Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Game Theory: Oligopolies

Ashesh Shah
Abstract. In a survey of oligopoly models, techniques of game theory are adapted to nd market equilibria and understand deciding forces for rms. We look at situations with increasing complexity, to develop a better understanding of the usefulness of the game theory in market analysis. Initially, we explore the basics of game theory. Then, following an introduction of basic microeconomics, we explore simple static games of competitive interaction. Finally, we add complexity to these models to explore the eects of time and repeated interactions.

1. Introduction. Game theory is generally known as the analysis of the interaction between multiple individuals in the decision making process. Unfortunately, many people confuse game theory as a whole with one of its simplest examples, The Prisoners Dilemma. The Prisoners Dilemma addresses the decision making of two individuals suspected of a crime. In this scenario, the two suspects are being questioned in separate rooms. While the police lack sucient evidence to convict the two, they are attempting to get a voluntary confession from either or both of the suspects. The police explain the outcomes of confessing and denial. If both confess, they face a reduced sentence of 5 months. If only one confesses, he is set free, and the other faces a maximum sentence of 10 months. If neither confesses, they are detained for a month on lesser charges. This problem is represented in a more succinct format in Table 1. By looking at Table 1-1 The Prisoners Dilemma Confess Deny Confess 5, 5 10, 0 Deny 0, 10 1, 1

the outcomes for a given strategy, we deduce the best strategy for the suspects to take. For example, if Suspect 1 chooses to confess, Suspect 2 either faces jail for 5 months if he confesses, or for 10 months if he denies the crime. Obviously faced with this choice, Suspect 2 confesses. If Suspect 1 chooses to deny, Suspect 2 will either be free if he confesses, or face a one month sentence if he denies the crime. Again, Suspect 2 chooses to confess. Here we have demonstrated a dominant strategy for Suspect 2 (Suspect 1s strategies are identical because the game is symmetric). Regardless of the strategy that Suspect 1 chooses, Suspect 2 always chooses to confess. This process is known as elimination of strictly dominated strategies. In this scenario, we discovered that for any selection by Suspect 2, confession was a better strategy. The denial strategy is said to strictly dominate the confession strategy. As such, the denial strategy is eliminated from the set of strategies from which either suspect is willing to choose. If this were a game that had more than two options, then the solution could 171

172

MIT Undergraduate Journal of Mathematics

be found by using a process called iterative dominated strategy elimination. Under this process, strategies that are strictly dominated are discarded from the game. The game is analyzed again to nd strictly dominated strategies. As the process is repeated, the game becomes simplied to the point where the best strategy is discovered. The Prisoners Dilemma is a good example of a static game of complete information. This is probably the simplest form of a game that involves the interaction of multiple individuals in a decision making process. Though this particular game may not appear to be that useful in modeling interactions, it does provide a solid base under which more complicated systems may be understood. With this basic understanding of game theory, we build the basic framework of multirm competition in Section 2. This framework is used to develop an understanding of complex interactions and a set of criterion by which they are solved. In each case, the aim is to understand all the information that each player has, the decisions that they are able to make, and the outcomes as a result of each strategy. From this information we deduce which strategies are used and how often they are employed. To develop an understanding of the workings of real oligopolies, we add levels of increasing complexity, starting from static games of complete information, in Section 2, to repeated interaction, in Section 4. Beginning with the simplest Bertrand model in which we assume that rms were price takers, to multiple stage Cournot collusion, we add realism to our originally naive models. In fact, the cartel system modeled in the nal section is like that used within the Middle East to set production quantities for each year. More complex treatments of multirm competition are found in Gibbons book [1], particularly in Chapters 2 and 3. 2. Simple Multirm Competition. To understand the interaction of multiple rms, we must understand a bit of economics concerning markets and pricing. When rms sell a product, they face a demand curve under which consumers are willing to consume a quantity dependent upon the price. A more complete account of rms actions is found in Chapters 8, 9, and 10 of Rubenfeld and Pindycks book [2]. Depending on the model of the marketplace, rms either produce a specic quantity or set a specic price. The rm makes these choices dependent upon the possible outcomes and what they expect the other rms to do. For simplicity, we examine a two-rm oligopoly. First, we look at the Bertrand model for oligopolies, where rms are price setters. The rms in this model face a demand curve characterized by an equation for QD (p), in which QD is the quantity demanded of the rm at a set price p. For simplicitys sake, we assume that demand is a linear function of the form, QD (p) = a bp. (2-1)

