Strategic Behavior vs DA

 Decision analysis : (optimizing against nature)

 Should I take an umbrella?
Game theory : Introduction  What crops should I plant?
 Shall Company x drill for oil?
1. Strategic behavior  Strategic behavior (optimizing against intelligent opponent)
2. two-person zero-sum games  Outcome depends on choices by both (or all) players
3. Pure strategy
gy  So what I should do depends on what you do
4. Mixed strategy  What you should do depends on what I do
5. Prisoner’s dilemma

Game theory Game theory

 Examples of competing players are :
 Game theory is a mathematical theory that deals with the
general features of competitive/conflicting  teams, armies, companies, political candidates,
situations. contract bidders,
bidd etc..
 Game Theory is a set of tools and techniques for  Game theory application areas:
decisions under uncertainty involving two or more  Economics,
intelligent opponents in which each opponent aspires to  Business competition,
optimize his own decision at the expense of the other  Politics/diplomacy,
opponents.  Military/war,
 Biology,
 etc.

3 4

1
 Game Theory Game theory: assumptions
 In game theory, an opponent is referred to as player. Each (1) Each decision maker has available to him two or more
player has a number of choices,
choices finite,
finite called strategies.
strategies p
well-specified choices or sequences
q of choices.
 The outcomes or payoffs of a game are summarized as
functions of the different strategies for each player. (2) Every possible combination of plays available to the
 Objective: Finding acceptable, if not optimal, strategies in players leads to a well-defined end-state (win, loss, or draw)
conflict situations. that terminates the game.
 Abstraction of real complex situation, highly mathematical
(3) A specified payoff for each player is associated with each
 Research on game theory continues to deal with complicated
end-state.
types of competitive situations.

5 6

Game theory: assumptions Maximin criterion

 Game theory deals with the determination of the “optimal”
(4) Each decision maker has perfect knowledge of strategies for each player.
th game andd off hi
the his opposition.
iti
 Assumption:
 in view of the conflicting nature and lack of information about
the specific strategies selected by each player, optimality is
(5) All decision makers are rational; that is, each based on a rather conservative criterion, namely, each player
player, given two alternatives, will select the one selects his strategy which guarantees a payoff that can never be
worsened byy the selections of his opponent.
pp This criterion is
that
h yields
i ld hi
him the
h greater payoff.
ff known as the maximin criterion (or minimax criterion).

7 8

2
 Types of Games Two-persons Zero Sum game
 Two competing players in the game.
 Zero sum vs. non-zero sum
 Zero-sum means the ggain (or
( loss)) for one player
p y is equal
q to the
 Two person or N-person game corresponding loss (or gain) for the other player.
 Perfect vs. Imperfect information  Two sides in conflict: What one player wins, the other player loses.
 A strategy is pure, whenever a game is repeated each player choose
 Cooperative vs. conflict the same strategy of his/her own – no change of strategy for a
 Sequential vs. Simultaneous moves repeated number of times
 A mixed
i d strategy
t t is i one where
h th the players
l iintroduce
t d a random
d
 Single Play vs. Iterated
element in their choice of a strategy each time.


9 10

Non-zero Sum Game N-person games

 The sum of payoffs is not constant during the course of game.  With a third player, the possibility of a coalition arises
 Players may co
co-operate
operate or compete  Within the coalition there are issues of payoff
 Some nonzero sum games are positive sums and some distribution
negative sum  Coalitions disintegrate / realignments occur
 Negative sum games are competitive , everybody loses (e.g. a  Solutions for N-person games can be difficult
strike, war)
 One examplep two-persons
p nonzero-sum ggame is the
prisoner’s dilemma (a non –cooperative game)

11 12

3
 Two-Person, Zero-Sum Game
Games of Perfect/imperfect Information – Pure Strategy
Perfect Information  Two players with a finite set of strategies
 The information concerningg an opponent’s
pp move is well  Zero
Zero-sum
sum implies that the losses of one player is the
known in advance. exact gain of the other.
 All sequential move games are of this type.  Pure strategy refers to a prescribed solution in which
Imperfect Information one alternative is repeatedly recommended to each
 Partial or no information concerning the opponent is given in player
advance to the pplayer’s
y decision.  Neither player can improve his/her outcome by
 Imperfect information may be diminished over time if the changing strategies even if he/she learns in advance the
same game with the same opponent is to be repeated. opponent’s strategy

