictly competitive,
we want to …nd two strategy pro…les (cells), s and s’for which
players’preferences are aligned, that is,
u1 (s ) > u1 (s 0 ), then u2 (s ) > u2 (s 0 )
Example 1 - Matching pennies

Player 2

Heads Tails
Player 1 Heads 1, 1 1, 1
Tails 1, 1 1, 1
One example of a strategy pro…le is s = (H, H ), and another
is s 0 = (T , H ), where
u1 (s ) > u1 (s 0 ) and u2 (s ) < u2 (s 0 )
Importantly, this is true for any two strategy pro…les: if one
player is improving his payo¤, the other player is reducing his.
In fact, many board games satisfy this condition: if we play in
such a way that I end up winning, it must be that my
opponent loses, and viceversa.
Examples: tennis, chess, football, etc.
Practice:

For the following games, determine which of them satisfy the

de…nition of strictly competitive games:
1 Matching Pennies (Anticoordination game),
2 Prisoner’s Dilemma,
3 Battle of the Sexes (Coordination game).
Matching Pennies

P2
Heads Tails

Heads 1, -1 -1, 1
P1
Tails -1, 1 1, -1

Check if this game satis…es the de…nition of strictly

competitive games.
Recall that we must check that, for any two strategy pro…les s
and s 0 ,
u1 (s ) > u1 (s 0 ) and u2 (s ) < u2 (s 0 )
Matching Pennies

Comparing each possible pair of outcomes

1 u1 (H, H ) > u1 (H, T ), i.e., 1 > 1
u2 (H, H ) < u2 (H, T ), i.e., 1 > 1

2 u1 (H, T ) < u1 (T , T ), i.e., 1 < 1

u2 (H, T ) > u2 (T , T ), i.e., 1 > 1

3 u1 (H, H ) = u1 (T , T ), i.e., 1 = 1
u2 (H, H ) = u2 (T , T ), i.e., 1 = 1

4 u1 (H, T ) = u1 (T , H ), i.e., 1 = 1
u2 (H, T ) = u2 (T , H ), i.e., 1 = 1
Prisoner’s Dilemma

Player 2
Confess Not Confess

Confess -5, -5 0, -15

Player 1
Not
Confess -15, 0 -1, -1

Check if this game satis…es the above de…nition of strictly

competitive games.
[Hint: What happens when you compare (C , C ) and
(NC , NC )? Preference alignment].
Battle of the Sexes

Wife
Football Opera

Football 3, 1 0, 0
Husband
Opera 0, 0 1, 3

Check if this game is strictly competitive game:

[Hint: What happens when you compare (F , O ) and (F , F )?
Preference alignment].
Zero-sum games
An interesting class of strictly competitive games: zero-sum
games.
A zero-sum game is an strictly competitive game in which the
payo¤s of the two players adds up to zero. That is
u1 (s1 , s2 ) + u2 (s1 , s2 ) = 0 for any strategy pair (s1 , s2 )
alternatively, u1 (s1 , s2 ) = u2 (s1 , s2 ).
For a general strictly competitive game we were saying that:
“if one strategy pro…le (s1 , s2 ) increases my payo¤, then...
...such strategy must reduce your payo¤”
but in a zero-sum game we are impossing an stronger
assumption:
“the payo¤ that I gain, is exactly what you lose”
Zero-sum games

The de…nition of a zero-sum game was satis…ed by the

matching pennies game....

Player 2

Heads Tails
Player 1 Heads 1, 1 1, 1
Tails 1, 1 1, 1

since u1 (s1 , s2 ) + u2 (s1 , s2 ) = 1 + ( 1) = 0, for any strategy

pro…le that especi…es one strategy for player 1 and one for
player 2, (s1 , s2 ).
Constant-sum games

Some games are not zero-sum games, but they are

constant-sum games. (They are of course an special type of
strictly competitive game; verify).
The following Tennis game is a constant-sum game

Player 2

Right Left
Player 1 Right 20, 80 70, 30
Left 90, 10 30, 70

since u1 (s1 , s2 ) + u2 (s2 , s1 ) = 100, for any strategy pro…le

(s1 , s2 ) that especi…es one strategy for player 1 and one for
player 2.
Constant-sum games

