GAME THOERY

QUANTITATIVE ECONOMICS
HOTELLING LOCATION MODEL GAME THOERY STUDY NOTES

KEYWORDS *
FIRST SITUATION:
What is hotelling location
model? A gets a large payoff (everyone to her left + half the people between A and B
Two (or more) enterprises
choosing a location for their B gets half the people between A and B + everyone to her right
business
NOTE: fixing market prices IF B were to move LEFT (A remains fixed) he will increase his profits
AND sell perfect substitutes
Nash equilibrium:
a set of strategies such that
no player has incentive to
unilaterally deviate i.e.,
conditional on what everyone A B
else does, no one wants to
change
Equilibrium: SECOND SITUATION:
people’s choices are
optimal simultaneously A has the entire beach
given their context,
including the behaviours
of other people BUT IF B were to jump over A, B will get almost the entire beach
when all players are
choosing optimal THIS WILL REPEAT UNTIL BOTH PLAYERS ARE IN THE MIDDLE (NASH EQUILBIRUM *)
strategies at the same
time, where an optimal When nobody can improve their payoff
strategy is the strategy
that will give the
relevant player the
highest possible payoff
in a game given the
strategies of the other
players. AB
SIDE NOTE:
no Nash equilibrium for n = 3

BA
NASH EQUILBIRUM*:

AB
SEQUENTIAL-MOVE GAME: GAME THOERY STUDY NOTES

KEYWORDS * To analyse a sequential-move game, USE BACKWARD INDUCTION*

What is a sequential-move
game?
Players take turns choosing Payoff to
their actions i.e., player one Payoff to
second
moves, then player two moves first player
(don’t move at same time) player
- Represented by a
game tree (extensive
form)
The extensive form game
specifies
1. The players in the game
2. When each player has a
move 1. How many possible strategies does B
3. What each player can do have?
at each opportunity - Must be a complete
move strategy*
4. What each player knows
at each opportunity - List all strategies for player
move B (e.g AXA or AYV)
5. Payoff received - Number of strategies = [all
Backward induction : the complete
process of working backwards strategies] i.e., multiply ALL
to ensure that sequential choices at each choice node
rationality is satisfied (start
at end node to route node) e.g. (4 ´ 2 ´ 3)
Complete strategy:
A choice at every node for
that player
ALWAYS give the solution EXPECTED PAYOFF’S/ NATURE/ LOTTERY
to a game as a full
strategy for each player, Find the expected utility/payoff for nature*
i.e a strategy profile/
joint strategy
(weight/probability ´ outcome utility)
FOR EXAMPLE: if person 1
chooses strategy G and e.g 0.7(-2) + 0.3 (2)
person 2 chooses strategy H,
then the chosen strategy bigger prob (e.g 0.7) will result in the
profile is (G, H)
NATURE: outcome being closer to utility associated
Nature isn’t a player, doesn’t
have preferences, doesn’t with that probability (e.g -2)
make choices, doesn’t have
strategies
- Probabilities MUST
ADD UP TO 1
- If the outcome of
the lottery is the
same as other
possible outcomes,
plays have 2 possible
solutions (list both +
mark both branches)

In summary
- All players MUST act rationally + know the rules of the game
Rational behavior means that a person acts to obtain the best possible outcome for themselves
given the circumstances
Rules of the game are precise descriptions of a strategic interaction
It includes
1. Who the players are
2. What actions they can take at all different point
3. How payoffs are assigned
4. What all players know
\ players have FULL UNDERSTANDING of the game
SIMULTANEOUS MOVE GAMES GAME THOERY STUDY NOTES

KEYWORDS * Simultaneous move games have:

What is a simultaneous move
game? 1. Imperfect information
A players move is not
conditional upon other players If something happens that is payoff-relevant for some player, but the player cannot observe it when making
moves/ players choose at the
same time her choice (Doesn’t mean games have incomplete information)
What is a prisoner’s dilemma?
The category of games in 2. Strategic uncertainty
which players gave a
dominant strategy to “cheat”,
preventing beneficial Strategic uncertainty is if the other player’s strategy is not observed when a player makes her choice
cooperation from occurring
(inferior outcome/ poor 3. Pure strategy
payoff/ Lower utility)
LESSON:
Self-interested 4. Discrete strategies
behaviour in games
may not lead to Can count them (not infinite) (can put them on a table)
socially optimal
outcomes EXAMPLES OF SIMULATNOUS MOVE GAMES
Strategic form = a matrix,
used for simultaneous games 1. Hotelling location model
(no order of moving, move at
the same time)
2. Prisoners’ dilemma*
Require complete and
imperfect information +
common knowledge
Common knowledge = both
players know each player is 1. Hotelling location model- discussed on page 1
rational + knows the rules of
the game
2. Prisoners’ dilemma
EXTRA: This is the strategic
Rational behavior means that form*: Examples: whether humans can cooperate in
a person accts to obtain the situations where they must give up certain benefits to
best possible outcomes for themselves to benefit other people]
herself given the
circumstances
1. negative externalities (pollution, global
Strategic form game specifies
- The players in the game warming)
- The strategies available
to each player 2. Public goods
- Payoff received

