6230 swann stackelberg 2010 (l5)
TRANSCRIPT
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
1/15
Recap:
Last class (January 21, 2010) Examples of games with continuous action sets
Tragedy of the commons
Duopoly models: Cournot and Bertrand
Today (January 26, 2010) Duopoly models
Comparison of duopoly models with Monopoly
1
& comparison with Cournot, Bertrand, and
Monopoly
Multistage games with observed actions
Stackelberg Model
Two competing firms, selling a homogeneous good
The mar inal cost of roducin each unit of the ood:c1 and c2
Firm 1 moves first and decides on the quantity to sell:q1
Firm 2 moves next and after seeing q1, decides on thequantity to sell: q2
Q= q1+q2 total market demand
The market rice P is determined b inverse market
2
demand: P=a-bQ if a>bQ, P=0 otherwise.
Both firms seek to maximize profits
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
2/15
Stackelberg Model
Qj: the space of feasible qjs, j=1,2
trateg es o rm :
s2: Q1 Q2(That is, the strategy of firm 2 can depend on firm 1s decision)
Strategies of firm 1: q1Q1
3
(q1
, q2
) = (q1
, s2(q1
))
j(q1, q2) = [a-b(q1+ q2)- cj] qj
Stackelberg Model: Strategy of Firm 2
Suppose firm 1 produces q1 , 22 = (P-c)q2 = [a-b(q1+ q2)]q2 c2q2
= (Residual) revenue Cost
First order conditions:
d 2/dq2= a - 2bq2 bq1 c2 =
4
q2=(a-c2)/2b q1/2= R
2(q1)
s2 = R2(q1) Strategy of firm 2
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
3/15
Stackelberg Model: Firm 1s decision
Firm 1s profits, if it produces q1 are:
= P-c = a-b + c
(Now, what is different from the Cournot game?)
We know that from the best response of Firm 2:q2=(a-c2)/2b q1/2
Substitute q into :
5
1 = [a-b(q1+ (a-c2)/2b- q1/2)]q1 c1q1
= [(a+ c2)/2-(b/2) q1-c1]q1 From FOC:
d1/dq1= (a+ c2)/2-b q1- c1 = 0
q1= (a-2c1+c2)/2b
Stackelberg Equilibrium
We have Firm 1s profits, if it produces q1:
1= - 1 2
And firm 2s best response
q2=(a-c2)/2b q1/2
Therefore:
q2=(a+2c1-3c2)/4b
6
1 2 q1= (a-c)/2b
q2= (a-c)/4b
Q = 3(a-c)/4b
Recall: qci = (a-c)/3b
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
4/15
Cournot vs. Stackelberg vs. Bertrand
Bertrand Stackelberg Cournot Monopoly
Price c (a+3c)/4 (a+2c)/3 (a+c)/2
Quantity(a-c)/b
3(a-c)/4b((a-c)/2b+(a-c)/4b)
2(a-c)/3b (a-c)/2b
Total Firm- 2 - 2 - 2
7
Profits
Example: Stackelberg Competition
P = 130-(q1+q2), so a=130, b=1
1 = 2 = =
Firm 2: q2=(a-c2)/2b q1/2 = 60 - q1/2
Firm 1:
1 = [a-b(q1+q2)]q1 c1q1 1 = [(a+c2)/2-(b/2)q1]q1c1q1 1 = [70-q1/2]q1 c1q1
8
q_1 = 60
Market price and demand
Q=90 P=40
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
5/15
Monopoly vs. Cournot vs. Bertrand vs. Stackelberg
Bertrand Stackelber Cournot Mono ol
Price10 40 50 70
Quantity120
90(60+30)
80 60
Total FirmProfits 0
27001800+900
3200 3600
9
Firm profits and prices:
Bertrand Stackelberg Cournot Monopoly
Stackelberg competitionP
130
P=130-Q
Consumer surplus=4050
Firm profits=2700
10
q
MC=10
130
40
90
Deadweight loss=450
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
6/15
Monopoly vs. Cournot vs. Bertrand vs. Stackelberg
Bertrand Stackelberg Cournot Monopoly
Consumersurplus
7200 4050 3200 1800
Deadweightloss
0 450 800 1800
11
Total FirmProfits 0
2700(1800+900)
3200 3600
Stackelberg and Information
Does player 2 do better or worse in thistac e erg game compare to t e ournot
game?
Does player 2 have more or less information inthe Stackelberg game compared to the Cournotgame?
12
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
7/15
Multi-Stage Games with Observed Actions
These games have stages such that
tage + s p aye sequent a y a ter stage
In each stage k, every player knows all theactions (including those by Nature) that weretaken at any previous stage Players can move simultaneously in each stage k
Some players may be limited to action set donothing in some stages
Each la er moves at most once within a iven
13
stage
Players payoffs are common knowledge
Stackelberg game
Stage 1
Firm 1 chooses its quantity q1; Firm 2 doesnothing
Stage 2
Firm 2, knowing q1, chooses its own quantityq2; Firm 1 does nothing
14
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
8/15
Multi-Stage Games with ObservedActions
k kh stageofstartat theHistory:
k
ki
is
hk
ihA
aaa
i specifiesthatplayerforstrategyPure:
historygivenstagein
playertoavailableactionsofSet:)(
,...,,,...,, ==
15
k
ki
h
khAa
historyeach
andeachfor)(actionan
Q: why would the strategy need to specific an action foreach stage and history?
