Lecture 7: Collusion and Dynamic Oligopoly Tom Holden io.tholden.

org

Why collude?

Examples and types.

Sustaining collusion in dynamic oligopoly. When is collusion easier?

In growing markets. In recessions/or in booms? Other factors.

Under Cournot or Bertrand competition, the price is always below the monopoly one (and quantity is higher).
Since a monopolist could choose either the Cournot or Bertrand price if they really wanted, aggregate industry profits must be lower under Cournot or Bertrand. So by agreeing to collude on a high price, all firms may increase their profits.

The business stealing effect underlying oligopoly competition drives this.

A firm that increases its quantity does not internalise the negative impact it will have on the profits of other firms. This externality means that (relative to the goal of maximising industry profits) firms will produce too much. A bit like a prisoners dilemma. There is a Pareto-dominant outcome (featuring collusion), that cannot be sustained as an equilibrium.

Open agreements.

E.g. the OPEC cartel. Generally just maximise joint profits, i.e. the cartel acts as if it was one firm not several. Illegal in most developed countries. E.g. between Sotherbys and Christies in the 90s.

Secret agreements.

CEOs of the two companies met in secret, and agreed on common commission charges. (Also shared client lists and removed the possibility of negotiating rates.) Eventually Christies came forward with information, leading to a $65 million fine of Sotherbys (combined US and EU), and a $7.5 million fine for the CEO of Sotherbys along with one year in jail. The two auction houses also paid customers over $0.5 billion in compensation.

Secret agreements (continued).

Or the Vitamin cartel of the 90s (Hoffman-LaRoche, BASF, Aventis, Solvay, Merck, etc.)

Regular exchange of sales data, price fixing. Eventually prosecuted, $0.5 billion fine for Hoffman-LaRoche in the US and $225m for BASF, plus additional fines in the EU. Prison time and personal fines for the executives.

Tacit agreements.

What we shall mostly focus on. E.g. the fact that almost all shops selling Sony TVs charge the same price (well above MC).

If my rival is selling vitamin C pills at 1 per 100 pills, no matter what informal agreement we might have in place, I will always be tempted to start selling them at 99p per 100. To sustain collusion then, my rival needs to be able to punish me for undercutting them.
E.g. by pricing at marginal cost for a prolonged period. Collusion is thus always a dynamic phenomenon.

Is having a finite number of sales periods enough to sustain > ?

Consider any game with a unique Nash equilibrium (e.g. Prisoners Dilemma, Cournot, Bertrand, Dixit-Stiglitz, Hotelling, Salop etc.), and imagine it is to be played times in succession, with final payoffs given by a (discounted) sum of payoffs from each periods game.

Thus finitely repeated symmetric marginal cost Bertrand competition results in = in each period.

To find the SPNE, as usual we start at the final period and work backwards. The final period is just the stage-game, so all players will play the Nash equilibrium of the stage-game. Given everyone is playing the Nash equilibrium in the stage-game in the final period (independent of the history up to there), in the penultimate period everyone will also play the Nash equilibrium of the stage-game. Etc.

Each firm 1, simultaneously sets its price , each period , to maximise the discounted sum of their present and future profits: where , is firm s profits in period . 1 , = ,1 + ,2 + 2 ,3 +
=1

Infinitely many periods, indexed by = 1,2, . firms, each with constant marginal cost . Market demand curve , finite monopoly profits , corresponding to a price .

A monopolist would maximise , meaning 0 = + .

Trivial equilibrium:

Maximum-profit collusive equilibrium:

In the first period, all firms set ,1 = . In subsequent periods , all firms set , = unless they have ever observed another firm setting a price other than , in which case they set , = . Thus if a firm ever deviates and sets a price below the monopoly one, from then on no firm makes a profit. This is a grim trigger strategy.

all firms set , = in each period .

A strategy is an SPNE if and only if there is no possible history up to a point such that some player would like to deviate from the strategy in period only. Proof:
Easy to see than in an SPNE there cannot be a period such that some player would like to deviate from the strategy in period only, since otherwise this profitable deviation would mean the strategy in the sub-game was not Nash. Proving the reverse is harder (and omitted).