DOMINAT STRATEGY:
[Strategy that yields a higher payoff than alternatives, regardless of the strategy choice by another player]

Strictly dominant: Weakly dominant

Leads to a better outcome leads to equal or better outcomes
than alternatives than alternatives

NB: A dominant strategy equilibrium is when ALL players play their dominant strategy

DOMINANT strategy example: NOTES:

- Calvin has a dominant strategy Confess- it is
strictly dominant as the payoffs are not equal
\If rational will choose dominant strategy
- Klein stickly dominate strategy= confess
(rational choice)

n \dominate strategy equilibrium

Dominate means it MUST IN ALL circumstances give a
HIGHER payoff
SIMULTANEOUS MOVE GAMES GAME THOERY STUDY NOTES

KEYWORDS * DOMINATED STRATEGY:

A dominated strategy is one that always gives a worse outcome than some alternative strategy (the dominating strategy) regardless of the strategy choice
NOTE : of the others
If strictly dominant , it is also
weakly dominant
Because: leads to better/equal strictly (or strongly) dominated*: weakly dominated*:
outcomes than alternatives
A weakly dominant strategy leads to worse outcomes than alternative if it leads to equal or worse outcomes
MAY also be strictly dominant than alternative strategies
BUT NOT NECESSARILY strategies
SUBGAMES

A subgame is any part of a SUBGAMES*

tree that begins at a decision
node and contains everything To identify if it is a SPNE*, first, use backward induction and mark off branches where
that “follows” that node
NOTE: only fits with perfect outcome is maximized. List the SPNE of all possible maximized outcomes at each different
information
Subgame perfect Nash subgame
equilibrium
SPNE requires rationality in all
subgames, including off- NOTE: (D, HH) is not a SPNE
equilibrium path ones (B, LH) is a SPNE

NB: all SPNE are NE, BUT not all

NE are SPNE

How to draw a strategic-form representation of a game when it is illustrated in extensive

form:

Identify player A and player B’s full strategies (i.e every possible choice at each players
choice node i.e decision nodes) + list player 1 strategies in rows and player 2’s strategies in
columns

To correctly associate payoff in the matrix: TIP use first letter of the row with last letter
of the columns

When asked to identify SPNE:

Use backward induction, \ mark all branches. A NE is an outcome/branch marked from
backward induction. A SPNE looks at each decision node

TYPES OF STATEGIC FORM GAMES 4. Assurance/ stag hunt

1. Prisoners’ dilemma

5. Battle of the sexes

2. Pure coordination

3. Chicken game 6. Matching pennies

SIMULTANEOUS MOVE GAMES GAME THOERY STUDY NOTES

KEYWORDS * WORKED EXAMPLE FROM TUT 2

IESDS:
iterated elimination of strictly
dominated strategies R1 (R2)
C1(C2)
R2 (R3)
What is IESDS? C2(C3)
A solution (strategy profile) IEDS solution: R3, C3
obtained by iterated
elimination of strictly
dominated strategies (IESDS)]
BEST REPSONSE FUNCTIONS AND CURVES GAME THOERY STUDY NOTES

KEYWORDS * WORKED EXAMPLE:

What is a duopoly? Suppose we have two firms, x and y, each with demand equations and cost per unit:
a market situation in which
two companies exclusively
provide a particular product
or service
Qx=44−2"X+"Y Qy=44−2"Y+"X #X=8 ; #Y=8
What is an oligopoly?
while oligopoly is an economic
condition in which a small
number of sellers exert
control over the market of a First construct a profit function for firm x:
commodity
What does taking a partial πx= TRx -TCx OR πx= (QxPx) – (QxCx) OR πx= (Px - Cx)Qx
derivative mean?
It means we take the other
variables (!" in this case) as Qx(Cx)
given , i.e we do not consider Qx(Px)
that they may change and
therefore treat them like
constants = (Px- 8) (44−2Px+Py)
If demand functions are
= 44Px - 2Px2 + PxPy - 8·44 + 16PX - 8Py
symmetrical – Best response
function for firm x and firm y
= - 2Px2 + 60PX + PxPy - 352 - 8Py
are equal

The point of intersection = To maximize profits, given the other firm’s price. Take the partial derivative of profit with respect to Px
Unique Nash equilibrium and Py and set this equal to zero
If a firm sells homogenous ∂πx ∂πx
goods = symmetrical cost =0 =0
structure = equal best ∂Px ∂Py
response function
\ -4Px + 60Px + Py c c
∂πx
\ = 0Px = 4 + 4 Py ∂πx = 0
60 1
∂Px ∂Px
1
= 4 Py+ 15 ¬ This is firm x’s best response function and it is vital that it
be expressed with Px on the left hand side
1
Py = 4
Px + 15 ¬ This is firm y’s best response function

