Independent scholarship · Open accessThe monthly digest

Indexed text attachment

games-and-information-in-higher-categories.tex

Source for Games and Information in Higher Categories

View paper · Download original file

Lines 4919–5018 of 6559. Source text is displayed literally.

4919\begin{exercise}4920Explain, using \cref{sec:thompson}, why the Dutch and first-price auctions are identified by Thompson moves4921while the English and second-price auctions are not, by describing the information sets of each.4922\end{exercise}4923%% ---- end ch/17-auctions.tex ----4924%% ---- begin ch/18-pricing.tex ----4925\chapter{Pricing}\label{ch:pricing}4926 4927The oldest application of game theory is to firms that are neither price-takers nor monopolists. This4928chapter revisits the models of \cref{ch:order} with the tools accumulated since: commitment, repetition,4929information and mechanism design each resolve, or relocate, a classical puzzle about how prices are4930set.4931 4932\section{Quantities or prices?}4933 4934Cournot's firms choose quantities and earn positive profits; Bertrand's choose prices and earn none4935(\cref{ex:cournot,ex:bertrand}). The two models make opposite predictions from the same primitives, and for4936a long time the choice between them was a matter of taste. The reconciliation is a two-stage game.4937 4938\begin{theorem}[Kreps--Scheinkman]\label{thm:KS}\Scl[\cite{KS83}]4939Suppose two firms first choose capacities, at a cost, and then, knowing both capacities, choose prices,4940with demand rationed efficiently when the low-priced firm cannot serve it all. Then in the unique4941subgame-perfect outcome the capacities are the Cournot quantities and both firms charge the Cournot price.4942\end{theorem}4943 4944Quantity competition is price competition preceded by a commitment. A firm that has installed a small4945capacity has burned the bridge to aggressive pricing: it cannot serve the customers it would win by4946undercutting. This is the logic of \cref{ch:perfection}---reducing one's own options changes the rival's4947context---and in the language of \cref{ch:order} the capacity stage converts a game with a discontinuous,4948knife-edge best response into one of strategic substitutes. Which model applies is then an empirical4949question about what is hard to change in the short run. Where capacity is the slow variable (cement,4950steel, shipping), Cournot; where it is not (software, generic drugs sold by firms with ample capacity),4951Bertrand.4952 4953\section{Escaping the Bertrand paradox}4954 4955The Bertrand outcome rests on a discontinuity: at equal prices firms share the market, and an arbitrarily4956small price cut captures all of it. Each of the following removes the discontinuity or its consequence.4957 4958\emph{Capacity constraints.} If neither firm can serve the whole market, undercutting has limited appeal and4959the high-priced firm retains residual demand. For intermediate capacities there is no pure-strategy4960equilibrium: prices cycle, in Edgeworth's phrase, and the equilibrium is in mixed strategies, which is how4961one may read sales and price dispersion.4962 4963\emph{Differentiation.} If products are imperfect substitutes, demand is continuous in prices, the price4964game is supermodular (\cref{ex:bertrand}), and prices exceed cost by a margin that measures the4965differentiation.4966 4967\emph{Repetition.} If the firms meet repeatedly, the folk theorem applies. With $n$ identical Bertrand4968competitors, sharing the monopoly profit $\pi^m$ under the threat of reversion to marginal-cost pricing is4969an equilibrium iff $\frac{\pi^m/n}{1-\delta}\ge\pi^m$, that is4970\[ \delta\ \ge\ 1-\frac1n. \]4971Collusion is easier with fewer firms, with more frequent interaction, and with quicker detection. The4972structure is the two-state automaton of \cref{ex:grim}. In the terms of \cref{prop:markov} the collusive4973price is sustained by strategies that are not Markov: the firms condition on a payoff-irrelevant history,4974and antitrust law's hostility to ``concerted practices'' is hostility to exactly that frame.4975 4976\emph{Search.} If consumers must pay a small cost to learn each additional price, the outcome flips to the4977other extreme. No firm gains by pricing below its rivals, since nobody will find out; each can raise its4978price by less than the search cost without losing a customer; and the unique equilibrium has every firm at4979the \emph{monopoly} price, however many firms there are \cite{Dia71}. Bertrand and Diamond are the same4980discontinuity seen from its two sides: with zero search cost, price equals marginal cost, and with any4981positive search cost it equals the monopoly price. The realistic middle ground, with dispersed prices,4982requires heterogeneity among consumers in what they know.4983 4984\section{Location}4985 4986Differentiation is itself chosen. In Hotelling's model \cite{Hot29} consumers are spread uniformly along a4987street of unit length and buy one unit from the firm with the lowest price plus transport cost.4988 4989If prices are fixed (by regulation, or because the ``firms'' are political candidates and the ``price'' is4990absent), two firms choosing locations both go to the centre: each can gain market share by moving towards4991the other, and the only equilibrium is back-to-back at the median. This is the \emph{principle of minimum4992differentiation}, and in politics the median voter theorem. With three firms there is no pure-strategy4993equilibrium at all.4994 4995If the firms also choose prices, the conclusion reverses.4996 4997\begin{example}[Price competition on a line]\label{ex:hotelling}4998Let the firms be at the two ends of the street, with marginal cost $c$, and let transport cost be4999quadratic, $t\cdot(\text{distance})^2$. The consumer at $x$ is indifferent when $p_1+tx^2=p_2+t(1-x)^2$, so5000firm~1's demand is $\hat x=\frac{p_2-p_1+t}{2t}$. Its best response is $p_1=\tfrac12(p_2+c+t)$, increasing in5001$p_2$---prices are strategic complements---and the equilibrium is5002\[ p_1=p_2=c+t,\qquad\text{with profits }\tfrac t2\text{ each.} \]5003The margin is the transport cost: firms earn exactly what their distance from each other protects.5004\end{example}5005 5006Anticipating this, firms choosing locations before prices face two forces: moving towards the rival steals5007customers at given prices, and intensifies price competition. With quadratic transport costs the second5008dominates and the firms locate at the opposite ends, \emph{maximal differentiation}; with Hotelling's5009original linear costs the price game has no pure-strategy equilibrium when the firms are close, so his5010argument for minimum differentiation was incomplete \cite{dGT79}. The location stage is a commitment to5011soft pricing, and the sign of the strategic effect is the one predicted in \cref{ch:order} for5012complements.5013 5014\section{Comparative statics without calculus}5015 5016Because differentiated-product price games are supermodular, \cref{thm:supermodular} applies wholesale.5017If any parameter raises each firm's marginal profitability of a higher price---an increase in its own5018cost, in a rival's cost, in demand, in a tax---then the greatest and least equilibrium price vectors rise.