Next, we must examine the rm. Firms face costs of production of two forms, xed and variable. Fixed costs are associated with capital investments and are one-time costs, whereas variable costs depend upon the number of goods produced. For this example, we assume that xed costs are zero and that marginal costs (the costs for producing a single added good) are constant, with value c. This assumption means that costs are linear in qs , the quantity supplied. So, the cost function is of the form, C(q) = cqs . (2-2)

Game Theory: Oligopolies

173

When examining the possibilities for the rms, we divide the pricing scheme into three ranges. The rst range, p < c, results in a rm selling the good for less than it costs to produce it. As such, it is impossible for a rm to sell at p < c. The second range is p > c. In this range, the rm whose price is lower captures the entire market. As a result, the best strategy is to price lower than what you estimate your competitor to price. Unfortunately, the other rm also has the same strategy. Thus we are lead to examine the nal possible price, p = c. At this price, neither rm is willing to lower price. Essentially, both rms sell at marginal costs (this is the competitive equilibrium and the most ecient outcome). Obviously, the assumptions made under this model reduce the problem of oligopoly to triviality. For this reason, it is important to look past this model to more complex reasons. A more interesting version of this rst model is the Cournot model. Under this model, the market assumptions are identical, except that in this case, rms are price takers (that is, they produce a certain quantity, and the price is determined by the market demand). The rms strategy is based on quantity rather than price. In this example, two rms are competing. The two rms, Firm 1 and Firm 2, set their production quantities q1 and q2 , respectively. Because there are two suppliers in the market, the market price is a function of the sum of the two rms production, (q1 + q2 ), which is the total production in the market. As with the previous case, we employ a linear demand curve and a constant marginal cost of production. The outcome from this strategy decision is the rms prot, which is a function of the market price, the quantity sold, and the marginal cost: (q1 ) = q1 P (q1 + q2 ) q1 c. (2-3)

In words, Equation 2-3 states that a rms prots are equal to the quantity it sells times the price at which it sells them, minus the costs associated with production. By substituting for the inverse demand function, P (q1 + q2 ), which is Equation 2-1, solved for price, we nd Firm 1s prot function to be (q1 ) = q1 a q1 + q2 b b q1 c. (2-4)

As is the case with any rm, Firm 1 chooses its quantity of production so as to maximize its own prots. As is evident from Equation 2-4, Firm 1s prots depend on its production (q1 ) and Firm 2s production (q2 ). Because Firm 1 is able to control only its own production, it must maximize its prots versus its quantity (q1 ). The prot maximizing quantity is found by setting the derivative of the prot function for Firm 1 equal to zero: a 2q1 + q2 (q1 ) = c = 0. q1 b b Solving Equation 2-5 for q1 in terms of q2 , we nd q1 = a cb q2 . 2 a cb q1 2 (2-6) (2-5)

The same process is repeated for Firm 2, resulting in an equation for q2 in terms of q1 : q2 = (2-7)

174

MIT Undergraduate Journal of Mathematics

Equations 2-6 and 2-7 are known as reaction curves. They predict what quantity each rm would produce, given the quantity that the other rm is producing. Solving these two equations, we nd a solution to this model: q1 = q2 = a cb . 3 (2-8)

The quantities found in Equation 2-8 are substituted into the original prot function, Equation 2-4, to calculate the prots of each rm (only Firm 1 is shown below as both are identical): 1 (q1 ) = 1 (a cb)2 . 9

Because the rms are identical, their strategies are identical. Because there is perfect information, both rms know the other rms best strategy. By solving the two reaction curves simultaneously, we nd the equilibrium levels of production known as the Cournot equilibrium. It is interesting to note that the Cournot equilibrium is in between the monopoly levels of production and the competitive levels. In addition, the prots earned by the two rms are less than half the monopoly prots. 3. Dynamic Multirm Competition. The next interesting model to look at is called the Stackelburg model. This model is essentially identical to the Cournot model, except that it is a dynamic game. This means that there is a sequence to the course of events in the decision making process. Rather than both rms setting quantities at the same time, one rm is able to set a quantity rst. For both rms, the prot function looks identical to the one above. However, rather than maximizing prot with Firm 2s quantity as an unknown, Firm 1 uses its knowledge of Firm 2s reaction curve to gain further prots. Firm 1s function now is solely a function of q1 : (q1 ) = q1 Substituting Equation 2-7, we nd (q1 ) = q1 And simplifying, we get (q1 ) = q1 a q1 c 2b 2 . a q1 + (a cb q1 )/2 b b q1 c. a q1 + q2 (q1 ) b b q1 c.

