Game Theory

Lecture 10

Leong Kaiwen

Assistant Professor in Economics

Nanyang Technological University

2021

Leong Kaiwen (NTU) Game Theory NTU 2021 1 / 19

Williams Sister Tennis

S
l r
L (50,50) (80,20) p* p*=0.7
V q*=0.6
R (90,10) (20,80) 1-p*
q* 1-q*
Now we want to check if p∗ is a BR to q∗ .
V’s payoff of choosing L against q∗ =50(q∗ )+80(1-q∗ )=0.62
V’s payoff of choosing R against q∗ =90(q∗ )+20(1-q∗ )=0.62
V’s payoff from p∗ = 0.7∗0.62+0.3∗0.62=0.62

Leong Kaiwen (NTU) Game Theory NTU 2021 2 / 19

Williams Sister Tennis

Is p∗ in fact a BR?
If it is not a BR, there must be some deviation V could make that would
make a strictly better off. If V deviates to either L or R, the payoff is still
0.62, which means V is not strictly better off. p∗ is as good as playing
L or R.
So there is no way to deviate to a pure strategy in a strict profitable way,
but how about deviating to another mixed strategy?
Any mix that V deviates to will be a mix between L and R and any mix
between L and R will be a mix between 0.62 and 0.62 and hence will be
0.62.
Recall:Any mixed strategy yields a payoff that is a weighted average of
the pure strategy payoffs that are in the mix.

Leong Kaiwen (NTU) Game Theory NTU 2021 3 / 19

Williams Sister Tennis

Key: If we have shown that there are no pure strategy deviation that’s
strictly profitable then there can’t be any mixed strategy deviation
that’s strictly profitable. Why? Because the mixed strategy deviations
must yield payoffs that lie among the pure strategy deviations.
Lesson: we only ever have to check for strictly profitable pure strategy
deviations.
Note: S is not trying to make V indifferent between L and R. Similarly,
V is not trying to make S indifferent between l and r. It is the case
that in equilibrium, V will make S indifferent and vice-versa. If p∗ >
0.7, then S will choose l all the time but V will not want to choose L
anymore. So it is not an equilibrium. It has to be the case that p∗ &
q∗ make both players indifferent.
Another application of mixed strategies: baggage checking machines
at airports.

Leong Kaiwen (NTU) Game Theory NTU 2021 4 / 19

Battle of Sexes (Revisited)

Player 2
AP P Pure strategy NE:
Apple Picking (AP) (2,1) (0,0) p (AP,AP), (P,P)
Player 1
Play (P) (0,0) (1,2) 1-p
q 1-q
Find mixed NE in this game
To find NE q, we use player 1’s payoffs
Player 1’s payoff for choosing AP = 2q+(1-q)0
Player 1’s payoff for choosing P=0q+(1-q)1
Since player 1 is mixing, these two payoffs must be equal.
⇒ 2q+(1-q)0=0q+(1-q)1
1
⇒ q=
3

Leong Kaiwen (NTU) Game Theory NTU 2021 5 / 19

Battle of Sexes (Revisited)

Find mixed NE in this game

To find NE p, we use player 2’s payoffs.
Player 2’s payoff for choosing AP=1p+0(1-p)
Player 2’s payoff for choosing P=0p+2(1-p)
Since player 2 is mixing, these two payoffs must be equal
⇒ 1p+0(1-p)=0p+2(1-p)
2
⇒ p=
3

Leong Kaiwen (NTU) Game Theory NTU 2021 6 / 19

Battle of Sexes (Revisited)

Check that what we have found is indeed a NE.

2
Check that p= is a BR for player 1.
3
1 2 2
Player 1’s payoff to AP=2q+0(1-q)=2* +0*( )=
3 3 3
1 2 2
Player 1’s payoff to P= 0q+1(1-q)=0*( )+1*( )=
3 3 3
2 1
Player 1’s payoff to choosing p= (AP payoff)+ (payoff to p)
3 3
2 2 2 1 2
= *( )+ *( )=
3 3 3 3 3
Since there is no strictly profitable pure deviation, there will be no other
strictly profitable mix deviation as well. So indeed for player 1, p is a
BR to q. We can do the same for player 2.
2 1 1 2 2 2
Mixed NE[( , ),( , )], payoffs to mixed NE=( , )
3 3 3 3 3 3

Leong Kaiwen (NTU) Game Theory NTU 2021 7 / 19

Battle of Sexes (Revisited)

Why is the payoff so bad for mixed NE?

2
In the pure NE, worst payoff was 1 & best was 2. Here, mixed NE is
3
What is forcing the payoffs down in the mixed NE is that player 1 & 2
are not meeting that often.
Player 2
AP P
AP (2,1) (0,0) p
Player 1
P (0,0) (1,2) 1-p
q 1-q
1 2 1 2 4
Probability the 2 players meet=q*p + (1-p)(1-q)=( )*( )+( )*( )=
3 3 3 3 9
4 5
of the time 2 players meet. of the time 2 players not meeting.
9 9

Leong Kaiwen (NTU) Game Theory NTU 2021 8 / 19

Battle of Sexes (Revisited)

Since randomizing is bad for both players, why are the 2 players doing
this?
We will think about a second interpretation for what we think mixed
strategy equilibria are.
Rather than think about both players as literally randomizing, it is prob-
ably better in this case to think about the following idea. We need to
think about player 2’s mixture as being a statement about what player
1 believes player 2 is going to do. Similarly, we can think of player 1’s
mixture as a statement about what player 2 believes about what player
1 will do.
In the NE, we found the beliefs such that these players are exactly indif-
ferent over what they do.
In this setting, better to think about mixed strategies being a statement
about what people believe in equilibrium.