Solve simultaneously

GRAPH OF BEST RESPONSE RULES

more
profit
BEST REPSONSE FUNCTIONS AND CURVES GAME THOERY STUDY NOTES

KEYWORDS * If the firms acted as a cartel, the cartel would set both Px and Py in order to maximize joint (or
What does cartel mean? total) profit. Total profit is given by:
Two or more independent
firms agree to manipulate the
price of a product/service (i.e
easy way to make a profit)
Π = πx + πy
= - 2Px2 - 2Py2 + 52Px - 52Py- 2Px2 + 2PxPy – 704
- Discourages innovation +
efficiency gains MAXIMISE ABOVE FUNTION: by getting partial derivates with respect to x and y and set
equal to 0
Cartel solution when drawn in
a 2 X 2 matrix= a prisoners’
dilemma ∂Π ∂Π
=0 =0
∂Px ∂Py

∂πx \ -4Px + 52Px + 2Pcy \ cx

- 4Py + 52Px + 2P
=0 ∂πx 52 1
∂Px ∂Px
=0 \ Py = + Px
4 2
AND Px = 26
SUBSTITIUE:
\ PY = 26
DECISION MAKING UNDER UNCERTAINTY GAME THOERY STUDY NOTES

KEYWORDS * Expected utility

we model uncertainty using The expected utility (EU*) approach considers the agents subjective value of an uncertain
lotteries, where there is a
distinct number of outcomes
with an associated probability situation to be a probability-weighted average (i.e [statistical expectation] of the utility of
\a lottery is a probability the possible outcomes) EU(L)= 0.8u(A) + 0.2u(B)
distribution (probabilities may be
subjective) FOR EXAMPLE:
Important properties of weighted averages
EU allows utility comparisons EU(L)= 0.8u(A) + 0.2u(B)
between lotteries and fixed 1. EU(L) lies between u(A) and u(b)
outcomes, and between
lotteries and other lotteries (L refers to the lottery) 2. EU(L) lies closer to u(A) and u(b)

gives a solution to EXAMPLE:

decision problems EU(L)= 10p + 3(1-p)
namely the choice
leading to the highest NOTE: if p is high enough EU(L) will be greater
possible EU than the outcome R U(R)

Two requirements for EU Two different ways a games rules can be

1. Decision must be changed
rational 1. Make probability of p bigger
2. Decision must reflect 2. Play around with payoff numbers
agents’ subjective a. i.e make 7 smaller relative to
beliefs 10 & 3
EMV is a fair value of the
lottery OR the average EXPECTED MONETARY VALUE (EMV*)® money amount
monetary return of the lottery
EMV(L)= (prob) (outcome1) + (prob) (outcome2) = R(amount)
Utility
Is a measure of value
Risk loving:
EXPECTED UTLITY® lottery
Risk loving
A person willing to take more EU(L) = (probX)· u (RX) + (probY)· u (RY) =
risks
Risk adverse:
Avoiding as much risk as
possible Certainty equivalent* (CE) ® money amount
Risk neutral
CE=EMV AND EU(L) = u(EMV(L)) U(CE) = EU(L) NOTE: Always indifferent between CE and the lottery
Risk loving* Risk averse*
A
r EU(L) > u(EMV(L)) OR CE(L) > EMV(L) EU(L) < u(EMV(L)) OR CE(L) < EMV(L)
i
s In words: compare the value the person places on a money In words: compare the value the person places on a money
k
l amount or taking a lottery. If someone puts a high utility amount or taking a lottery. If someone puts a high utility
o
v (i.e prefers) to take the lottery when given the choice. The (i.e prefers) to take the money amount over taking a
i
n person is risk loving lottery (i.e taking risk). The person is risk adverse
g
p
e
r
RISK LOVING
s
o RISK ADVERSE
n
i
s
w
i
l
l
i
n
g
t
o
t
a
k
e
MIXED STRATEGIES GAME THOERY STUDY NOTES

KEYWORDS * Consider the following game:

FISRT STEP

Say: Let p be the probability of Evert playing DL and let q be the probability of Navratilova playing
DL
SECOND STEP: find the values of p and q
To find the probability of p, set expected payoff of player 1 = expected payoff of player 2
For Navatilova
πNAV (p, DL) = πNAV (p, CC) side note: NAV is player 2 \ look at columns for DL & CC

50p + 10(1-p) = 20p + 80 (1-p)

\p= 0.7
For Evert,
πEV (q, DL) = πEV (q, CC) side note: EV is player 1 \ look at rows for DL & CC

50q + 80(1-q) = 90q + 20(1-q)

\q= 0.6
THIRD STEP:
draw best response graph
Expected utility/payoff

When p = 0
NAV EU = (0)50 + 10- 10(0)

P value bounded from 0 -1

Want to know for each value of p,
what is the BR for NAV
MIXED STRATEGIES GAME THOERY STUDY NOTES