Use one-stage deviation principle to prove that the collusive equilibrium is an SPNE:

If at some point a firm has ever set a price other than , then all firms are pricing at cost, which means we are effectively in the trivial equilibrium (from which no firm wants to deviate).

Use one-stage deviation principle to prove that the collusive equilibrium is an SPNE:
Profits from now on from not deviating are . =
1

If up to now all firms have priced at , does a firm want to deviate?

+

Profits from deviating are . (The deviating firm sets a price just below , then all firms price at from then on.) So the proposed strategy is an SPNE if and only if
1 ,

+ 2

i.e. if and only if 1 1 , which is true if and only if i.e. if and only if firms are sufficiently patient.

Firms with rational owners who set prices 1 where is the once per year will set = 1+ real interest rate. If they set prices times per year, then there will be periods in each year (so the period one year from period is period + ), so 1 firms will set = .
1+

Due to arbitrage between shares and bonds. Discount factors will be time varying.

Also suppose that each period, a firm has a probability of of being hit by a death-shock which would force it to exit the industry. (Assume exiting firms are immediately replaced.) Finally suppose that the discount factor was some .

Now suppose that demand was given by = 1 + so if > 0 demand is growing.

1

A monopolist would maximise 1 + in period , so the 1 . monopoly price is constant at , and monopoly profits are 1 +

Then the expected profits from setting the monopoly price 2 2 2 would be: + 1 + 1 + 1 + 1 +

Therefore behaviour in this model is the same as behaviour in the model without growing demand, and without death shocks, but with = 1 + 1 .

Putting the last two slides together with our original criterion, the maximal-profit collusive equilibrium (with growing demand and death shocks) is an SPNE if and only if 1 + 1 + 1 1 .

So, the likelihood of observing collusion is:

Decreasing in the number of firms. Increasing in the speed with which the market is growing. Decreasing in the probability of a death shock. Decreasing in real interest rates. Increasing in the frequency of price adjustment, providing the market is not growing too quickly.

Guess that = 1 + . Then the profits from setting the monopoly price (in the set up of 2 slides ago) would be: + 2 2 2 1 + 1 + 1 + 1 2 + .
1+ 1 1+

In a growing market, fixed is implausible.

So behaviour in this model is the same as behaviour in the model without growing demand, and without death shocks, but with = 1 . Our guess is verified as profits are not growing over time, so this is consistent with constant entry cost. So when there is free entry, growing demand neither makes collusion more nor less likely.

Yes, because of: The Folk Theorem (Fudenburg and Maskin 1986):

Consider any infinitely-repeated player game, with payoffs given by the discounted sum of payoffs from each periods stage-game. For each player , let be the lowest payoff that players other than can force on player in the stage-game. (So this is the payoff gets if the other players choose actions under which the payoff to from playing its best response is minimised.)

Let be the convex-hull of the set of possible payoff-vectors of the stage game, and let be the subset of this set containing payoff vectors 1 , 2 , , for which > for all (so all players are better off than the worst they could possibly be). Then providing either = 2 or is of dimension , then if players are sufficiently patient (i.e. discount factors are sufficiently high), and if 1 , 2 , , , there is an SPNE in which player attain an average payoff off (for all 1, , ).

Highest possible profit that a single firm may make is , when other firms make zero profits.
Easy to see this set is of dimension .

In symmetric Bertrand competition, the lowest payoff (profit) other players (firms) can force on a player is zero. (A firm can always set its price equal to and either sell nothing, or sell at cost.)

Thus any vector of firm profits 1 , 2 , , with > 0 for all and 1 + 2 + + is in . Hence, for sufficiently patient firms, there is an SPNE of repeated Bertrand in which average profits are given by any vector 1 , 2 , , with > 0 for all and 1 + 2 + + .

In symmetric Cournot competition, the lowest payoff (profit) other players (firms) can force on a player is also zero, by producing a very large quantity. (The remaining firm can always produce nothing.) Thus any vector of firm profits 1 , 2 , , with > 0 for all and 1 + 2 + + is in .
Easy to see this set is of dimension .