13 14

Two-persons game Two-persons game

Thus the game is represented by the payoff matrix to player A A1,A2,…..,Am are the strategies of player A
as B1,B2,…...,Bn are the strategies of player B
B1 B2 ……… Bn
aij is the payoff to player A (by B) when the player A plays
A1 a11 a12 ……… a1n strategy Ai and B plays Bj (aij is –ve means B got |aij| from A)
A2 a21 a22 …........ a2n
. .
. .
Am am1 am2 ………. amn
15 16

4
 Two-Person Zero-Sum game Example 1
Determine the saddle-point solution, the associated pure strategies,
 Basic assumption: Each player chooses a strategy that
and the value of the game for the following game. The payoffs are for
enables him/her to do the best he/she can,, given
g that player A are given.
given
his/her opponent knows the strategy he/she is following.
 Solution Strategy : Minimax – Maximin Approach B1 B2 B3 B4 Row min
 Minimax - to minimize the maximum loss (defensive)
A1 8 6 2 8 2
 Maximin - to maximize the minimum gain (offensive)
A2 8 9 4 5 4
 A two-person
t zero-sum game hhas a saddle
ddl point i t if andd
only if A3 7 5 3 5 3
max min
 Max (row minimum) = min (column maximum)
Max Col 8 9 4 8

17 18 min max

Two-person zero-sum games Example 2 (advertising)

 The solution of the game is based on the principle of securing the  Two companies, A and B, sell two brands of products. Company A
best of the worst for each player.
p y advertises in radio (Al), television (A2), and newspapers (A3). Company
B in
B, i addition
dditi tot using
i radiodi (B1),television(B
) t l i i (B2),and
) d newspapers (B3)also
)l
 For player A: mails brochures (B4). Depending on the effectiveness of each advertising
campaign, one company can capture a portion of the market from the
 To maximize the minimum return, he/she should play
other. The following matrix summarizes the percentage of the market
strategy 2 captured or lost by company A.
 For Player B:
B1 B2 B3 B4
• To minimize the maximum loss,
loss B should play strategy 3.
3
A1 8 -2 9 -3
 max (row minima) = min (column maxima) = is called the value
The maximin
of the game. 4 is called the saddle-point (A2, B3) A2 6 5 8 8 equals to Minimax
At the saddle point, neither player can improve their payoff at A2 &B2
by switching strategies A3 -2 4 -9 5
19 20

5
 Solution Example 3: Labor Dispute
 The optimal solution: strategies A2 and B2  The labor contract between a company and the union will terminate in the near
future. A new contract must be negotiated. After a consideration of past
 Company A share will increase by 5%. The value of the experience the Company agrees that the feasible strategies for the company to
experience,
game is 5% follow are:
 C1 = all out attack; hard aggressive bargaining
 If B moves to another strategy (B1, B3, or B4), Company  C2 = a reasoning, logical approach
A can stay with strategy A2, which ensures that B will  C3 = a legalistic strategy
lose a worse share of the market (6% or 8%).  C4 = an agreeable conciliatory approach
 Assume that the union is considering one of the following set of approaches:
 Similarly, A does not want to use a different strategy
 U1 = all out attack; hard aggressive bargaining
because if A moves to strategy A3, B can move to B3 and  U2 = a reasoning, logical approach
realize a 9% increase in market share. A similar  U3 = a legalistic strategy
conclusion is realized if A moves to A1, as B can move to  U4 = an agreeable conciliatory approach

B4 and realize a 3% increase in market share.

21 22

Example 3 Pure Strategy : Remarks

 With the aid of an outside mediator, the following game  Change in Strategy – Since games are repetitive, both
matrix is constructed: players may change. But in pure strategy games, there is no
 Conditional Gains of Union incentive to change.
 Any player deviating from the prescribed strategy will usually
find a worsening payoff.
 Multiple Optimal Solutions – Some games may involve
multiple optimal strategies.
 Dominance : D Dominated
i t d rows andd columns
l can bbe ddeleted
l t d
from the table.

