HTML from LaTeXML, with custom CSS/JS. The PDF is more accurate.

A Note on Undominated Bertrand Equilibria††thanks: I thank Jonathan Vogel for stimulating this note and Uliana Loginova for proofreading.

Navin Kartik Department of Economics, Columbia University. Email: nkartik@columbia.edu.
December 2010

The standard model of Bertrand competition with homogenous products and identical marginal costs has a unique Nash equilibrium: each firm prices at marginal cost. Unfortunately, these strategies are weakly dominated. Blume (2003) has shown that when marginal costs are non-identical, there is an equilibrium (in fact, more than one) in undominated strategies. These equilibria achieve the conventional market outcome of the lowest-cost firm serving the entire market at a price equal to the next-to-lowest marginal cost. A natural question in this case is whether all undominated equilibria have the same outcome in terms of market price and share. To my knowledge, this issue has not been settled in the literature.111Blume (2003) notes that lower market prices can be supported in equilibria with weakly dominated strategies. This note provides an affirmative answer under mild assumptions.

Model.

Consider the standard two-firm homogenous-products Bertrand pricing model with constant but non-identical marginal costs, 0≤c1<c2.222The analysis extends straightforwardly to more than two firms. What is important is that there is only one firm with the lowest cost. Market demand is given by a function Q⁢(p), where Q:ℝ→ℝ+. Denote Πi⁢(p):=Q⁢(p)⁢[p−ci] as the monopoly profit for firm i at price p. Firms choose prices simultaneously, denoted p1 and p2 respectively. If pi<pj, firm j’s payoff is zero and firm i’s payoff is Πi⁢(pi); if p1=p2, each firm i gets 12⁢Πi⁢(pi).

I make the following five assumptions, all of which are satisfied in textbook examples:

(A0): There exists ε^>0 such that Q⁢(⋅) is Lipschitz continuous on [c2,c2+ε^).

(A1): For any p>c2 with Q⁢(p)>0, there exists some i∈{1,2} and p′<p such that Πi⁢(p′)>12⁢Πi⁢(p). In other words, firm i would rather charge the price p′<p and sell to the whole market than split the market at p.333Note that this assumption is automatically satisfied if the demand function, Q⁢(p), is non-increasing.

(A2): For each i∈{1,2}, max⁡Πi⁢(p) exists and is finite. In other words, for each firm, there exists an optimal monopoly price.

(A3): For any p<c2, Π1⁢(p)<Π1⁢(c2). In other words, as a monopolist, any price below c2 would be strictly worse for firm 1 than pricing at c2.

(A4): There exists ε~>0 such that Π2⁢(p) is strictly increasing on (c2,c2+ε~).  In other words, monopoly profits for firm 2 are strictly increasing just above price c2.

Role of the Assumptions.

(A0) and (A3) are used to construct an equilibrium where firm 1 serves the entire market at price c2; (A3) and (A4) ensure that this equilibrium is in undominated strategies. (A1) and (A2) guarantee uniqueness of the desired market outcome; in particular, (A2) rules out the kinds of equilibria constructed by Baye and Morgan (1999).

Given that there are a continuum of pure strategies, say that a mixed strategy is weakly undominated if, viewed as a probability measure, it assigns zero probability to any (Borel) set of pure strategies that are all weakly dominated.

Proposition.

Any Nash equilibrium in weakly undominated strategies has the property that with probability one, firm 1 sells to the entire market at price c2. Moreover, such an equilibrium exists.

Proof.

(A0) and (A3) imply existence of a Nash equilibrium (NE, hereafter) with the desired market price and share using Blume’s (2003) construction: firm one plays a pure strategy of charging price c2 and firm 2 mixes, say with a uniform distribution, on [c2,c2+δ] for some small enough δ>0.444Plainly, firm 2 is playing a best response to firm 1.  It is routine to verify that (A0) ensures that δ>0 can be chosen small enough so that firm 1 does not want to raise price above c2, while (A3) obviously implies that it does not want to decrease price below c2. Firm 1’s strategy is undominated because, by (A3), it is playing its unique best response to firm 2. Similarly, any p2>c2 in firm 2’s support would be the unique best response to firm 1 playing a uniform distribution on [p2,p2+ε⁢(p2)] for some small enough ε⁢(p2)>0 (by (A0), (A4), and that δ can be chosen small enough); hence firm 2 is also playing an undominated strategy.

To prove the first statement of the Proposition, fix any undominated NE with strategies (σ1,σ2). For i∈{1,2}, denote p¯i:=sup[S⁢u⁢p⁢p⁢[σi]] and p¯i:=inf[S⁢u⁢p⁢p⁢[σi]]. By (A4), it is weakly dominated for firm 2 to charge any price less than or equal to c2, so p¯2≥c2 and σ2 must put zero probability on c2, which implies p¯2>c2. By (A3), any p1<c2 is not a best response for firm 1, hence p¯1≥c2. Therefore, it suffices to show that p¯1=c2.

Assume, to contradiction, that p¯1>c2. Notice that by (A4), firm 2 can get a positive expected profit by choosing some p2=c2+ε for small enough ε>0; hence σ2 must put zero probability on prices that yield a zero expected profit against σ1. This implies that p¯2≤p¯1. Similarly, since (A3) implies Π1⁢(c2)>0, it also follows that p¯1≤p¯2.  Combined, we must have p¯:=p¯1=p¯2>c2.  Moreover, since charging any price above an optimal monopoly price is weakly dominated for a firm, (A2) implies that p¯<∞.  There are now two cases to consider:

1) Suppose first that σ2 puts positive probability on p¯. Then p¯ must yield firm 2 a positive expected profit, hence Q⁢(p¯)>0 and σ1 must also put positive probability on p¯. But then (A1) implies that one of the two firms is not playing a best response.

2) Suppose next that σ2 puts zero probability on p¯. This implies that Prσ2⁡(p2≥p¯−ε)→0 as ε↓0. But then, σ1 is not a best response, because firm 1’s expected profit from p1 becomes arbitrarily small as p1↑p¯ (since (A2) implies that monopoly profits are bounded) whereas charging price c2 gives firm 1 some strictly positive expected profit (by (A3)). ∎

References

HTML from LaTeXML, with custom CSS/JS. The PDF is more accurate.