How to calculate expected payoffs when both players are using mixed strategies
The expected payoffs for the players in the mixed strategy NE can be calculated using any one of
these methods (using Evert as an example):
1. First calculate the expected payoff for each of Evert’s pure strategies, considering that
Navratilova’s mixed strategy creates a lottery for Evert if Evert plays a pure strategy:
πEV (DL, q) = 50q + 80(1-q)

πEV (CC, q) = 90q + 20(1-q)

Then apply another expected payoff calculation to account for the lottery that Evert creates with
her p-mix:

2. Π!"#$%(!, ") = ! ⋅ Π!"#$%($%, ") + (1 − !) ⋅ Π!"#$%('', ")

= p(50" + 80(1 − ")) + (1 − !)(90" + 20(1 − "))

Substitute the NE p and q values to calculate the expected payoff for Evert (answers found in first
page) \ p= 0.7 and q= 0.6

BEST RESPONSE RULES:

Navratilova’s best response (BR)

rules:
BEST RESPONSE CURVE
If p < 0.7, then play CC (q=0)

If p > 0.7, then play DL (q=1)

If p = 0.7, then play CC or DL

or any mix (0 ≤ q ≤ 1)

Evert’s best response (BR) rules:

If q < 0.6, then play DL (p=1)
If q > 0.6, then play CC (p=0)
If q = 0.6, then play CC or DL
or any mix (0 ≤ p ≤ 1)
FINITE REPETITION GAME THOERY STUDY NOTES

KEYWORDS * CONTINGENT STRATGIES

NOTE: With finite re- 1. Trigger strategies:
petition, backwards
induction is impossible Player cooperates (C) as long as other player does. Any D leads to a period of
Punishment
TIME PREFERENCE 2. Crim strategy
PRESENT VALUE FORMULA
1 C if another player has never chosen D
PV= FV• Otherwise, D
(1+r)n Discount factor 3. Tit for Tat (TFT)
Discount rate
Big r- more you care First move: C
about present rather
than future payoffs/
Afterwards: what another player did previous round *
The bigger the r, the
less u care about the
future also known as
impatient people Either a Tit for Tat or grim strategy
Time preferences are
SUBJECTIVE \ r is a way
to summarise those
preferences
Steam of equal sized-future
payments
1
PV=
r
MUST discount each payoff
as it happens in NEXT
period
NOTE: DO NOT discount first
period
IF loss > gain
Deducted 140 from all payoffs
If gain > loss \x strategy WHY?
TFT is not a sufficient
incentive for x to keep - because it makes (C, C) = 0
pricing high \ not a NE NOTE: It does not change preferences
when both payers play TFT
It is useful because:

- It Can see if players is better or worse by

changing thier strategy

EXAMPLE FROM TUT BOTH COOPERATE \ BOTH CHOOSE HIGH

ROUND B A GAINS/LOSSES

1 L H + 250

2 H L -275/ (1+r) = 229.16

275/ (1 + r) > 250
3 H L 0 r < 0.1
*Note* SIGN CHANGES
KEYWORDS * EVOLUTIONARY GAME THEORY

Evolutionary model gives explanation for Problems with classical game theory:
boundedly rational behavior changing &
adapting over time 1. Multiple Nash equilibriums : we cannot give a clear prediction of what will happen, this can cause problems between
2 players as they can’t predict what will happen
Evolution principles
à evolution: Change of population
2. Unreasonable Nash equilibriums
characteristic over a long period of time 3. Assumptions about rationality : rationality we expect in normal game theory is extreme & unrealistic \not
(not a change in an individual) applicable to real world
à population: distribution of (pheno- 4. Agnostic regarding the process of rationality & movement to equilibrium: don’t discuss how people get to
)types
a NE, we want to discuss if and when people get to NE’s. People can learn to become rational \ more realistic to the real
Side note:
Phenotype: is a set of observable characteristics world
or traits of an organism
Dynamics of evolution: Evolution as a metaphor:
As a metaphor, evolution in game theory is about behaviours/characteristics that gets reproduced over time
Human behaviour/change behaviours
- Firms with low-payoff strategies become insolvent and are replaces by better payoffs
- Bad strategies will die off \ good strategies will survive and pro-create and others will start copying it

