Indexed text attachment
games-and-information-in-higher-categories.tex
Source for Games and Information in Higher Categories
View paper · Download original file
Lines 2401–2500 of 6559. Source text is displayed literally.
2401bank runs, arms races, team production.2402 2403\section{Quantity competition}2404 2405\begin{example}[Cournot]\label{ex:cournot}2406Two firms choose quantities $q_1,q_2\ge0$ of a homogeneous good with inverse demand $P=a-b(q_1+q_2)$ and2407constant marginal cost $c<a$. Firm $i$'s profit is $q_i(a-c-b(q_i+q_j))$, and her best response is2408\[ q_i=\frac{a-c-bq_j}{2b}\quad(\text{or }0). \]2409Best responses slope \emph{downward}: quantities are \term{strategic substitutes} \cite{BGK85}. The unique equilibrium2410is $q_1=q_2=\frac{a-c}{3b}$, with price $\frac{a+2c}{3}$. With $n$ firms each produces $\frac{a-c}{(n+1)b}$2411and the price $\frac{a+nc}{n+1}$ tends to marginal cost.2412\end{example}2413 2414The Cournot duopoly has \emph{decreasing} differences in $(q_i,q_j)$, so it is not supermodular as it2415stands. But reverse the order on firm~2's action set and it is: with two players, strategic substitutes2416become strategic complements under the duality $L\mapsto L^{\op}$ applied to one of them. All of2417\cref{thm:supermodular} then applies, with ``greatest'' meaning ``most favourable to firm~1''. The trick is2418categorical---an adjunction with one side dualized is still an adjunction---and it is special to two2419players: with three firms there is no consistent reversal, and genuinely different behaviour appears.2420The best-response map of the linear duopoly has slope $-\tfrac12$, so it is a contraction and iterated2421best responses converge; with three firms the slopes sum to $-1$ and the iteration cycles; with more it2422diverges \cite{The60}.2423 2424\section{Price competition}2425 2426\begin{example}[Bertrand]\label{ex:bertrand}2427If instead the two firms choose prices and the good is homogeneous, the lower-priced firm takes the whole2428market. Any price pair above cost invites undercutting, and the unique equilibrium is $p_1=p_2=c$: two2429firms suffice for the competitive outcome. This is the \emph{Bertrand paradox}. The payoff is discontinuous, and the2430conclusion is fragile: capacity limits, product differentiation, repeated interaction and incomplete2431information each overturn it, and \cref{ch:pricing} examines them.2432 2433With differentiated products, let demand be $q_i=a-p_i+sp_j$ with $0\le s<2$. Profit $(p_i-c)(a-p_i+sp_j)$ has best response2434\[ p_i=\frac{a+c+sp_j}{2}, \]2435which slopes \emph{upward}: prices are \term{strategic complements}. The game is supermodular, and its unique equilibrium is2436$p^*=\frac{a+c}{2-s}$. An increase in $a$ or in $c$ raises both prices, by part (iii) of the theorem,2437without any computation.2438\end{example}2439 2440The distinction between substitutes and complements governs how a firm should try to influence a rival's2441later choice. If actions are strategic substitutes, a commitment to be aggressive makes the rival retreat;2442if complements, it makes the rival aggressive too. The whole taxonomy of strategic investment \cite{FT84}2443is the sign of a cross partial derivative composed with the sign of the effect of the investment, and we2444return to it under the heading of commitment in \cref{ch:perfection}.2445 2446\section{Ordinal conditions}2447 2448Supermodularity is a cardinal condition: it is not preserved by monotone transformations of the2449objective, although the conclusion of Topkis's theorem obviously is. From the standpoint of2450\cref{prin:invariance} this is a defect of the hypothesis, and Milgrom and Shannon repaired it.2451 2452\begin{theorem}[Monotone comparative statics]\label{thm:MS}\Scl[\cite{MS94}]2453Let $X$ be a lattice and $T$ a poset. The correspondence $(t,S)\mapsto\argmax_{x\in S}f(x,t)$ is2454increasing in $t$ and in the sublattice $S$ (in the strong set order) if and only if $f$ is2455\term{quasi-supermodular} in $x$ and satisfies the \term{single-crossing property} in $(x,t)$: for $x'>x$2456and $t'>t$, $f(x',t)\ge f(x,t)$ implies $f(x',t')\ge f(x,t')$, and likewise with strict inequalities.2457\end{theorem}2458 2459The conditions are ordinal, and the theorem is an equivalence: single crossing is exactly what monotone2460comparative statics requires, no more. For games with von Neumann--Morgenstern payoffs the relevant2461invariance is only affine and the cardinal conditions are appropriate; for decision problems under2462certainty the ordinal ones are. Single crossing will carry Part~II: it is the condition under which types2463can be sorted by their choices, and so the condition under which screening and signalling work2464(\cref{ch:mechanism,ch:signalling}).2465 2466\section{The metric analogue}\label{sec:banach}2467 2468Lawvere observed \cite{Law73} that a category enriched in the ordered monoid $([0,\infty],\ge,+)$ is a2469generalized metric space: the hom-object $d(x,y)$ is a distance, composition is the triangle inequality2470$d(x,y)+d(y,z)\ge d(x,z)$, and the identity is $0\ge d(x,x)$. Enriched functors are the maps that do not2471increase distance. The table compares the two enrichments.2472 2473\begin{center}\small2474\renewcommand{\arraystretch}{1.25}2475\begin{tabular}{@{}lll@{}}2476\toprule2477 & \textbf{enriched in $\mathbf{2}$} & \textbf{enriched in $[0,\infty]$}\\\midrule2478object & preorder & generalized metric space\\2479functor & monotone map & nonexpansive map\\2480completeness & complete lattice & Cauchy completeness\\2481good endomaps & monotone & contractions\\2482fixed-point theorem & Tarski: a lattice of fixed points & Banach: a unique fixed point\\2483proof & transfinite iteration & geometric convergence\\2484\bottomrule2485\end{tabular}2486\end{center}2487 2488\noindent Game theory uses both columns, and the difference between them is the difference between2489multiplicity and uniqueness. A monotone operator has a least and a greatest fixed point and possibly many2490between; a contraction has one. When a strategic situation is governed by a monotone operator---best2491responses in a supermodular game, the Abreu--Pearce--Stacchetti operator of \cref{ch:repetition}, the2492unraveling operator of \cref{ch:adverse}---expect a lattice of equilibria and ask which end is selected.2493When it is governed by a contraction---the Bellman operator of a discounted decision problem, the2494Shaked--Sutton recursion for alternating-offer bargaining in \cref{ch:bargaining}---expect uniqueness.2495Discounting is what turns the first kind of problem into the second, when it does.2496 2497\section*{Notes}2498 2499Tarski's theorem is \cite{Tar55}. The lattice-theoretic approach to optimization and games is due to2500Topkis \cite{Top78,Top79}; his book \cite{Top98} is the standard reference. Vives \cite{Viv90} and Milgrom