econs 424 - repeated games ii - washington state university
TRANSCRIPT
EconS 424 - Repeated Games II
Félix Muñoz-García
Washington State University
March 25, 2014
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 1 / 28
Harrington, Ch. 14 "Check your Understanding" 1
Recall the Christie�s and Sotheby�s in�nitely repeated game fromclass:
0,0 2,1
1,2 4,4
4%
4%
Christies
Sothebys
1,4 1,7
4,1
7,1
5,5
6% 8%
6%
8%
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 2 / 28
Harrington, Ch. 14 "Check your Understanding" 1
Suppose that Christie�s and Sotheby�s want to collude to charge acommission rate of 8%, giving payo¤s of 5 for both auction houses.Suppose that if deviation were detected, the punishment is a twoperiod price war in which both auction houses set a commission rateof 4%.
Derive the conditions for this strategy pair to be a subgame perfectNash equilibrium.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 3 / 28
Harrington, Ch. 14 "Check your Understanding" 1
First, note that the Nash equilibrium of the unrepeated game isf6%, 6%g:
0,0 2,1
1,2 4,4
4%
4%
Christies
Sothebys
1,4 1,7
4,1
7,1
5,5
6% 8%
6%
8%
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 4 / 28
Harrington, Ch. 14 "Check your Understanding" 1
Histories can be partitioned into three types.
First, suppose the history is such that both auction houses are to seta rate of 8%. Then the equilibrium condition is
51� δ
� 7+ δ� 0+ δ2 � 0+ δ3�
51� δ
�where the 7 represents the instantaneous gain in utility from deviatingtowards 6%, when the other auction house still chooses thecooperative rate of 8%. In the following two periods the grim-triggerstrategy prescribes that both players punish by setting a rate of 4%(price war), yielding a payo of zero in both of these periods.Afterwards, cooperation is reestablished with a corresponding payo¤of 5 thereafter.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 5 / 28
Harrington, Ch. 14 "Check your Understanding" 1
Now consider a history whereby there is to be a punishment startingin the current period. Then the equilibrium condition is
0+ δ� 0+ δ2�
51� δ
�� 1+ δ� 0+ δ2 � 0+ δ3
�5
1� δ
�In this case note that in the right hand side we consider the payo¤that I obtain if I deviate towards 6% when my opponent still selects4% (implementing the price war), which constitutes my mostpro�table deviation. If I were to do that, however, I would bepunished with a price war for two periods, returning to thecooperation rate thereafter.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 6 / 28
Harrington, Ch. 14 "Check your Understanding" 1
Finally, if the auction houses are in the second period of thepunishment, then the equilibrium condition is
0+ δ
�5
1� δ
�� 1+ δ� 0+ δ2 � 0+ δ3
�5
1� δ
�A similar situation as in the previous slide applies.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 7 / 28
Harrington, Ch. 14 "Check your Understanding" 3
Recall the ABM Treaty in�nitely repeated game with imperfectmonitoring game from class:
10,10 6,12
12,6 8,8
NoABMs
NoABMs
USA
USSR
18,0 14,2
0,18
2,14
3,3
LowABMs HighABMs
LowABMs
HighABMs
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 8 / 28
Harrington, Ch. 14 "Check your Understanding" 3
Suppose a technological advance improves the monitoring technology,so that the probability of detecting ABMs are as follows:
Number of ABMs Probability of Detecting ABMsNone 0Low .3High .75
Using the strategy pro�le just described, derive the equilibriumconditions.
If you answer correctly, then you will �nd that the restriction factor isless stringent, indicating that better monitoring makes cooperationeasier.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 9 / 28
Harrington, Ch. 14 "Check your Understanding" 3
No ABMs is preferred to Low ABMs when
101� δ
� 12+ δ
�.3�
�3
1� δ
�+ .7�
�101� δ
��=) δ � 2
4.1� .49
where 0.3� 31�δ = 0.3�
�3+ 3δ+ 3δ2 + ...
�indicates the expected
and discounted stream of pro�ts that the deviating country obtains ifits deviation to Low is detected and therefore punished thereafter byreverting to the psNE of the unrepeated game, (High, High), with acorresponding payo¤ of 3.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 10 / 28
Harrington, Ch. 14 "Check your Understanding" 3
Similarly, 0.7� 101�δ = 0.7�
�10+ 10δ+ 10δ2 + ...
�represents the
expected and discounted stream of pro�ts that the deviating countryobtains if its deviation to Low is undetected, and then returns to thecooperative outcome (NoABMs, NoABMs) thereafter.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 11 / 28
Harrington, Ch. 14 "Check your Understanding" 3
No ABMs is preferred to High ABMs when
101� δ
� 18+ δ
�.75�
�3
1� δ
�+ .25�
�101� δ
��=) δ � 8
13.25� .60
This strategy pair is a subgame perfect Nash equilibrium when thediscount factor is at least 0.6, since δ � 0.6 is more restrictive thanδ � 0.49. With the weaker monitoring technology discussed in class,the discount factor had to be at least 0.74.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 12 / 28
Harrington, Ch. 14 "Check your Understanding" 3
A similar intuition as above is now applicable for the expected streamof payo¤s that the deviating country obtains if its deviation to High isdetected, 0.75� 3
1�δ = 0.75��3+ 3δ+ 3δ2 + ...