Evolutionary framework:
1. Each participant/individual has a specific type which is the strategy that the participant will play i.e type= strategy
à Selection 2. Symmetrical games
Fitness: survival of the form 3. No rationality assumed but is an outcome of the game
(phenotypic or genotypic that will leave 4. Fitness: how well you fit in the environment i.e population or is the frequency distribution of types (expected payoff)
copies of itself in successive
generations [ i.e well fitted/suited to the 5. Selection: The frequency of strategies with high payoff’s increase a& the frequency of strategies with low payoffs
environment, will not survive] decrease
1. Natural selection 6. Mutants: potential strategies may enter the population
2. Sexual selection: reproduce
3. Artificial selection:
à Mutation
Creates variations or introducing EXAMPLE TEBOGO
x:
novelties VIDA M&B ÕVida = 2 (1 - x)+ 0x frequency (i.e percentage) of
à Perturbations population following y
Changes in environment i.e climate change VIDA 2, 2 0,0 strategy & X describes
can disturb an equilibrium and make LISA M&B 0, 0 1, 1 ÕM&B = 0 (1 - x)+ 1x population’s start at a point
population more resilient in time
Side note: phenotypic refers to physical
characteristics (behavior) in economics we refer
to it as the strategy & genotypic refers to
generic characteristics
INTERPRETATION OF A GRAPH
dynamic equilibrium is when the
endogenous variables change at the same
rate over time X= 2/3 is the Nash equilibrium
It is unstable { tipping point } because if we move to the
Dynamic equilibria/rest points (where x left Vida becomes more favourable and tends to 0 i.e if you
has no tendency to change) = Nash equilibriums push it all the way to the right it will go to 1
BUT Not ALL dynamic equilibriums are
stable when there are perturbations X=0 & x=1 stable equilibria
dx Large BASIN OF ATTRACTION
ESS: evolutionary stable strategies >0 Selection: x increases over time if: ESS
dt
Strategy is impermeable when adopted by
a population
No evolutionary pressure to the L or R i.e equally happy
KEYWORDS * REPEATED GAMES & THE PRISONER’S DILEMMA
Oligopoly game
- Higher payoff= higher profit WHat we already know about prisoner’s dilemma
- Payoffs are money amounts [firms - Two prisoners
are risk neutral] - Can confess OR remain silent
- NE: deflect BUT both be better off if - Dominant strategy to confess {fewer years in jail irrespective of other player}
choose to cooperate - NE for both to confess = poor outcome
- Rational action leads to a bad outcome {because they cannot make a binding agreement/act independently (look
Sucker payoff at own payoffs)}
Worst payoff, you cooperate and other - However, this model can get a cooperative outcome {when players play the game repeatedly against each other}
player deflects - When a player changes from one strategy to the other, they get a better payoff i.e the payoff of this player is
Temptation much less than the cost of the other {both confess and are worse off than they were when cooperating}
Highest payoff, tempted to break
agreement i.e high payoff if you deviate
from cooperation Prisoners’ dilemma as a model for conflict + cooperation
Sucker payoff
- Conflict: dominant strategy implies worse outcome for other player
La Frasca - Cooperation: possibility of better outcome
LOW HIGH - not a zero-sum game ! Ground for an agreement

massimos LOW 90,90 180,70 - many situations look like:

1. Oligopolies:
HIGH 70,180 140,140 2. Arms race:
3. Public goods:
4. Common resources:
temptation 5. Contracts:
Standard terms used:
Deflect/ cooperate ASSUMPTIONS OF REPEATED GAMES
Arms race: - Player plays the game multiple rounds against the same player.
Two countries each have a choice between - Rationality: people only cooperate when it’s rational.
high or low level of arms. - Simultaneous moves – not possible to have strategy like “do what the other player does in this round”. You can “do
Most preferred outcome is when what the other player did last round”.
country A chooses high, and their rival
chooses low = a clear military advantage - Communication is non-credible à no matter communication, they will play rational choice.
NE = both agree to not set up arms - No manipulation of rules or payoffs possible.
Public goods:
everyone makes a contribution to a good CONCEPTS OF REPEATED GAMES
where benefit is divided BUT everyone has - Stage game = once-off round of a PD game within the larger game of repeated PD’s.
an incentive to free-ride and benefit from - Repeated games’ payments are summed (over a period of time).
others contributions - But provision can be made for time preferences.
NE= no one contributes
Common resource: Time preferences OF OF REPEATED GAMES
private incentive to overexploit a resource - R10 in future is worth less than R10 now: discounting.
everyone is worse off because there e.g. R1 today can be invested to yield (1+r) rand one year from now.
are no resources left 1
PV= Discount rate
Contracts: (1+r)
requires an exchange
cooperation is necessary
KEYWORDS * REPEATED GAMES & THE PRISONER’S DILEMMA

Conditional strategy:
Must provide an incentive/ be worthwhile
for the other players to cooperate Cooperation in the PRISONERS DILEMMA
- Possibility of cooperation only if C (cooperate) now will lead to the other player playing C later .
Finite repetition - Therefore, give up a ST immediate benefit in exchange for a larger LT benefit.
Game is being played a specific (and - Large discount rate (r) or an interest rate = { big difference between future and current amounts of money. Care more
finite) number of rounds\ game ends
after certain amount of round have been about present rather than future payoffs \ impatient people- impatient person wants instant gratification}
played {backward induction} - Small discount rate (r)= {small difference between future and current amounts of money. Care more about future
rather than present value payoffs\ patient people}
infinite repetition - Possibility of cooperation depends on people’s time preferences
Game is being played a infinite number of - Players’ actions must therefore discriminate based on other players’ previous actions
times- players do not know for how many - Not discriminate à choosing C/D irrespective of what the other player did in the past à what is the other player
rounds the game is being {backward
induction is impossible} [because there is no going to do? He/she has no reason whatsoever to cooperate.
“last round”] - If you choose a conditional strategy, you have to end up playing C most of the time à if not, you’re not creating an
- Larger number of games = more incentive for the other to cooperate
complicated