 HW: Solution: C3 and U1 represents the pure strategy

solution to this game theory problem.
23 24

6
 Example 4: Multiple pure strategy Two-person zero-sum games
 Example 4: Given game matrix showing the conditional gains
Definition:
of A.
A strategy is dominated by a second strategy if the second
strategy is always at least as good (and sometimes better)
regardless of what the opponent does. Such a dominated
strategy can be eliminated from further consideration.
pp y concept
Apply p of dominated strategies
g on Example p 1 to rule
out succession of inferior strategies until only 1 choice remains.
 Multiple Pure Strategy Solutions: A2, B2 and A4, B2

25 26

Two-person zero-sum games Two-person zero-sum games

Example 1: for player A, strategy A3 is dominated by
B1 B2 B3 B4
the strategy A2 and so can be eliminated.
A1 8 6 2 8
B1 B2 B3 B4 A2 8 9 4 5

A1 8 6 2 8
For player B, strategies B1, B2, and B4 are dominated by the
A2 8 9 4 5 gy B3.
strategy
A3 7 5 3 5 Eliminating the strategies B1 , B2, and B4 we get the reduced
payoff matrix:
Eliminating the strategy A3 , we get the following reduced
27
payoff matrix: 28

7
 Two-person zero-sum games Two-person zero-sum games: Mixed strategy
B3
A1 2  A mixed strategy problem is one where players change
from alternative to alternative when the game is
A2 4
repeated.
 A mixed strategy problem does not yield a saddle point.
For player A, strategy A1 is dominated by the strategy A2.
Eliminating the strategy A1 we thus see that A should always
play A2 and B always B3 and the value of the game is 4 as
before.

29 30

Mixed Strategy Problems Two-persons game

Assumptions: Example 5
 The pplayers
y ppractice a maximum secrecyy with their pplans so Consider the game of the odds and evens.
e ens This game consists of
that the opponent will not guess their move
two players A,B, each player simultaneously showing either of
 The average payoff is determined by the fraction of the time
one finger or two fingers. If the number of fingers matches, so
that each of the alternatives is played and there is a certain
that the total number for both players is even, then the player
fraction that is best for each player.
taking evens (say A) wins $1 from B (the player taking odds).
 The best strategy for a mixed strategy game is a random
Else,, if the number does not match,, A pays
p y $$1 to B.
selection of alternatives which conform in the long run to
predetermined proportions.

31 32

8
 Two-persons game
Game: odd or even
 The payoff matrix to Player A  The maximin and the minimax values of the games are
- $1 and $1,, respectively
p y
 maximin value ≤ value of the game ≤ minmax value
B1 B2
 unstable games where such a saddle point does not exist.
A1
1 -1  Because the two values are not equal, the game does not
have a pure strategy solution. (Both players can randomly
A2 -11 1 mix their respective strategies).

33 34

Two-Person, Zero-Sum Game Mixed Strategy

Two-Person, Zero-Sum Game Mixed Strategy
 Consider again the labor dispute problem but replacing
 Interpretation: From the above game matrix,
(U3,C3) value by 1.9
 If the
h Union
U i adopts
d U1,
U1 the
h Company
C will
ill adopt
d C3 C3.
 If the Company adopts C3, the Union will adopt U3.
 If the Union adopts U3, the Company will adopt C2.
 If the Company adopts C2, the Union will adopt U1.
 The shift from alternative to alternative becomes a cycle

 (U1 and C2) is not an equilibrium or saddle point

35 36

9
 Mixed strategy Games with mixed strategies
 Read materials provided  It is necessary to choose among alternative acceptable
strategies on some kind of random basis.
 Whenever a game does not possess a saddle point, game
theory advises each player to assign a probability distribution
over his/her set of strategies.
 xi = probability (proportion of the time) that player 1 will use
strategy i (i = 1, 2, . . . , m)
 yj = probability that player 2 will use strategy j ( j = 1, 2, . . . , n)
 Probabilities need to be nonnegative and add to 1.
 These plans (x1, x2, . . . , xm) and (y1, y2, . . . , yn) are usually
referred to as mixed strategies, original strategy
37 38

Mixed strategies Mixed Strategy

 The solution of the mixed strategy problem is based also on  Applying minimax theorem of game theory
the minimax criterion.  Player
y A selects xi
 Player I selects the ratios xi (instead of the pure strategies i) which
maximize the minimum expected payoff in a column, while II that will yield
selects the ratios yj (instead of pure strategies j) which minimize
the maximum expected payoff in a row.