Hence, for any strategy pro…le (s1 , s2 )

u1 (s1 , s2 ) + u2 (s1 , s2 ) = 0, in zero-sum games
u1 (s1 , s2 ) + u2 (s1 , s2 ) = Constant, in constant-sum games
The Constant is exactly equal to zero in zero-sum games
Therefore

Zero-Sum Games

Constant-Sum Games

Strictly Competitive Games:

u1(s1,s2) + u2(s1,s2) Constant
Constant-sum games

Compact representation of the Tennis game:

Player 2

Right Left
Player 1 Right 20 70
Left 90 30

We don’t need to represent player 2’s payo¤, since we know

that in this constant-sum game

u1 (s1 , s2 ) + u2 (s1 , s2 ) = 100

for all strategy pro…les (s1 , s2 ).

Hence, player 2’s payo¤s are 80, 30, 10, and 70
Constant-sum games

How to solve this class of games?

We could use the NE solution concept (implying the need to
rely on msNE for most of these games).
An alternative, historically developed before John Nash
introduced his "NE solution concept," is to use the so-called:
Security strategies
(also referred as Max-Min strategies).
Security or Max-min strategy

Compact representation of the Tennis game:

Player 2

Right Left
Player 1 Right 20 70
Left 90 30

Note that Player 1 wants to maximize his own payo¤s, and...

Player 2 also wants to maximize his own payo¤s, which
implies minimizing Player 1’s payo¤s, since we are in a
constant-sum game.
Security or Max-min strategy
Let us put ourselves in the worst case scenario:
First, for a given strategy s1 that player 1 selects, choose the
strategy of player 2’s that minimizes player 1’s payo¤s.
w1 (s1 ) = min u1 (s1 , s2 )
s2

we refer to w1 (s1 ) as the worst payo¤ that player 1 could

achieve by selecting strategy s1 .
Alternatively, we can interpret that, if player 1 select strategy
s1 , he guarantees to obtain a payo¤ of at least w1 (s1 ).
A Security strategy gives player 1 the best of the worst case
scenarios:
max w1 (s1 ) = max min u1 (s1 , s2 )
s1 s1 s2

The strategy that solves this maximization problem is referred

as the Security strategy, or Max-min strategy.
Security or Max-min strategy

max w1 (s1 ) = max min u1 (s1 , s2 )

s1 s1 s2

The payo¤ max w1 (s1 ) is usually referred as the

s1
Security-payo¤ level.
Note what is happening here:
I maximize my payo¤, given that I know that my opponent will
minimize it (because he wants to maximize his own payo¤).
Security or Max-min strategy

We can generalize the above de…nition to mixed strategies,

i.e., talking about σi rather than si .
Player 1’s security payo¤ level is

max w1 (σ1 ) = max min u1 (σ1 , s2 )

σ1 σ1 s2

And similarly for player 2:

max w2 (σ2 ) = max min u2 (σ2 , s1 )

σ2 σ2 s1
Security or Max-min strategy

Let us …rst apply Security strategies to the example of the

Matching pennies game.
Afterwards, we will apply the same methodology to the Tennis
game.
Security or Max-min strategy

Note that, in order to …nd the security (or max-min) strategy

for player 1, we need to …nd

max min u1 (s1 , s2 )

s1 s2

We hence need to …rst …nd:

EU1 (p jH ) conditional on player 2 choosing H.
EU1 (p jT ) conditional on player 2 choosing T.
We can then …nd the min of these two expressions (i.e., their
"lower envelope").
Finally, we can …nd the max of the min.
Confused? Ok, let’s do one example together. !
Security (Max-min) Strategy - Matching Pennies Game

Player 2
Heads Tails

p Heads 1, -1 -1, 1
Player 1
1 - p Tails -1, 1 1, -1

1st step: Find the expected payo¤ of player 1

If player 1 chooses H (In the …rst column), player 1’s EU
becomes:
EU1 (p jH ) = 1 p + ( 1)(1 p ) = 2p 1
If player 1 chooses T (In the second column), player 1’s EU
becomes:
EU1 (p jT ) = ( 1) p + 1(1 p) = 1 2p
Security (Max-min) Strategy - Matching Pennies Game

