Indexed text attachment
games-and-information-in-higher-categories.tex
Source for Games and Information in Higher Categories
View paper · Download original file
Lines 4499–4598 of 6559. Source text is displayed literally.
4499Exhibit two different experiments with two signals each that induce the same $\tau$.4500\end{exercise}4501\begin{exercise}4502In \cref{ex:career} let $c(a)=\tfrac12a^2$. Compute equilibrium effort, and compare it with the efficient4503effort $c'(a)=1$. For which parameters does the manager work too hard?4504\end{exercise}4505%% ---- end ch/15-signalling.tex ----4506\part{Applications}4507%% ---- begin ch/16-bargaining.tex ----4508\chapter{Bargaining}\label{ch:bargaining}4509 4510Two parties can create a surplus by agreeing and must divide it. There are two classical theories of the4511division. Nash's axiomatic theory asks what a solution must look like if it respects the symmetries of the4512problem. Rubinstein's strategic theory writes down a game of offers and counteroffers and finds its4513equilibrium. This chapter presents each in the terms developed in Part~I---the first as an equivariant4514natural map, the second as the unique fixed point of a contraction---and then shows, with the tools of4515\cref{ch:mechanism}, why bargaining under two-sided private information cannot be efficient.4516 4517\section{The axiomatic solution}4518 4519\begin{definition}4520A \term{bargaining problem} is a pair $(S,d)$ of a compact convex set $S\subset\R^2$ of feasible utility4521pairs and a disagreement point $d\in S$, with some $s\in S$ strictly better than $d$ for both. A4522\term{solution} assigns to each problem a point of $S$.4523\end{definition}4524 4525\begin{theorem}[Nash]\label{thm:nash-bargaining}\Scl[\cite{Nas50b}]4526There is exactly one solution that is (i) Pareto efficient, (ii) symmetric, (iii) invariant under positive4527affine transformations of either player's utility, and (iv) independent of irrelevant alternatives: if4528$S'\subseteq S$ contains the solution of $(S,d)$, then it is the solution of $(S',d)$. It is4529\[ \phi(S,d)=\argmax_{s\in S,\ s\ge d}\ (s_1-d_1)(s_2-d_2). \]4530\end{theorem}4531 4532Read the axioms as statements about a category. Let the objects be bargaining problems. Axioms (ii) and4533(iii) say that the solution is \emph{equivariant} under the group generated by exchanging the players and4534rescaling utilities: it is a function on the type of bargaining problems \emph{up to} the equivalences that4535von Neumann--Morgenstern utility imposes, in the sense of \cref{prin:invariance} and \cref{thm:vnm}. They are4536not substantive assumptions about bargaining; they are the requirement that the question be well posed.4537Axiom (iv) is of a different kind. It says that the solution is \emph{natural} with respect to a class of4538non-invertible morphisms, the inclusions $S'\hookrightarrow S$ that keep the solution feasible. This is the4539substantive axiom, and different choices of morphisms give different theories: requiring instead4540monotonicity with respect to expansions of $S$ that preserve each player's best feasible payoff4541characterizes the solution of Kalai and Smorodinsky \cite{KS75}. The axiomatic method in bargaining is the4542choice of a category on which the solution is to be a natural transformation.4543 4544\begin{example}\label{ex:nash-risk}4545Two players divide one unit. Player~1 has utility $x$ and player~2 has utility $\sqrt{1-x}$; disagreement4546gives both zero. The Nash solution maximizes $x\sqrt{1-x}$, so $x=\tfrac23$. The more risk-averse player4547receives the smaller share: concavity makes her less willing to hold out for more.4548\end{example}4549 4550\section{The strategic solution}4551 4552\begin{definition}4553In the \term{alternating-offers game}, player~1 proposes a division of a unit surplus; player~2 accepts,4554ending the game, or rejects, in which case a period passes and player~2 proposes; and so on. Player $i$4555discounts by $\delta_i\in(0,1)$ per period.4556\end{definition}4557 4558With a finite number $T$ of rounds, after which disagreement gives zero, the game is an element of the4559initial algebra of \cref{ch:perfection} and is solved by a fold. In the last round the proposer takes4560everything. In the round before, the proposer must offer the other her discounted continuation value. With4561a common discount factor the first proposer's share is4562\[ x_T=1-\delta+\delta^2-\cdots\pm\delta^{T-1}=\frac{1-(-\delta)^T}{1+\delta}, \]4563which oscillates and converges to $\frac{1}{1+\delta}$.4564 4565\begin{theorem}[Rubinstein]\label{thm:rubinstein}\Scl[\cite{Rub82,SS84b}]4566The infinite-horizon alternating-offers game has a unique subgame perfect equilibrium. Player~1 proposes4567to keep4568\[ x^*=\frac{1-\delta_2}{1-\delta_1\delta_2}, \]4569player~2 accepts at once, and each player always proposes the same division and accepts exactly the offers4570at least as good as her discounted proposer share.4571\end{theorem}4572 4573\begin{proof}[Proof (Shaked--Sutton)]4574The game is stationary: the subgame beginning at each of player $i$'s proposals is the same game. Let $M_i$4575and $m_i$ be the supremum and infimum of $i$'s subgame-perfect payoffs in a subgame where she proposes. A4576responder gets at most $\delta_jM_j$ by rejecting, so she accepts anything better, and the proposer can4577secure $m_i\ge1-\delta_jM_j$. A responder can secure $\delta_jm_j$ by rejecting, so $M_i\le1-\delta_jm_j$.4578Combining the four inequalities,4579\[ M_1\le\varphi(M_1),\qquad m_1\ge\varphi(m_1),\qquad\text{where }\varphi(x)=1-\delta_2\,(1-\delta_1x). \]4580The affine map $\varphi$ is increasing with slope $\delta_1\delta_2<1$, so $x\le\varphi(x)$ iff $x\le x^*$4581and $x\ge\varphi(x)$ iff $x\ge x^*$, where $x^*$ is its fixed point. Hence $M_1\le x^*\le m_1$.4582\end{proof}4583 4584The proof is worth comparing with the analysis of repeated games in \cref{ch:repetition}. There too one4585considered the set of subgame-perfect payoffs and an operator under which it is a fixed point, and the4586operator $B$ was monotone, with a greatest fixed point much larger than the least. Here the operator4587acts on the interval $[m_1,M_1]$ and is a \emph{contraction}, with modulus $\delta_1\delta_2$; a contraction4588has one fixed point, and the interval collapses. In the terms of \cref{thm:guarded}, bargaining is a4589\emph{guarded} recursion: the continuation value enters each player's decision only as a single number,4590scaled by a discount factor. In a repeated game the continuation enters as a function from histories to4591promised values, whose \emph{spread} supplies incentives, and discounting does not shrink the spread4592relative to what it has to deter. That is the structural reason why one infinite-horizon game has a unique4593equilibrium and the other has a folk theorem.4594 4595A corollary: the infinite-horizon solution is the limit of the finite-horizon ones, $x_T\to\frac{1}{1+\delta}$.4596Under a guard the initial algebra and the terminal coalgebra give the same answer, and the question4597whether there is a last period---so vexed for repeated games---does not matter.4598