two-stage games apec 8205: applied game theory fall 2007
Post on 22-Dec-2015
247 views
TRANSCRIPT
Two-Stage Games of Imperfect Information
• The dynamic games we have played so far, have been ones of perfect information.
• The games we want to look at know are dynamic games of imperfect information.– For these games, the subgame perfect equilibrium will serve us well.
• Note that the two-stage game taxonomy is not particularly standard.
Application: Bank Run Game
• Who are the players?– Two investors denoted by i = 1, 2.
• Who can do what when? – Investors choose to withdraw savings (W) or not (N) in 1st stage.
– Investors choose to withdraw savings (W) or not (N) in 2nd stage.
• Who knows what when?– Investors do not know each other’s choice in each stage.
– Stage 1 choices are reveal to each player before period 2 choices. .
How are firms rewarded based on what they do?
(R, R)
2
(2R-D, D) (D, 2R- D) (R, R)
1
W N W N
W N
(r, r)
2
(D, 2r-D) (2r-D, D)
1
W N WN
W N
Assumptions:
R > D > r > D/2
Strategies?
{W, (N,W), (N,N)}
Subgame Perfect Equilibrium
Want to start by solving for Nash in stage 2?
(R, R)
2
(2R-D, D) (D, 2R- D) (R, R)
W N W N
W N
1
Stage 2 Extensive Form Game:
Player 2
W N
Player 1
W R
R D
2R - D
N 2R - D
D R
R
Stage 2 Normal Form Game:
Assumptions:
R > D > r > D/2
* **
*
W is a dominant strategy for Player 1!
W is a dominant strategy for Player 2!
(W, W) is a unique Nash equilibrium!
Subgame Perfect Equilibrium Continued
Lets use the Nash strategy (W, W) to rewrite the game and solve for Stage 1?
(R, R)
(r, r)
2
(D, 2r-D) (2r-D, D)
1
W N WN
W N
Assumptions:
R > D > r > D/2
Revised Extensive Form Game in Stage 1:
In Normal Form:
Player 2
W N
Player 1
W r
r 2r – D
D
N D
2r – D R
R
*
*
*
*
(W, W) is a Nash equilibrium!
So is (N, N)!
There is also a mixed strategyNash equilibrium!
Application: Tariffs & Imperfect International Competition
• Who are the players?– Two countries denoted by i = 1, 2.– Each country has a government.– Each country has a firm where firms produce a homogeneous product.
• Who can do what when? – First: Government in country i sets tariff (ti) on exports from firm in country
j.
– Second: Firm in country i chooses how much to produce for domestic markets (hi) & how much to produce for export (ei)..
• Who knows what when?– Governments do not know each others tariffs or firm outputs when choosing
tariffs.– Firms know tariffs, but not each other outputs when choosing outputs.
How are governments and firms rewarded based on what they do?
• Firm i’s reward includes– Domestic Profit: (a – c – hi – ej)hi
– Export Profit: (a – c – tj – hj – ei)ei
i(ti, tj, hi, ei, hj, ej) = (a – c – hi – ej)hi + (a – c – tj – hj – ei)ei
• Government i’s reward includes: – Domestic Consumer Surplus: Qi
2/2 where Qi = hi + ej
– Domestic Firm Profits: i(ti, tj, hi, ei, hj, ej)
– Tariff Revenues: tiej
– W(ti, tj, hi, ei, hj, ej) = Qi2/2 + i(ti, tj, hi, ei, hj, ej) + tiej
Subgame Perfect Equilibrium
iijjijijjiijiieh
eehtcahehcaehehttii
,,,,,max0,0
02
jii
i ehcah
02
ijji
i ehtcae
Need to start by solving each firm’s optimal output decision.