Finite games of perfect information
A multistage game has perfect information if each la er knows all revious moves when makin
a decision
for every stage k and history hk, exactly one playerhas a nontrivial action set, and all other playershave one-element action set do nothing
In a finite game of perfect information, thenumber of stages and the number of actionsat an sta e are finite.
16
Theorem (Zermelo 1913; Kuhn 1953): Afinite game of perfect information has a pure-strategy Nash equilibrium (Note, this is a little stronger than the Nash result!)
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
9/15
Backward induction
Determine the optimal action(s) in the finalsta e K for each histor hK
For each stage j=K-1,,1
Determine the optimal action(s) in stage j foreach possible hj given the optimal actionsdetermined for stages j+1,,K.
17
The strategy profile constructed by backward
induction is a Nash Equilibrium.Each players actions are optimal at every
possible history.
Example: Stackelberg competition
P = 130-(q1+q2), c1 = c2 = c = 10
Backward induction Firm 2 strategy: s2(q1)= q2= 60 - q1/2
Firm 1 strategy: q1= 60
The outcome (60,30) is a Nash equilibrium(Stackelberg outcome)
18
Is (60,30) the unique equilibrium in this game?
Lets consider the Cournot equilibrium (40,40) forthe Stackelberg game s2(q1)= 40 q1= 40
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
10/15
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
11/15
Subgame-perfect equilibrium
Denote:
G(hk): game from stage k on with history hk
For each playerj, sj|hk is the restriction ofstrategies sj to the histories consistent with hk
A strategy profile s of a multistage game withobserved actions is a subgame-perfect equilibriumif, for every hk, the restriction s|hk is a Nash
21
equilibrium of subgame G(hk).
Classroom exercise: Strategic investment
Firm 1 does not invest
= - = -
q2=(a-c2)/2b q1/2 = 6 - q1/2 (q1, q2)=(4,4)
Payoffs: (16,16)
Firm 1 does invest q1=(a-c1)/2b q2/2 = 7 - q2/2
q2=(a-c2)/2b q1/2 = 6 - q1/2 =
22
, ,
Payoffs: (256/9-f,100/9)
Firm 1 choice:
Invest if 256/9-f>16, i.e., if f
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
12/15
Example: Stackelberg competition
P = 130-(q1+q2), c1 = c2 = c = 10
By backward induction the outcome (60,30)is a subgame-perfect equilibrium
The outcome (40,40) is NOT subgame
23
per ec , ecause e s ra egy s q1 =does not induce a Nash equilibrium in
stage 2 for player 2, for histories otherthan q1= 40
Extensive form of a game
The set of players
The order of moves
The players payoffs as a function of themoves that were made
The set of actions available to the playerswhen they move
24
ac p ayer s n orma on w en e ma eshis move
The probability distributions over anyexogenous events (Nature)
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
13/15
Example 11
U D
L R L R
2 2
2 1 0 0 - 3 2
25
, , , ,
Player 1 moves first. After observing player 1s action, player 2 moves
Player 1 action set: {U,D} Player 2 action set:{L,R}Player 1 strategies: {U,D}
Player 2 strategies: {(L,L), (L,R), (R,L), (R,R)}
(Strategies specify a complete plan of action for all contingencies)
Normal form representation of
extensive-form games
Player 2
(L,L) (L,R) (R,L) (R,R)
U 2,1 2,1 0,0 0,0
D -1,1 3,2 -1,1 3,2Player1
26
contingent plan made in advance
We usually do not use normal-formrepresentation of an extensive-form game
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
14/15
Example 2
1 2
U D
L R L R
2 2
2 1 (0,0) -1 1 3 2
L R
U D U D
1 1
2 1 -1.1 0 0 3 2
27
,
Player 1 moves first, player 2
moves next. Player 2 does not
know player 1s action when
he chooses his action
,
Player 2 moves first, player 1
moves next. Player 1 does not
know player 2s action when
he chooses his action
Classroom exercise
1
U D
(2,0)2
L R
(1,1)1
28
X Y
(3,0) (0,2)
-
8/6/2019 6230 Swann stackelberg 2010 (L5)
15/15
Example to Draw
Pla er 1 chooses an action from thefeasible set {L,R}
Player 2 observes player 1s actionand then chooses an action fromthe feasible set {L,R}
29
the history of actions is (R,R) and
then chooses an action from thefeasible set {L,R}
Example (cont.)
1
L R L R
2 2
3 3
L R
30
L R L R L R L R