Again, maximizing prots over quantity, Firm 1 produces a Stackelburg equilibrium quantity, and Firm 2 produces a quantity that maximizes its prots, given Firm 1s quantity as dened by its reaction curve. To nd Firm 1s production, we must take the derivative of its prot function with respect to Firm 1s quantity and set it equal to zero: a 2q1 c d(q1 ) = = 0. dq1 2b 2 (3-1)

Game Theory: Oligopolies

175

Solving Equation 3-1 for q1 , we nd the production strategy that maximizes Firm 1s prots: q1 = a cb . 2

Firm 2 has no option but to react to Firm 1s production level in a way that maximizes its prots. From the original Cournot model, we know that this strategy is determined by Firm 2s reaction curve, given by Equation 2-7. Subtracting the value for q1 given by the equation above, we nd q2 to be q2 = a cb . 4

Thus, by having the rst mover advantage, Firm 1 captures more of the market share. In addition, it increases its own prots. 4. Multistage Simple Multirm Competition. One of the more interesting adaptations of these theories is the multiple stage game. One good example is a Cournot model, in which the two rms try to collude over time. Collusion is the cooperation of two or more rms to restrict their production to maintain prices at a higher level. The aim of the cartel (the group of rms that is colluding) is to produce a total quantity that is the monopoly quantity, a level at which the total prots would be highest. By examining our previous models, we see that there is no incentive for the rms to stick to the collusion quantity. Specically, if both rms are producing such that the total supply is the monopoly level, there is prot to be made by either rm if they were to raise production. To see this incentive, we must rst examine the monopoly setting. To nd the monopoly level of production, assume that both rms act as one, and set a total quantity and total prots. The prot function is then simply a function of total quantity: (q) = q aq b q c. (4-1)

Again, we must maximize prots by setting the derivative of prots with respect to quantity equal to zero: d(q) = dq Solving this equation for q, we nd q= a cb . 2 a 2q b c = 0.

Because the two rms are acting equally in this monopoly, their individual quantities are half of the total, and therefore q1 = and q2 = a cb . 4 a cb 4 (4-2)

176

MIT Undergraduate Journal of Mathematics

The prots for both rms are equal to half of the monopoly prot and are found by substituting q into Equation 4-1: 1 (q1 ) = 2 (q2 ) = 1 (a cb)2 (q) = . 2 4

However, under the quantity strategy of the rm, if it knows that its competitor is producing the colluding quantity, its prot maximizing strategy is to produce more. In other words, assume that Firms 1 and 2 are competing and that Firm 1 produces its collusion quantity given by Equation 4-2. Subtracting this value into Firm 2s reaction curve, we nd that Firm 2s prot maximizing production quantity is q2 = 3a 3cb a cb q1 = . 2 8

This production level is greater than the collusion quantity, and results in the breakdown of cooperation between the two rms. The prots derived from this level of production are calculated by substituting the two quantities into the original Firm 2 prot function: 2 (q2 ) = 3a 3cb 9 1 q1 + q2 c . b

Simplifying this equation, we nd that this prot level is greater than that associated with the monopoly level of production: 2 (q2 ) = 9 1 (a bc)2 > m (qm ). 64 2

Each rm is then faced with an incentive to cheat and produce more than half of the monopoly quantity to reap more prots. Because of this, the rms must create a system to insure collusion and punish cheaters by making it unprotable in the long run. Here is where the game aspect becomes important. If a rm cheats, the other rm has the ability to return to Cournot production quantities, causing a loss in prots to both rms. Instead of making half of the monopoly prot, the rm would make only the Cournot prot. For this punishment to be eective, the term of the Cournot production level must be long enough so that the benets to cheating are outweighed by the losses: 1 1 1 9 (a cb)2 m (qm ) < T m (qm ) a cb 64 2 2 9 Solving for T , we nd T > 9 2. 8
2

If the punishment for cheating is two periods of Cournot production, the cartel is able to prevent cheating and maintain monopoly levels of output. A punishment of two periods of Cournot production is suciently large as to outweigh the gains in prots from overproducing. The ndings in Equation 26 assume that there is no discount rate, which would make future prots less valuable than current prots. If there were a discount rate, or a time value to money, the punishment period would be longer.

Game Theory: Oligopolies

177

References [1] Gibbons, R., Game Theory for Advanced Economists, Princeton University Press, 1992. [2] Rubinfeld, D., and Pindyck, R., Microeconomics, Prentice-Hall.

178

This page will be blank.

You might also like