Sie sind auf Seite 1von 21

Unpredictability

In zero-sum games: the min-max theorem


In non zero-sum games
Payoff changes
Unpredictability
Correctly anticipating and responding to
unpredictability is useful:
The IRS wants to audit tax evaders; cheaters hope to avoid an audit
In standard setting: the dominant firm, with a large installed base,
wants to retain control of its standard; a small, less well-established
firm wants to adopt the standard through an adapter
Invader wants to achieve tactical surprise in its choice of a place to
attack; the defender wants to concentrate its forces on the place of
attack
The setters of fashion in restaurants, clothing wants exclusivity;
the general public wants to mingle with the trend-setters
Unpredictability
While the taxmans or the attackers decision may
be unpredictable, there are rules which govern the
selection
The right amount of unpredictability should not be
left to chance
The odds of choosing one move or another can be
precisely determined from the particulars of the
game
Zero-sum game: the tennis case
The Server wants to minimize the probability that
the Receiver can return the service,
while the Receiver wants to maximize this
probability
Here the Servers gains are the Receivers loss: the
game is zero-sum
S can choose to serve on forehand or backhand
R can move appropriately and get prepared to
receive on its forehand or backhand
Zero-sum game: the tennis case
The percentages of Rs successfully returning the
service are given in the payoff matrix:

S Forehand Backhand
R
Forehand 90 20
Backhand 30 60

No Nash-equilibrium in pure strategy


Zero-sum game: the tennis case
If S always aims to forehand, R successfully returns 90%
of times if it anticipates, and gets prepared to, a service on
his forehand
If S always aims to backhand, R successfully returns 60%
of times if it gets prepared to a service on his backhand
If S randomizes, and aims to forehand half of times:
if R waits for F, he returns only 55% of times
if R waits for B, he returns only 45% of times
Hence S is better off randomizing: what is the best mix?
Zero-sum game: the tennis case
Plot the percentage of successful returns against
percentage of times S aims to Forehand:

R wants to keep return percentage as RMF 90


high as possible, along the green line

60
55
48
45
30
20 RMB

40 50
Zero-sum game: the tennis case
The mix 40:60, where S aims 40 percent of times to
forehand is the one that minimizes the percentage of
successful returns when R plays its best response (along
the green line): it wants to minmax (minimize the
maximum payoff of R); it is then 48%
Thus 40:60 is Ss best mix
Now let us look at the game from the Receivers point of
view by plotting the percentage of successful returns
against percentage of times R moves to Forehand.
By correctly exploiting Rs move, S can
keep return percentage as low as possible,
along the red line; hence the best choice of
R is to move to its forehand 30% of times:
this keeps its return percentage to the
highest, i.e., 48% SAF 90

60
48

30
SAB
20

30
Zero-sum game: the tennis case
The mix 30:70, where R anticipates 30 percent of
times, and gets prepared to, a service on its
forehand is the one that maximizes the percentage
of successful returns when S plays its best
response (along the red line): it wants to maxmin
(maximize its minimum payoff); it is then 48%
From both point of view, the best mix leads to a
percentage of successful returns of 48%
This is a general property of zero-sum games,
called the min-max theorem
Zero-sum game: the tennis case
When, in zero-sum games, one player attempts to
minimize his opponents maximum payoff, while
his opponent attempts to maximize his opponents
minimum payoff, the surprising conclusion is that
the minimum of the maximum payoffs equals the
maximum of the minimum payoffs
Neither player can improve his position, and so
these strategies form an equilibrium of the game
If R moves 50% on F, then his percentage of
return, as a function of the probability p S aims
at F is: 0.5(90p+20(1-p))+0.5(30p+60(1-p)) =
20p+40 is minimum for p=0