0
ii
i hh
0ih
0
ii
i ee
0ie
First Order Conditions:
For an Interior SolutionThese First Order Conditions Imply
0*2*
0**2
0*2*
0**2
jii
ij
ijj
ji
ehtca
ehca
ehtca
ehca
3* i
i
tcah
3
2* j
i
tcae
Solving yields:
Such that:
9
2
9,
22ji
jii
tcatcatt
18
2
2
* 22ii tcaQ
3
2* ii
ji
tcatet
Now to the Government’s Optimization Problem:
3
2
9
2
918
2,max
222iijii
jiit
tcattcatcatcattW
i
0
3
4
9
22
9
2
iii
i
i tcatcatca
t
W
First Order Condition:
3*
cati
9
4*
cahi
9
*ca
ei
Such that:
But what is the socially optimal tariff scheme?
3
2
9
2
918
2
3
2
9
2
918
2,max
222
222
,
jjijj
iijiiji
tt
tcattcatcatca
tcattcatcatcattW
ji
0
9
84
3
4
9
2
9
2
iiii
i
tcatcatcatca
t
W
First Order Conditions:
Such that:
0
9
84
3
4
9
2
9
2
jjjj
j
tcatcatcatca
t
W
2*
cati
6*
cahi
3
2*
caei
Implications of Socially Optimal Policy
• Subsidize Exports
• Produce More for Export Markets & Less for Domestic Markets
• Total Output is Greater
What is going on here?
Application: Tournaments
• Who are the players?– Two Workers & Boss
• Who can do what when? – Boss determines how much to pay the most and least productive worker: wH &
wL where wH > wL.
– Workers choose how hard to work: ei for i = 1, 2.
• Who knows what when?– Boss knows output of each worker before making payment.
– Boss cannot observe effort perfectly due to random productivity shock: i with density f(i) & cumulative distribution F(i) for i = 1, 2.
• Assume E(i) = 0 for i = 1, 2 and independence of 1 & 2.
– Workers know pay schedule, but not the productivity shocks or other worker’s effort before choosing their own effort..
How are players rewarded based on what they do?
• Boss:– yi(ei) = ei + i is ith workers output
– E(y1(e1) + y2(e2) – wH – wL) = e1 + e2 – wH – wL
• Worker i:– wHPr(yi(ei) > yj(ej)) + wL(1 - Pr(yi(ei) > yj(ej))) – g(ei) for i ≠ j
• Assume g’(ei) > 0 and g’’(ei) > 0.
Subgame Perfect Equilibrium
0'Pr
ii
jjiiLH
i
i ege
eyeyww
e
g 0
ii
i ee
g
ii
jjiiLH eg
e
eyeyww '
Pr
The last stage in this game is the workers’ choices of effort.
ijjiiLjjiiHie
egeyeyweyeywgi
Pr1Prmax0
First Order Conditions:
For an Interior Solution:
0ie
Now what?
jjjiji
jjii dfeefe
eyey
Pr
jijijjiijjii eeeeeyey PrPrPrNote:
such that
Bayes Rule Implies:
jjjijjjjijijiji dfeeFeeee 1Pr|PrPr
Still, so what?
jji
jjii dfe
eyey
2Pr
ijjLH egdfww '2
The workers are identical, so why not assume they will choose the same equilibrium effort:
such that
A useful result from this equation:
0
''
2
i
jji
eg
df
wd
de where w = wH – wL
A workers effort only depends on the difference in wages.
Aside
Suppose i is normally distributed with variance 2.
jjj dedfj
2
22 2
2
2
1
jde
j
2
2
22
2
22
1
22
2
2
1
jdej
2
2
22
1
Forging Ahead, We Now Turn to the Boss
Assume workers can work for someone else earning Ua.
For the boss to get these workers to work for him, he must pay at least Ua on average:
wHPr(yi(ei) > yj(ej)) +wL (1 - Pr(yi(ei) > yj(ej))) – g(ei) ≥ Ua
But, if workers use the same effort in equilibrium:(wH + wL)/2 – g(ei) ≥ Ua
Optimization Problem for the Boss
LBww
wwwegL
2*2max0,0
subject to
(w + 2wL)/2 – g(e*(w)) ≥ Ua
a
LL Uweg
wwwwweL *
2
22*2
First Order Conditions:
0
ww
L 0*'*'2
11*'2
wewegwew
L 0w
02
Lw
L0
LL
ww
L 0Lw
0*2
2
aL Uweg
wwL
0
L
0
For an interior solution, w > 0 & wL > 0 implies L/ w = 0 & L/ wL = 0, such that = 2, g’(e*(w)) = 1, and (w + 2wL)/2 – g(e*(w)) = Ua.