FINITE REPITITION
Cooperation in Prisoner’s dilemma second round last round
First round
La Frasca La Frasca
Possibility of cooperation only if C now will La Frasca LOW HIGH
lead to the other player playing C later LOW HIGH
LOW HIGH
Therefore, give up a short-term immediate LOW 90,90 180,70 LOW 90,90 180,70
benefit in exchange for a larger longer-term LOW 90,90 180,70 massimos HIGH 70,180
benefit massimos HIGH 140,140
massimos HIGH 70,180 140,140
Players’ actions must therefore discriminate 70,180 140,140
on the basis of other players’ previous Second round: both players
Last round: both players (knowing that
actions First round: same thing à know that they will both defect
there are no more future rounds) will
Strategy to so discriminate and to play C play their dominant strategies à
whatever they do, it won’t in the last round à no reason
“usually” (to elicit positive discrimination) affect what happens going to cooperate in second round, defect strategies à both end up with
must be best response in equilibrium forward, so they will play because it doesn’t affect what poor playoffs
their dominant strategy happens in the last round à {No reason for conditional strategy in
defect last round because they’ve got an
unconditional dominant strategy}
No cooperation is possible !!

Not possible when a game has a DEFINITE END

Infinite repetition
La Frasca
LOW HIGH Cooperation is possible
massimos LOW 90,90 180,70
\ have to care about CURRENT payoffs and what you know will happen in the FUTURE
HIGH 70,180 140,140

….. ¥
KEYWORDS * REPEATED GAMES & THE PRISONER’S DILEMMA

Conditional strategy:
Must provide an incentive/ be worthwhile CONTINGENT STRATAGIES
for the other players to cooperate
Trigger strategies:
Finite repetition {Player cooperates (C) as long as the other player does. Any D leads to period of punishment}
Game is being played a specific (and
finite) number of rounds\ game ends
after certain amount of round have been 1. Grim strategy:
played {backward induction} C if the other player has never chosen D, otherwise D forever.
2. Tit for tat (TFT):
infinite repetition
First move: C
Game is being played a infinite number of Afterwards: what other player did previous round.
times- players do not know for how many P2 P2
rounds the game is being {backward DEFECT COOPERATE DEFECT COOPERATE
induction is impossible} [because there is no DEFECT
“last round”] DEFECT 90,90 180,70 -50,-50 40,70
- Larger number of games = more P1 COOPERATE 70,180 P1 COOPERATE -70,40 0,0
complicated 140,140
practice
Normalized version of
Is a strategy that cooperates a best response when the other player is playing TFT? the game the players
Two possible deviations [More complicated deviations will effectively be the same as one of these since TFT has only one round of memory] preferences are
unchanged

I
1. To play D for one round for temporary gain
2. To play D always
*NB*
Is it worthwhile to play D only once against a rival playing TFT?
{Depends on the Present Value (PV) - compare value of the gain with the PV of the loss}
Yes if:
70
40> \r>0.75 TFT in equilibrium
(1+r)

If ! > 0.75, then D once is beneficial relative to C always • If neither deviation from full cooperation
against TFT is worthwhile, then cooperation
*NB* is a best response
Is it worthwhile to play [D always] against a rival playing TFT? • In the example, if ! < 0.75
• TFT itself will then also be a best response
• Gain: 40 in first round because TFT will cooperate always if the
• Loss of 50 in every round from round 2 onwards rival is playing TFT
• If TFT is a best response for both players
when the other is playing TFT, then (TFT,
ONLY Worthwhile if: TFT) is a NE
• This shows that it is possible for rational
50 50 50 50 50 players to cooperate in a PD game!
40> + 2 + 3 + 4 +… = \ r > 1.25
(1+r) (1+r) (1+r) (1+r) r
KEYWORDS * EXPERIMENTAL GAME THEORY

Normative behavior: WHAT IS THE POINT OF GAME THEORY?

How people ought to behave - what rational people - Just an exercise in pure math’s
ought to do when they play against other rational - Study real people in real situations to understand their behavior Theories are simplified models \ not
people {sets a benchmark for how these perfect a. Normative behavior perfect representations of real world
players should behave} b. Descriptive behavior
descriptive behavior:
should be able to describe what happens in real Centipede game
world using theory – explain what real people do
{implies we should be able to predict what people
do}

Experimental game theory:

Assumes people behave in labs the way they would
behave in the real world

- Two players taking turns moving

- At every choice, the player can choose to continue/stop the game
- Continues up to a certain end
- Backwards induction to see what a rational player should do:
o Last round: second player can continue for a payoff of 64, or stop for 128 \ P2 would stop
o If P1 knows this at the 5 th node, if she continues, she will get 32, but if she stops she would get 64 \
P1 would stop
o For P2 at the 4 th node, stop is the best option
o For P1 at the 3 rd node, stop is the best option
o For P2 at the 2 nd node, stop is the best option
o For P1 at the 1 st node, stop is the best option
- Payoff’s at the 1 st node are very low for both players compared to the others à every time the game
continues the payoffs are doubled
- However, 4 is still < 2 for P1
- People do not apply backwards induction in the real world à it’s rare that player will stop playing at the
first round. They will typically continue for a number of rounds and stop at some point = get higher payoffs
- According to GT, they are not rational. BUT are they really so stupid? Because they are walking away with
higher payoffs?
- P1 chooses continue with a degree of trust that P2 would do the same.
Why don’t people use backwards induction?
they may not be trained/good with it
KEYWORDS * EXPERIMENTAL GAME THEORY
Normative behavior:
How people ought to behave - what rational people the ultimate game:
ought to do when they play against other rational Two players:
people {sets a benchmark for how these perfect a. Proposer: has to divide a certain fixed amount of money to share to the Responder
players should behave} b. Responder: chooses to accept or reject
• If the Responder accepts, they both get their shares. If the Responder rejects, both get nothing
descriptive behavior: • Proposer should propose a share that the Responder would accept
should be able to describe what happens in real
world using theory – explain what real people do
{implies we should be able to predict what people Ultimatum minigame:
do}

Experimental game theory:

Assumes people behave in labs the way they would
behave in the real world

Backwards induction indicates that the Respondent should always accept the offer à the Proposer should then
offer R1 to get a larger payoff (rationally)

Experimental results:
- Very few Proposers make an offer of R1.
- On average, most Proposers offer between 30% and 40%, and 50% is a very common (‘modal’) offer à not
what GT expects
- Respondents often reject low offers: <20%
- Criticism: money amounts are too small and the players aren’t motivated to act rationally.
- Address problem by increasing money amounts ($100 instead of $10) or offer money in countries where
incomes are lower
- Turned out that even with higher money amounts, Responders still reject very low offers, and the Proposers
still offered about 50
 Extra Notes on Theory

The Problem of Cooperation Readings problem of when to cooperate with

them and when not to.
Column player
Thomas Hobbes: cooperation cannot
develop without a central authority. Cooperate Defect
Cooperate R = 3, R = 3, S = 0, T = 5,
Reward for Sucker ,