Remark
Note that EU1 (p jH ) represents the expected utility that player
1 obtains from randomizing between H (with probability p)
and T (with probability 1 p), conditional on player 2
selecting Heads (in the …rst column).
Do not confuse it with EU1 (H ) that we used in msNE, which
re‡ects player 1’s expected utility from selecting H with
certainty but facing a randomization from his opponent (e.g.,
player 2 randomizing between H and T with probability q and
1 q respectively.
Security (Max-min) Strategy - Matching Pennies Game

2nd step: Let’s graphically depict EU1 (p jH ) and EU1 (p jT )

EU1(p| )
1

EU1(p|H) = 2p - 1

0 p
1
½

EU1(p|T) = 1 –2p

-1
Security (Max-min) Strategy - Matching Pennies Game
3rd step: Identify the lower envelope, i.e.,
min u1 (s1 , s2 )
s2

EU1(p| )
1

EU1(p|H) = 2p - 1

0 p
1
½

EU1(p|T) = 1 –2p

-1
min u1(s1,s2) is the lower envelope
s2
Security (Max-min) Strategy - Matching Pennies Game
4th step: Identify the highest peak of the lower envelope: i.e.,
max min u1 (s1 , s2 )
s1 s2

EU1(p| )
1

EU1(p|H) = 2p - 1

Highest point of the lower

envelope, that is max min u1(s1,s2)
s1 s2
0 p
1
½

EU1(p|T) = 1 – 2p

-1
min u1(s1,s2) is the lower envelope
s2
Security (Max-min) Strategy - Matching Pennies Game

Summarizing our results...

We just found that the Security (or Max-Min) strategy for
player 1 is:
To choose Heads with probability p = 21 .
What about player 2?
Well, we have to follow the same procedure we used with
player 1.
Practice on your own (see next two slides).
[Hint: you should …nd that player 2 also randomizes with
probability q = 21 ].
Security (Max-min) Strategy - Matching Pennies Game

Similarly for player 2 (Practice!)

Player 2
q 1-q
Heads Tails

Heads 1, -1 -1, 1
Player 1
Tails -1, 1 1, -1

Expected payo¤ for player 2:

If player 1 plays H (…rst row):

EU2 (q jH ) =

If player 2 plays T (second column):

EU2 (q jT ) =
Security (Max-min) Strategy - Matching Pennies Game

Graphical depiction for player 2

EU2(q| )
1

0 q
1

-1
Security (Max-min) Strategy - Tennis Game

Let’s go back to the Tennis game:

It is a constant-sum game, since the sum of players’payo¤s is
equal to a constant (100), for all possible strategy pro…les (i.e.,
for all possible cells in the matrix), but...
It is not a zero-sum game, since the sum of players’payo¤s is
not euqal to zero for all strategy pro…les.

Player 2
Right Left

Right 20, 80 70, 30

Player 1
Left 90, 10 30, 70

Let’s start with player 1:

Security (Max-min) Strategy - Tennis Game

Player 2
Right Left

p Right 20, 80 70, 30

Player 1
1 - p Left 90, 10 30, 70

Player 1’s expected payo¤:

If player 2 chooses Right:

EU1 (p jR ) = 20p + 90(1 p ) = 90 70p

If player 2 chooses Left:

EU1 (p jL) = 70p + 30(1 p ) = 30 + 40p

Security (Max-min) Strategy - Tennis Game

Graphical depiction for player 1:

EU1(p| )
90 EU1(p|R) = 90 – 70p

70

570 EU1(p|L) = 30 + 40p

11

30
20

0 6
p
11 1
Lower envelope: min u1(s1,s2)
s2
Security (Max-min) Strategy - Tennis Game

Trick:
max min u1 (s1 , s2 )
s1 s2

coincides with the value of p for which

EU1 (p jR ) = EU1 (p jL).