We are Almost There
ijjjjLH egdfwdfww '22
jj dfw
2
1
jj
aLdf
wegUw
22
1*
Recall that
jj
aHdf
wegUw
22
1*
such that
How does this all really work out?
2w
1aL Uw
Suppose g(e) = ee where 1 > > 0 and that i is normally distributed with variance 2:
1aH Uw
g’(e*) = ee* = 1 implies e* = -ln()/
Implications
• Increasing the marginal cost of effort for a worker ()– decreases equilibrium effort.– increases the high and low equilibrium wage offered by the boss.– does not affect the difference in equilibrium wages.
• Increasing a workers opportunity cost (Ua)
– does not affect equilibrium effort.– increases the high and low equilibrium wage offered by the boss.– does not affect the difference in equilibrium wages.
• Increasing the variability of output (2)– does not affect equilibrium effort.– decreases the low equilibrium wage offered by the boss. – increases the high equilibrium wage offered by the boss.– increases the difference in equilibrium wages.
Application: Rent Seeking with Endogenous Timing
• Who are the players?– Two firms denoted by i = 1, 2 competing for a lucrative contract worth Vi.
• Who can do what when? – Stage 1: firms cast ballots to choose who leads.
– Stage 2: firms choose effort (xi for i = 1, 2).
• Who knows what when?– In 1st stage neither firm knows the other vote or effort.– In 2nd stage, firms know each others 1st stage votes:
• If both vote for Firm i in 1st stage, Firm j sees Firm i’s effort before choosing.• If both vote for different leader in 1st stage, a firm’s effort is chosen without
knowing opponent’s effort..
• How are firms rewarded based on what they do?– gi(xi, xj) = Vi xi / (xj + xj) – xi for i ≠ j.
Subgame Perfect Equilibrium
• How many subgames are there?– The whole game.
– Firm 1’s choice of effort, after Firm 2 when Firm 2 leads.
– Firm 2’s choice of effort, after Firm 1 when Firm 1 leads.
– Firm 1’s choice of effort, before Firm 2 when Firm 1 leads.
– Firm 2’s choice of effort, before Firm 1 when Firm 2 leads.
– Firm 1 and 2’s choice of effort when moving simultaneously.
• So there are lots of subgames, actually an infinite number.
We have actually seen the solution for all of these subgames except the last one previously!
Here are Those Solutions
• i Leads & j Follows– Strategies
– Rent Dissipation
– Payoffs
• Simultaneous Moves– Strategies
– Rent Dissipation
– Payoffs
otherwise
xVforxxV
V
V ijiij
j
i
0
0,
4
2
221
221
221
22
1 ,VV
VV
VV
VV
j
iij
j
i
V
VVV
V
V
4,
4
22
221
32
221
23
1 ,VV
V
VV
VV
2** i
ji
Vxx
21
2121 **
VV
VVxx
Lets Focus on the Solution to the Whole Game
Given the previous slide, the Normal form game is:
Player 2 Vote for 1 Vote for 2 Player 1
Vote for 1
2
21
12 4V
VVV
2
11 4V
VV
221
22
2VV
VV
221
21
1VV
VV
Vote for 2
221
22
2VV
VV
221
21
1VV
VV
1
22 4V
VV
1
22
21 4V
VVV
What is Firm i’s best response to Firm j voting for Firm i?
02 22 jjii VVVV
2
2
4 ji
ii
j
ii
VV
VV
V
VV
Firm i should vote for itself (Firm j) if
02 ji VV
Firm i should prefer to vote for itself if Firm j votes for i!
What is Firm i’s best response to Firm j voting for itself?
ij VV
2
22
2
2
11
4
111
1
ii
iii
i
ii VVV
i
jji
ji
ii V
VVV
VV
VV
4
2
2
2
Firm i should vote for itself (Firm j) if
Firm i should prefer to vote for itself if Firm j values winning more!