Row player
The problem of cooperation extends mutual and temptation
to many central issues of cooperation to defect
international politics. Defect T = 5, S = 0, P = 1, P = 1
The most important problem being Temptation Punishment for
the security dilemma to defect, mutual defection
and sucker
~ nations often seek their own payoff
security through means which Note: payoffs for the row player are
challenge the security of others. listed first
In this game, individual rationality
Related problems also occur in the leads to a worse outcome for both
form of competition with alliances than is possible, hence the dilemma.
and tariff negotiations. Two egoists playing the game once
will both choose their dominant
choice: defection, and each player
Questions to ask: How many times will will get less than if they both
you invite a friend over for dinner if cooperated.
they never invite you over in return?
If the game is played a known finite
number of times, the players will still
The Cooperation Theory: based on an have no incentive to cooperate. This
investigation of individuals who is because there is no future to
pursue their own self-interest influence.
without the aid of a central authority
to force them to cooperate with each
other.
Assumption: self-interest.
This allows for the examination of a
difficult case in which cooperation is
not completely based upon the
concern for others or upon their
welfare of the group as a whole.
NB: this assumption is much less
restrictive than it appears to be.
This assumption is really just an
assumption that concern for others
does not completely solve the
This reasoning does not apply if Why the future is less important
players will interact an indefinite than the present?
number of times, the players cannot 1. Players tend to value payoffs
be sure when the last interaction will less as the time of their
take place which means cooperation
can emerge. obtainment recedes into the
future
1. There is no mechanism 2. There is also some chance that
available to the players to the players might not meet
make enforceable threats or again.
commitments. Players cannot
commit to a particular THEREFORE, THE PAYOFF OF THE NEXT
strategy; they have to take MOVE ALWAYS COUNTS LESS THAN THE
into account all the possible PAYOFF OF THE CURRENT MOVE
strategies available to A discount parameter is used to
themselves. calculate the payoffs over time.
2. There is no way to be sure
what the other player will do A different illustration:
on a given move. This Two players interact, one plays the
eliminates the possibility of strategy of always defecting whilst
programme analysis the other plays the policy of Tit for
The only information available to the Tat.
players about each other is the Tit for Tat is the policy of
history of their interaction so far. cooperating on the first move and
3. There is no way to eliminate then doing whatever the other player
the other player or run away did on the previous move. This means
from interaction = each player that Tit for Tat will defect once
retains the ability to after each defection of the other
cooperate or defect on each player.
move. A strategy (decision rule) is a
4. There is no way to change the specification of what to do in any
other player s payoff. situation that may arise.
Therefore, words not backed by This situation depends on the history
actions are meaningless. The players of the game so far.
can only communicate to each other
through the sequence of their own Iterated Prisoner s Dilemmas are
behaviour. This is the problem of the quite different in that the interest
Prisoner s dilemma in its functional of the players are not in total
form. conflict. Both players can do well for
mutual cooperation or both can do
Cooperation can emerge if the poorly by getting the punishment.
players meet again, this means that
choices made today not only Using the assumption that the other
determine the outcome of this move, player will always make the move you
but can also influence the later fear most, where the player will
choices of the players. The future never cooperate, will lead you to
can therefore affect the current defect and this results in unending
strategic situation. punishment.
Therefore, it is not safe to assume A computer tournament: study the
that the other player is always out to effective choice in an iterated
get you in the prisoner s dilemma prisoner s dilemma.
game. Each entrant writes a program that
In fact, the strategy that works embodies a rule to select the
best depends directly on what cooperative or noncooperative choice
strategy the other player is using on each move.
and whether this strategy leaves The program also has available to the
room for the development of mutual
cooperation. history of the game so far and can
use this history when making a
This principle is based on the wight choice.
of the next move relative to the The participants also are familiar
current move being sufficiently large
enough to make the future important. with the Prisoner s Dilemma.
The discount parameter must be In the tournament, TIT FOR TAT
large enough to make the future loom came out as the best strategy.
large in the calculation of the payoff. None of the more complex programs,
Conclusions: and improvements of TIT FOR TAT
submitted could perform as well as
If the discount parameter is the original and simple TIT FOR TAT.
sufficiently high, there is no one best
strategy independent of the There is one single property that
strategy used by the other player. distinguished relatively high scoring
entries from relatively low entries
What is excluded in the Prisoner s that being: being nice , in other
Dilemma? words, never being the first to
- The possibility of verbal defect, or not the first to defect
communication before the last few moves.
- The direct influence of third There is a substantial gap in the
parties score of the nice entries and the
- The problems of implementing others.
a choice
- Uncertainty about what the What accounts for TIT FOR TAT S
other player actually did on robust success is its combination of
the preceding move. being nice, retaliatory, forgiving and
clear.
The evolution of cooperation requires
that individuals have a sufficiently Its niceness prevents it from getting
large chance to meet in the future so into unnecessary trouble. Its
that they have a stake in their retaliation discourages the other side
future interactions. from persisting whenever defection
is tried. Its forgiveness helps
restore mutual cooperation and its
clarity makes it intelligible to the
other player, eliciting long-term
cooperation.
Explanation of Constant sum: A game is constant
sum if the sum of the payoffs to
features every player are the same for every
single set of strategies. In these
Games can have several features, a games one player gains if and only if
few of the most common are listed another player loses. A constant sum
here. game can be converted into a zero-
Number of players: Each person who sum game by subtracting a fixed
makes a choice in a game or who value from all payoffs, leaving their
receives a payoff from the outcome relative order unchanged.
of those choices is a player.
Move by nature: A game includes a
Strategies per player: In a game random move by nature.
each player chooses from a set of
possible actions, known as pure
strategies. If the number is the Further concepts explained:
same for all players, it is listed here. A pure strategy provides a complete
definition of how a player will play a
Number of pure strategy Nash game. Pure strategy can be thought
equilibria: A Nash equilibrium is a set about as a plan, subject to the
observations they make during the
of strategies which represents course of the game of play. In
mutual best responses to the other particular, it determines the move a
strategies. In other words, if every player will make for any situation
player is playing their part of a Nash they could face. A player's strategy
equilibrium, no player has an set is the set of pure strategies
incentive to unilaterally change his or available to that player.
her strategy. Considering only
situations where players play a single A mixed strategy is an assignment of
strategy without randomizing (a pure a probability to each pure strategy.
strategy) a game can have any When enlisting mixed strategy, it is
number of Nash equilibria. often because the game doesn't allow
for a rational description in
specifying a pure strategy for the
Sequential game A game is sequential game. This allows for a player to
if one player performs her/his randomly select a pure strategy.
actions after another player; Since probabilities are continuous,
otherwise, the game is there are infinitely many mixed
a simultaneous move game. strategies available to a player. Since
probabilities are being assigned to
strategies for a specific player when
Perfect information: A game has discussing the payoffs of certain
perfect information if it is a scenarios, the payoff must be
sequential game and every player referred to as "expected payoff".
knows the strategies chosen by the
players who preceded them.
A totally mixed strategy is a mixed
strategy in which the player assigns
a strictly positive probability to
every pure strategy.
The best response is
the strategy (or strategies) which
produces the most
favourable outcome for a player,
taking other players' strategies as
given
A simultaneous game or static
game is a game where each player
chooses their action without
knowledge of the actions chosen by
other players. Simultaneous games
contrast with sequential games,
which are played by the players
taking turns (moves alternate
between players). In other words,
both players normally act at the
same time in a simultaneous game.
Even if the players do not act at the
same time, both players are
uninformed of each other's move
while making their decisions
A zero-sum game is a mathematical
representation of a situation in which
an advantage that is won by one of
two sides is lost by the other. If the
total gains of the participants are
added up, and the total losses are
subtracted, they will sum to zero.