Highest possible profit that a single firm may make is , when other firms make zero profits.

Hence, for sufficiently patient firms, there is an SPNE of repeated Cournot in which average profits are given by any vector 1 , 2 , , with > 0 for all and 1 + 2 + + .

This is identical to the result under Bertrand, and contains points with profits lower than the Cournot profits. (So merely punishing by reverting to the Nash equilibrium would not work. Instead the strategies require punishing deviations from punishment etc etc.)

Rotemberg and Saloner (1986) extend the repeated-Bertrand setting to allow for stochastic demand.
In each period it is random whether demand is high or low (boom or recession). Firms can observe which of the two states demand is in. Changing demands level does not change the monopoly price, but it will change monopoly profits.

In the model we looked at the key condition stated that the discounted profits from sharing the monopoly outcome each period, must be bigger than the profits from taking it all now, and getting no profits in future.
In the Rotemberg and Saloner (1986) model, there are now two conditions, one for the high state, and one for the low state.
Expected future profits are the same in both cases, but in the high demand state the gain today from deviating is greater. So for intermediate values of , the monopoly price may be sustained in recessions, but it may not be sustained in booms. In booms a lower price is set which makes deviating less tempting. Thus mark-ups are counter-cyclical.

Green and Porter (1984) tell a different story and reach opposing conclusions.

Chief difference: firms cannot see the state of aggregate demand, they can only see the quantity they sold. They work with quantity competition, but the story may be told with price competition too if we assume that in the low demand state demand is zero . Then if a firm sees that it has sold few units, it does not know if it is has been undercut (in the high demand state), or if demand is just low.

In this price-setting version of the Green and Porter (1984) model, it cannot be optimal for firms to punish with = for ever after zero demand is observed, because this may have been the result of low demand, rather than another firm undercutting. Likewise, it cannot be optimal for firms to never punish, else the other firm would undercut. Instead, firms should punish with = for a fixed number of periods following an observed zero demand state, before reverting to the monopoly price. Thus in the Green and Porter (1984) model, we get procyclical mark-ups instead.
This agrees with the recent whole-economy empirical evidence of Nekarda and Ramey (2010). Nice evidence for this particular story is provided by the Porter (1983) case study of 19th century US railroad cartels.

Rather than thinking about common business cycle shocks, we can also think about shocks that hit one firm and not another.
E.g. a firms bank going bankrupt. Or a vital machine breaking. Or a new invention.

Such shocks may lead a firm to have no alternative but to cut its price, so to punish them with = for ever cannot be optimal. But the less often such shocks arrive, plausibly the easier sustaining collusion will be.

Evans and Kessides (1994) perform an empirical study of competition amongst US airlines.

Find that prices are higher on routes served by airlines that compete on a lot of other routes. Deviating on one route may be punished by price cuts on all other routes on which the deviating airline flies.
If all routes were identical this would not make any difference, because the cost of deviation would exactly balance the benefits of it (you could deviate in several markets simultaneously, thus earning higher profits). However, if airlines have natural cost advantages in some markets, then the maximal-profit collusive equilibrium would feature each airline serving the market in which it is most efficient, but ready to serve the other market should their rivals deviate.

Albraek et al. (1997) look at the consequences of the Danish Competition Councils 1993 decision to begin publishing the various prices charged by firms selling ready-mixed concrete.
Idea was that the increased transparency would make it easier for consumers to shop around. Reality was that prices increased by 15%-20% within a year, and price dispersion fell. Natural explanation is that the public price information made it easier to detect firms deviating from collusive equilibria, and thus made collusion more likely.
Cannot be explained by e.g. the business cycle.

We are more likely to observe in dynamic oligopoly when:

There are few firms in an industry. Markets are growing, but firm entry is blocked. Firms rarely leave the market. Real interest rates are low. Prices are adjusted frequently. Demand is observable and low. Demand is unobservable and high. Firms experience few asymmetric shocks. Firms compete in multiple markets. Prices are transparent.

Collusion with monopoly profits is sustainable as an SPNE of Cournot or Bertrand competition.

As are many other profit levels, thanks to the Folk Theorem.