Let i = Vi/Vj, which implies
i1 or
Firm i should prefer to vote for Firm j if Firm j values winning less!
Summary of Subgame Perfect Equilibrium
• If Vi > Vj
– both firms vote for Firm j to lead.
– Firm j chooses effort first:
– Firm i chooses effort second:
– Rent Dissipation is Vj/2
i
jj V
Vx
4*
2
otherwise
xVforxxVxx jijji
ji0
0*
Implications
• Both Firms Agree About Who Should Go First
• Less Total Effort is Expended
• No Interventions Warranted
How did you do?
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
What is the subgame perfect Nash equilibrium?
How many subgames are there?
• Seven:– (1) The Game As a Whole
• (2) Player 1’s Choice After Both Players Vote For Player 1 to Lead– (6) Player 2’s Choice After Player 1
• (3) Player 2’s Choice After Both Players Vote For Player 2 to Lead– (7) Player 1’s Choice After Player 2
• (4) Both Player’s Choices After Both Players Vote for Themselves to Lead
• (5) Both Player’s Choices After Both Players Vote for Their Opponent to Lead
(7) Player 1’s Choice After Player 2
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
If Player 2 chooses L, Player 1 should choose U.
If Player 2 chooses C, Player 1 should choose D.
*
*
If Player 2 chooses R, Player 1 should choose U.
*
(3) Player 2’s Choice After Both Players Vote For Player 2 to Lead
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
Player 2 should choose C.
*
*
*
*
(6) Player 2’s Choice After Player 1
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
If Player 1 chooses U, Player 2 should choose L.
If Player 1 chooses M, Player 2 should choose R.
*
*
If Player 1 chooses D, Player 2 should choose R.
*
(2) Player 1’s Choice After Both Players Vote For Player 1 to Lead
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
Player 1 should choose M.
*
*
**
(4) or (5) Both Player’s Choices After Both Players Vote for Themselves or Their Opponent to Lead
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
If Player 1 chooses U, Player 2 should choose L.
If Player 1 chooses M, Player 2 should choose R.
*
*
If Player 1 chooses D, Player 2 should choose R.
*
If Player 2 chooses L, Player 1 should choose U.
If Player 2 chooses C, Player 1 should choose D.
*
*
If Player 2 chooses R, Player 1 should choose U.
*
Therefore, (U, L) is the Nash Equilibrium!
(1) The Game As a Whole
Player 2 Player 1 Leads Player 2 Leads Player 1
Player 1 Leads
34 32
27 24
Player 2 Leads
27 24
30 15
If Player 1 votes for itself, Player 2 should vote for Player 1.
If Player 1 votes for 2, Player 2 should vote for itself.
If Player 2 votes for 1, Player 1 should vote for itself.
If Player 2 votes for itself, Player 1 should vote for itself.
**
**
The Nash Equilibrium is for both players to vote for Player 1 to Lead!
Summary of Subgame Perfect Equilibrium Strategies
• Player 1– Vote for Player 1 to Lead
– If Lead, Choose M.
– If Follow,• Respond with U to L
• Respond with D to C
• Respond with U to R
– If Simultaneous, Choose U
• Player 2– Vote for Player 1 to Lead
– If Lead, Choose C
– If Follow,• Respond with L to U
• Respond with R to M
• Respond with R to D
– If Simultaneous, Choose L
Back to: How did you do?
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
43%
43%
14%
Treatment 1
Player 2 L: Nash C: Leader R: Follower
U: Nash 27
24 24
10 26
34 Player 1
M: Leader 33
18 29
12 34
32
D: Follower 31
20 30
15 32
28
10%
20%
30%
Treatment 2
Agreed 2 Leads
Disagreed on Leader
Agreed 1 LeadsAgreed 2 LeadsSimultaneous
Only 1 out of 7 subgame perfect Nash!
Only 1 out of 15 subgame perfect Nash equilibrium strategies submitted!
20%
10%
10%