39 40

10
 Mixed Strategy Solving a Game
 Player B then selects yi  Finding the optimal mixed strategies for both players in a
two-person zero-sum game is called solving a game
that will yield  An Optimal Mixed Strategy is a mixed strategy that
minimizes the potential loss against the opponent’s best
counter-strategy
 If xi* and yj* are the optimal solutions for both players, then
 No matter what a zero-sum game being played, there is
the optimal expected value of the game is: at least one optimal mixed strategy for each player.

41 42

Solving a Game Games 2x2 Matrix

 There are several methods for solving this type of game.  Consider the following (2 x 2) game in which we assume there is
 Algebraic
g Method no saddle point
 Graphical method
 Linear programming approach
 It is important to first use the principle of dominance to
be able to reduce the total number of alternatives.  Player I's expected payoff corresponding to the pure strategies
 The above non-linear optimization
p model is convertible of II are given by :
to a Linear Programming Model.

43 44

11
 Games 2x2 Matrix mixed strategies
 Since the optimal x1* and x2* have been chosen to make I's  Solution of (mxn) Games by Linear Programming
mixture of moves optimal against any of the two possible II's  The optimal
p strategy
gy for pplayer
y A
moves, the two expected payoff for I must be equal if x1* and x2*
are optimal. Hence we have

 A similar analysis for II leads to

 This model can be converted to linear programming using

the following: Let
 x1* and y1* can be determined as well x2* and y2*

45 46

mixed strategies mixed strategies

 The LP model is given by:  LP model for player A

 Using the same principle, LP model for player B

 Assuming that v>0, we divide all constraints by v and let

Xi=xi/v. since max v is the same as min 1/v, the model for
player A becomes:
47 48

12
 Linear Programming Method Exercise
 Steps in Linear programming approach  Verify that the solutions to the following (2 x 4) game, are:
 Step 1: (Optional but highly recommended)  x1* = x2* =1/2
Reduce the payoff matrix by dominance  y1* and y4* =0 and y2* and y3* = 1/2 and v* =5/2
 Step 2: Add a suitable fixed number to all the entries of the
resulting payoff matrix to convert it to a non-negative entry
matrix (scale up all entries by adding a fixed number T)
 Step 3: Solve the associated Standard LP
 Step 4: Calculate the optimal strategies
 Step 5: Value the game
 Hint: Reduce the payoff matrix by the dominance
principles
49 50  Note that strategy 1 of II is dominated by strategy 2

Prisoner’s Dilemma Prisoner’s Dilemma

 Two members of a criminal gang are arrested and imprisoned.  The district attorney offers each the chance to turn state’s
 They are placed under solitary confinement and have no evidence (confess).
chance of communicating with each other.  If only one prisoner turns state’s evidence and testifies against
 The district attorney would like to charge them with a recent his partner he will go free while the other will receive a 10
major crime but has insufficient evidence. year sentence.
 He has sufficient evidence to convict each of them of lesser  Each prisoner knows the other has the same offer
charge.
 The catch is that if both turn state’s
state s evidence,
evidence they each
 If he obtains a confession from one or both the criminals, he
receive a 5 year sentence
can convict either or both on the major charge.
 If both refuse, each will be imprisoned for 2 year on the lesser
charge
51 52

13
 Prisoner’s Dilemma Prisoner’s Dilemma
.
What would you do if you were Prisoner A?–  In such games, one player’s gain is not equal to the other’s loss.
This is not a constant-sum two-player
p y game
g  Prisoner A thinks: If the other prisoner don’t confess then I am
better off if I confess. If B confess, I am also better off if I confess.
Prisoner B  Prisoner B thinks similarly.

Confess Don’t  For each prisoner the “confess” strategy dominates the “don’t
Confess confess” strategy. Each prisoner seeks to eliminate any dominated
strategies from consideration.
Confess ((5,5)) ((0, 10))  Individually,
Individually there is a strong temptation to defect
defect.
Prisoner A Don’t Confess (10,0) (2, 2)

53 54

Prisoner’s Dilemma
 Clearly the best thing for the two as a group is to
p
cooperate with each other. Because there is no
communication and no mutual trust, the rational
prisoners obtain outcomes that are worst off than if they
had cooperated.

55

14