Leong Kaiwen (NTU) Game Theory NTU 2021 9 / 19

Battle of Sexes (Revisited)

Interpretation of mixed strategy equilibria.

So far we have 2.
People are literally randomizing
Mixed strategies are expressions about what people believe in equilib-
rium.
Now consider 3rd interpretation.

Leong Kaiwen (NTU) Game Theory NTU 2021 10 / 19

Tax Game

Taxpayer
H C Where, H is ”honest report”, C is
A (2,0) (4,-10) ”cheat”, A is ”audit”, N is ”not audit”
Auditor
N (4,0) (0,4)
Interpretation of payoff
Auditor is happy if she dose not audit & taxpayer doesn’t cheat.
Auditor is happy if she audits & finds taxpayer cheating
Auditor is not happy if she audits & taxpayer doesn’t cheat because
audits are costly.
The auditor is really unhappy if she did not audit & the taxpayer cheated.
For taxpayer, if honest and pay what they are supposed to pay, get payoff
of 0
For taxpayer, if cheat & not audited, get to keep a chunk of money.
For taxpayer, if cheat & caught, get a huge fine.

Leong Kaiwen (NTU) Game Theory NTU 2021 11 / 19

Tax Game

No pure strategy NE in this game

For taxpayer, if he is going to be audited, they will rather choose H.
For taxpayer, if he is going not to be audited, they will rather choose C.
For auditor, if taxpayer chooses H, auditor will prefer N.
For auditor, if taxpayer chooses C, auditor will prefer A.
No box in which the BR coincides, there is no pure strategy NE.

Leong Kaiwen (NTU) Game Theory NTU 2021 12 / 19

Tax Game
Find mixed strategy NE.
Taxpayer
H C
A (2,0) (4,-10) p
Auditor
N (4,0) (0,4) 1-p
q 1-q
Find proportion q of taxpayers that pay their taxes.
Look at auditor’s payoff
Auditor’s payoff from choosing A=2q+4(1-q)
Auditor’s payoff from choosing N=4q+0(1-q)
If the auditor is mixing, 2q+4(1-q)=4q+0(1=q)
2
q=
3
To make the auditor exactly indifferent between audit or not, it must
2 1
be the case that of taxpayers are honest and of all taxpayers not
3 3
honest.
Leong Kaiwen (NTU) Game Theory NTU 2021 13 / 19
Tax Game

Look at taxpayer’s payoff

Payoff taxpayer chooses H=0
Payoff taxpayer chooses C=-10p+4(1-p)
If taxpayer is mixing, 0=-10p+4(1-p)
2
p=
7
2 5 2 1
We should check that [( , ),( , )] is a NE. Go home & do.
7 7 3 3

Leong Kaiwen (NTU) Game Theory NTU 2021 14 / 19

Tax Game

2 5 2 1
How can we interpret [( , ),( , )]?
7 7 3 3
From the point of auditor: We can think of the auditor as literally ran-
domizing. By law, the auditors literally have to randomize.
From the point of taxpayers: Some taxpayers are paying their taxes
& some are not. There are a lot of taxpayers in the population. In
2 1
equilibrium, of taxpayers in the population will pay taxes & will
3 3
cheat. We can think of mixed strategies not as randomizing but as a
mix in a large population in which some people are doing one thing and
the other group are doing another thing. Here the thing = proportion
of people paying taxes.

Leong Kaiwen (NTU) Game Theory NTU 2021 15 / 19

Tax Game

Experiment: Let’s raise the fine to cheating from -10 to -20.

Taxpayer
H C
A (2,0) (4,-20) p
Auditor
N (4,0) (0,4) 1-p
q 1-q
What happens to tax compliance(q) in this game?
Work out the q in equilibrium by looking at auditor’s payoffs.
Payoff to A= 2q+4(1-q)
Payoff to N= 4q+0(1-q)
If auditor is mixing, then 2q+4(1-q)= 4q+0(1-q)
2
q=
3

Leong Kaiwen (NTU) Game Theory NTU 2021 16 / 19

Tax Game
In equilibrium, tax compliance has not changed at all. Why did the
increase in fine make no difference to tax compliance?
What determines the equilibrium mix for the column player are the row
player’s payoffs. We did not change the row player’s payoffs. So we are
not going to change the equilibrium mix of the column player.
What would have changed the new equilibrium?
The probability of audit would have changed.
To find the new p, we need to look at the tax payer’s payoff
payoff to H = 0p+(1-p)0
payoff to C = -20p+4(1-p)
Since taxpayers are mixing,
0=-20p+4(1-p)
1
p=
6
2 1
Audit rate has gone down from to . Since audits are costly, it is
7 6
good we lowered it. But we didn’t manage to increase the compliance
rate of taxpayers.
Leong Kaiwen (NTU) Game Theory NTU 2021 17 / 19
Tax Game

Lesson from the model:

Suppose we now have,
Taxpayer
H C
A (2,0) (4,-10)
Auditor
N (4,0) (0,6)
In this setting, the population of the cheaters aren’t necessarily going
to increase, but the probability of getting audited will increase. Gains
from cheating increased means you need to audit more to push back
to equilibrium.

Leong Kaiwen (NTU) Game Theory NTU 2021 18 / 19

Tax Game

How do we change incentives in the game to increase compliance rate?

We should change the payoffs to the auditor.

Or have somebody could set audit rates exogenously.
Key lesson reminder:
When checking for mixed strategy deviations, you only have to check
for pure strategy deviations. Make sure to check for all pure strategy
deviations, not just the pure strategies involved in the mix. If the guy
has 7 pure strategies & is mixing on 2, have to check all 7.

Leong Kaiwen (NTU) Game Theory NTU 2021 19 / 19