�, or undetected,
0.25� 101�δ = 0.25�
�10+ 10δ+ 10δ2 + ...
�. Therefore, cooperation
can be sustained for a larger set of discount factors as the monitoringtechnology becomes more accurate (closer to perfect monitoring).
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 13 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Assume that there are two �rms competing in quantities (a laCournot) with zero marginal costs and a demand function given by:
p(q1, q2) = 1� q1 � q2
so the pro�t function for �rm 1 is:
π1(q1, q2) = p � q1 = (1� q1 � q2) � q1= q1 � q21 � q1q2
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 14 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Find the equilibrium if the game is played just one time.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 15 / 28
Oligopoly and Collusion when �rms compete a la Cournot
The FOCs with respect to q1 are
1� 2q1 � 2q2 � 0
assuming we have an interior solution (meaning that both q1 and q2are positive), this FOC holds with equality. Solving for q1, we have�rm 1�s best response function
q1 =12� 12q2
Similarly by �rm 2 (since �rms 1 and 2 are symmetric):
q2 =12� 12q1
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 16 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Hence,q1
q2
BR2(q1)
BR1(q2)
1
1
½
½
⅓
⅓
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 17 / 28
Oligopoly and Collusion when �rms compete a la Cournot
The Nash equilibrium is given by:
q1 =12� 12
�12� 12q1
�=) qCournot1 = qCournot2 =
13
So the total production in the market is QCournot = 13 +
13 =
23 and
the market price is
pCournot = 1� 13� 13=13
In terms of pro�ts, each �rm i will get:
πi = p � qiπCournoti =
13� 13=19
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 18 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Find the equilibrium in a cartel (that is if the two �rms collude)
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 19 / 28
Oligopoly and Collusion when �rms compete a la Cournot
In a cartel, they maximize their joint pro�ts, as a monopoly.
Π(Q) = p �Q = (1�Q) �Q= Q �Q2
The FOC with respect to Q is:
1� 2Q = 0
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 20 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Solving for Q
QCartel =12
So each �rm will product half of QCartel :
qCartel1 = qCartel2 =QCartel
2=14
Thus, the price is
pCartel = 1� 12=12
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 21 / 28
Oligopoly and Collusion when �rms compete a la Cournot
So the pro�ts for each �rm i in the cartel are
πCarteli =12� 14=18
and as we can see
πCarteli =18> πCournoti =
19
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 22 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Can the two �rms achieve cooperation with the cartel quantity?
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 23 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Using the following Grim-Trigger strategy, in time period t:
If t = 1qi = q
Carteli =
14, thus πi =
18
If t > 1
qi =�qCarteli = 1
4 as long as qi = qj =14 in every previous period
qCournoti = 13 forever otherwise
with corresponding pro�t levels
πi =
�πCarteli = 1
8 as long as qi = qj =14 in every previous period
πCournoti = 19 forever otherwise
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 24 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Thus, by cooperating:
18+ δ
18+ δ2
18+ ... =
18� 11� δ
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 25 / 28
Oligopoly and Collusion when �rms compete a la Cournot
Let us now evaluate the �rm�s pro�ts from deviating : (But towardswhat?)The optimal deviation is:
maxqi
�1� qCartelj � qi
�qi
where �rm j sticks to the collusive agreement (producing qCartelj = 14 )
and �rm i deviates. The pro�t maximization problem for �rm i is now
maxqi
�1� 1
4� qi
�qi
We are now ready to �nd out which value of qi minimizes �rm i�spro�ts given that the other �rm (�rm j) repects the collusiveagreement (j cooperates). i.e.,
maxqi
�qi � qi
14� q2i
�Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 26 / 28
Oligopoly and Collusion when �rms compete a la Cournot
The FOC with respect to qi is
1� 14� 2qi � 0
and like before, we assume an interior solution and solve for qi ,
qDevi =38
With pro�ts as
πDevi =
�1� 1
4� qDevi
�qDevi
=
�1� 1
4� 38
�38=964
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 27 / 28
Oligopoly and Collusion when �rms compete a la Cournot
So the cooperation will be observed if and only if:
Payo¤ fromcooperatingz }| {
πCarteli � 11� δ
�
Payo¤ fromdeviatingz}|{πDevi +
Punishmentthereafterz }| {
πCournotiδ
1� δ18� 11� δ
� 964+19� δ
1� δ
Solving for δ, this condition holds when
δ � 917
so collusion is still possible among �rms is they do not discount thefuture too much.
Félix Muñoz-García (WSU) EconS 424 - Recitation 7 March 25, 2014 28 / 28