Whether the game is repeated infinitely or there is some finite number of times;
Whether there is a full information available to each firm about the objectives of, and opportunities available to, other firms;
How much weight the firms attach to the future in their calculations;
Whether the "cheating" can/can not be detected due to the knowledge/lack of knowledge about the prior moves of the firm's rivals.
The fact of repetition broadens the strategies available to the players,
because they can make their strategy in any currant round contingent on the others' play in previous rounds. This introduction of time dimension permits strategies, which are damaging to be punished in future rounds of the game. This also permits players to choose particular strategies with the explicit purpose of establishing a reputation, e.g. by continuing to co- operate with the other player even when short-term self-interest indicates that an agreement to do so should be breached.
b.) Finite game case.
But repetition itself does not necessarily resolve the prisoner's dilemma. Suppose that the game is repeated a finite number of times, and that there is complete and perfect information. Again, we assume firms to maximise the (possibly discounted) sum of their profits in the game as a whole. The collusive low output for the firms again, unfortunately for the firms, could not be sustained. Suppose, they play a game for a total of five times. The repetition for a predetermined finite number of plays does nothing to help them in achieving a collusive outcome. This happens because, though each player actually plays forward in sequence from the first to the last round of the game, that player needs to consider the implications of each round up to and including the last, before making its first move. While choosing its strategy it's sensible for every firm to start by taking the final round into consideration and then work backwards. As we realise the backward induction, it becomes evident that the fifth and the final round of the game would be absolutely identical to a "one-shot" game and, thus, would lead to exactly the same outcome. Both firms would cheat on the agreement at the final round. But at the start of the fourth round, each firm would find it profitable to cheat in this round as well. It would gain nothing from establishing a reputation for not cheating if it knew that both it and its rival were bound to cheat next time. And this crucial fact of inevitable cheating in the final round undermines any alternative strategy, e.g. building a reputation for not cheating as the basis for establishing the collusion. Thus cheating remains the dominant strategy.
* NOTE: the is however one assumption about slightly incomplete information, which allows collusive outcome to occurin the finitely repeated game, but I will left it for the discussion some paragraphs later.
c.)_ Infinite game case.
Now lets consider the infinitely repeated version of the game. In this kind of game there is always a next time in which a rival's behaviour can be influenced by what happens this time. In such a game, solutions to the problems represented by the prisoners dilemma are feasible.
i.) "Trigger" strategy
Suppose that firms discount the future at some rate "w", where "w" is a number between O and 1. That is, players attach weight "w" to what happens next period. Provided that "w" is not too small, it is now possible for non-co-operative collusion to occur. Suppose that firm B plays "trigger" strategy, which is to choose low output in period 1 and in any subsequent period provided that firm A has never produced high output, but to produce high output forever more once firm A ever produces high output. That is B co-operates with A unless A "defects", in which case B is triggered into perpetual non-co-operation. If A were also to adopt the "trigger" strategy, then there would always be collusion and each firm would produce low output. Thus the discounted value of this profit flow is:
If fact A gets this pay-off with any strategy in which he is not the first to defect. If A chooses a strategy in which he defects at any stage, then he gets a pay-off of 3 in the first period of defection (as B still produces low output), and a pay-off of no more than 1 in every subsequent period, due to B being triggered into perpetual non-co-operation. Thus, A's pay-off is at most
If we will compare these two results, we will get that it is better not to defect so long as
W > (or =)
We can conclude that is the firms give enough weight to the future, then non-co-operative collusion can be sustained, for example, by "trigger" strategies. The "trigger" strategies constitute a Nash equilibrium = self-sufficient agreement. However it is not enough for a firm to announce a punishment strategy in order to influence the behaviour of rivals. The strategy that is announced must also be credible in the sense that it must be understood to be in the firm's self-interest to carry out its threat at the time when it becomes necessary. It must also be severe in a sense that the gain from defection should be less than the losses from punishment. But because it is possible that mistakes will be made in detecting cheating (if, for example, the effects of unexpected shifts in output demand are misinterpreted as the result of cheating), the severity of punishment should be kept to the minimum required to deter the act of cheating.
Trigger strategies are not the only way to reach the non-co-operative collusion. Another famous strategy is Tit-for-Tat, according to which a player chooses in the current period what the other player chose in the previous period. Cheating by either firm in the previous round is therefore immediately punished by cheating, by the other, in this round. Cheating is never allowed to go unpunished. Tit-for-Tat satisfies a number of criteria for successful punishment strategies. It carries a clear threat to both parties, because it is one of the simplest conceivable punishment strategies and is therefore easy to understand. It also has the characteristics that the mode of punishment it implies does not itself threaten to undermine the cartel agreement. This is because firms only cheat in reaction to cheating be others; they never initiate a cycle of cheating themselves. Although it is a tough strategy, it also offers speedy forgiveness for cheating, because once punishment has been administered the punishing firm is willing once again to restore co-operation. Its weakness is in the fact that information is imperfect in reality, so it is hard to detect whether a particular outcome is the consequence of unexpected external events such as a lower demand than forecast, or cheating, Tit-for-Tat has a capacity to set up a chain reaction in a response to an initial mistake.
d.) Finite game case, Kreps approach.
Lets now return to the question of how collusion might occur non-co-operatively even in the finitely repeated game case. Intuition said that collusion could happen- at least at the earlier rounds- but the game theory apparently said that it could not. Kreps et al. (1982) offered the elegant solution to this paradox. They relax the assumption of complete information and instead suppose that one player has a small amount of doubt in his mind as to the motivation of the other player. Suppose A attaches some tiny probability p to B referring- or being committed- to playing the "trigger" strategy. In fact it turns out that even if p is very small, the players will effectively collude until some point towards the end of the game. This occurs because its not worth A detecting in view of the risk that the no-collusive outcome will obtain for the rest of the game, and because B wishes to maintain his reputation for possibly preferring, or being committed to, the "trigger" strategy. Thus even the small degree of doubt about the motivation of one of the players can yield much effective collusion.
The motives for retaliation.
The motives for retaliation differ in three approaches. In the first approach, the price war is a purely self-fulfilling phenomenon. A firm charges a lower price because of its expectations about the similar action from the other one. The signal that triggers such a non-co-operative phase is previous undercutting by one of the firms. The second approach presumes short-run price rigidities; the reaction by one firm to a price cut by another one is motivated by its desire to regain a market share. The third approach (reputation) focuses on intertemporal links that arise from the firm's learning about each other. A firm reacts to a price cut by charging a low price itself because the previous price cut has conveyed the information that its opponent either has a low cost or cannot be trusted to sustain collusion and is therefore likely to charge relatively low prices in the future.
So far I have discussed the collusion using some simple example with a choice of output levels made by the two firms. But there may be several firms in the industry, and in fact firms have a much broader choice. It may be that their decision variable is price, investment, R&D and advertising. Nevertheless the more or less the same analysis could be applied in each of the case.
I have examined different assumptions and predictions, which allow or do not allow the possibility of collusion. In reality such thing as collusion definitely takes place, if it had not, there would not have been any strong an ambiguous discussion of this topic. But I think it would be appropriate to end this essay with an explicit reminder that once we leave the world of perfect competition, we lose the identity of interests between consumers and producers. So, the discussion of benefits to firms in oligopoly that arise from finding strategies to enforce collusive behaviour might well have been the discussion of the expenses of consumers.
J.Vickers, "Strategic competition among the few- Some recent developments in the economics of industry".
J.Tirole, "The theory of industrial organisation". Ch 6.
Estrin & Laidler. "Introduction to microeconomics". Ch 17.
W.Nicholson, "Microeconomic theory". Ch 20.