By choosing to move 30% on


F, R secures the highest
Min minimum payoff: this mix
with R RMF 90
makes S indifferent about
50% F Min
his own mix with R
60 20p+40 30% F
48
40
30
20 RMB
Min
Min
with R
with R
40 0% F
100% F
Non zero-sum games
Game between the sponsor of a network which wants to
retain control of its standard and a firm which wants to
hook to this standard (ex: a big car constructor and a
company selling parts: Renault and Valeo)
Two possible choices of standard for gear boxes of
Meganes: A and B
In case they make the same choice, they share the market,
but R is much better with standard A
In case R manages to make a different choice than V, R
gets the whole market for replacement gear boxes of
Meganes, and extract a monopolistic rent: it gets 300
Non zero-sum games
Payoff matrix is:

V A B
R
A 150 , 50 250 , 0
B 300 , 0 100 , 100

We look at the problem from Valeos point of view


It is a simultaneous move game
Non zero-sum games
Probability that R chooses standard A: p
Probability that V chooses standard A: q
R chooses p so as to maximize its expected payoff:

p[150q+250(1-q)]+(1-p)[300q+100(1-q)] =
(100+200q)+(150-300q)p

Similarly, V chooses q so as to maximize expected payoff:


q[50p]+(1-q)[100(1-p)] = (100-100p)+(150p-100)q
Non zero-sum games
V should expect that R is going to anticipate qe
If qe < 0.5, then V knows that R will respond p = 1
If p = 1, then V should respond q = 1 hence the
conjecture qe < 0.5 leads to q = 1!
The same, the conjecture qe > 0.5 leads to q = 0
Only when qe = 0.5 is Renault indifferent between A and
B, hence Renault is unpredictable
The same, only when V expects R to play pe = 2/3 is V
indifferent between A and B, hence unpredictable
Non zero-sum games
Expectations are at equilibrium when qe = 0.5; then p = 2/3
is a best response of Renault, to which q = 0.5 is actually a
best response of Valeo
Hence the Nash-equilibrium in mixed strategy:
(p,q) = (2/3,1/2)
And equilibrium payoffs are: 200 for R, 33 for V
In fact there is always a Nash equilibrium (possibly in
mixed strategy) as soon as
There is a finite number of players
Each of whom has a finite number of pure strategies
Payoff changes
Suppose the receiver practices to improve his backhand:
percentage of returns rises to 65

The receiver best mix rises to 33.3, and his SAF 90


percentage of successful returns to 50
65
60
48 50

30
SAB
20

30 33.3
Payoff changes
Paradoxically, a higher percentage of successful returns is
achieved by using less often the improved backhand
The reason lies in the interaction of the two players
strategies: when the receiver is becoming better with his
backhand, the server aims at the forehand more often (43%
instead of 40%), in response the receiver moves too his
forehand more often too.
A better backhand unlocks the power of your forehand
Similarly for a basket-ball player: suppose her right-hand
shooting is better than her left-hand; an improvement in
her left-hand shooting changes the way she is defended
and allows her to shoot right-hand more often
Equilibrium mix
If one player is not pursuing his equilibrium mix, his
opponent can exploit this to his advantage
So, even though at your opponents equilibrium mix you
are indifferent between your strategies: play your best
mix! otherwise your opponent is going to exploit it
In the tennis example, each serve has to be unpredictable
(if your best mix is to play 40% on forehand, do it
whenever the second hand of your (analogue) watch is
between 1 and 24)
The best way to surprise the others is to surprise
yourself and toss a coin if your best mix is 50/50
Examples
Airlines and discount/stand-by tickets: if last minute ticket
availability were more predictable, there would be a
greater possibility to exploit the system
To motivate compliance at a lower monitoring cost: tax
audits, drug testing, parking meters, etc Explains why
the punishment should not necessarily fit the crime
If parking meter costs 1 euro an hour, then a fine of 20
euros will keep people honest if they believe the
probability of a fine is 5% or higher (risk-neutral);
No enforcement induces misuse of scarce parking places
100% enforcement would be too expensive
But enforcement mechanism should not be any: the expected fine
should be high enough to induce compliance
But then: random walks of officers!

Das könnte Ihnen auch gefallen