That is,
6
70 70p = 30 + 40p =) 60 = 110p =) p =
11
Hence,
6 570
EU1 (p jR ) = 90 70 =
11 11
|{z}
This is the height of the highest
peak in the lower envelope.
Security (Max-min) Strategy - Tennis Game

Similarly, for player 2

Player 2
q 1-q
Right Left

Right 20, 80 70, 30

Player 1
Left 90, 10 30, 70

Player 2’s expected payo¤:

If player 1 chooses Right (…rst row):

EU2 (q jR ) = 80q + 30(1 q ) = 30 + 50q

If player 2 chooses Left (second row):

EU2 (q jL) = 10q + 70(1 q ) = 70 60q

Security (Max-min) Strategy - Tennis Game

Graphical depiction for player 2 (Practice!):

EU2(p| )

0 q
1
Security or Max-min strategy

Security strategies were introduced at the beginning of the

century before Nash came out with his equilibrium concept...
For this reason, solving a game using security strategies does
not necessarily give us the same equilibrium prediction as if we
use Nash equilibrium.
Although there is one exception!
Such exception is, of course, stricty competitive games.
Security or Max-min strategy

Hence,
Security (Max-min) Strategies

Strictly Competitive Games

Nash Equilibrium
Strategies

Let us see the relationship between the equilibrium predictions

using Security strategies and that using NE.
Security or Max-min strategy

Relationship between Security strategies and NE

strategies:
If a two-player game is strictly competitive and has a Nash
equilibrium s = (s1 , s2 ), then...
s1 is a security strategy for player 1 and s2 is a security
strategy for player 2.
That is, if s1 is a NE strategy for player 1 in a strictly
competitive game, then s1 guarantees player 1 at least his
security payo¤ level, regardless of what player 2 does.
In other words, by playing the NE strategy a player guarantees
a payo¤ equal or higher than that he would obtain by playing
the Security (or Maxmin) strategy.
Practice:
2. Consider the following game:
Player 2

Left Right
Player 1 Top 6, 0 0, 6
Bottom 3, 2 6, 0
1 Find every player’s maxmin strategy.
2 What is every player’s expected payo¤ from playing her
maxmin strategy?
3 Find every player’s Nash equilibrium strategy, both using pure
strategies (psNE) and using mixed strategies (msNE).
4 What is every player’s expected payo¤ from playing her Nash
equilibrium strategy?
5 Compare players’payo¤ when they play maxmin and Nash
equilibrium strategies (from parts (b) and (d), respectively).
Which is higher?
What if a game is not strictly competitive?

Consider the following game:

Player 2

X Y
Player 1 A 3, 5 1, 1
B 2, 6 1, 2

1 This is indeed an example of a game that does not satisfy

the de…nition of strictly competitive games. In particular, we
can …nd two strategy pro…les, (A,X) and (A,Y) for which

u1 (A, X ) > u1 (A, Y ) for player 1,

but also u2 (A, X ) > u2 (A, Y ) for player 2!!
What if a game is not strictly competitive?

Player 2
X Y

A 3, 5 -1, 1
Player 1
B 2, 6 1, 2

1 The game has a unique psNE: (A, X ).

2 But, is A the security strategy for player 1?
We know that this is the case in strictly competitive games,
but...
this is not necessarily true in games that are not strictly
competitive (such as this one).
In order to check if A is a security strategy for player 1, let’s
…nd player 1’s security strategies !
What if a game is not strictly competitive?

Player 2
X Y

p A 3, 5 -1, 1
Player 1
1-p B 2, 6 1, 2

1 In order to check if A is the security strategy for player 1, we

must …nd EU1 (p jX ) and EU1 (p jY ).

EU1 (p jX ) = 3p + 2(1 p) = 2 + p
EU1 (p jY ) = 1p + 1(1 p) = 1 2p
What if a game is not strictly competitive?

Graphical representation for player 1:

EU1(p| )
1

2
EU1(p|X) =2 + p

1
EU1(p|Y) = 1 –2p

1
0
p

-1
min u1(s1,s2) is the lower envelope
s2
What if a game is not strictly competitive?

EU1 (p jX ) and EU1 (p jY ) do not cross for any probability

p 2 (0, 1).
EU1 (p jY ) is the minimum of EU1 (p jX ) and EU1 (p jY ), i.e.,
the "Lower envelope."
The lower envelope is maximized at p = 0.
Hence, player 1 does not assign any probability to action A,
but full probability to B
That is, B is player 1’s security strategy (which di¤ers from his
Nash Equilibrium strategy, A).
What if a game is not strictly competitive?

This con…rms our previous result that:

1 NE and Security strategies coincide for strictly competitive
games, but...
2 NE and Security strategies do not generally coincide for
games that are not strictly competitive.

Security (Max-min) Strategies

Action B in the
previous example Action A in the
previous example

Strictly Competitive Games

Nash Equilibrium
Strategies