Indexed text attachment
games-and-information-in-higher-categories.tex
Source for Games and Information in Higher Categories
View paper · Download original file
Lines 4641–4740 of 6559. Source text is displayed literally.
4641receives $v$: each captures the entire gain $v-c$. The lowest-value buyer and the highest-cost seller never4642trade and get zero, so participation forbids lowering either constant. The two traders together must4643therefore receive at least twice the expected gains from trade, while only the gains themselves are4644available. The shortfall is $\E[(v-c)^+]>0$, and someone outside would have to pay it. (When the intervals4645merely overlap, the worst types may earn something in the Groves mechanism, the constants can be lowered by4646that much, and the computation of \cite{MS83} shows that a positive shortfall remains.)4647\end{proof}4648 4649The theorem is an \emph{obstruction}, and its proof has the form that obstructions have had throughout this4650book. Incentive compatibility and the allocation rule determine payments up to constants, because the type4651spaces are connected. Participation bounds the constants on one side; budget balance fixes their sum. The4652three requirements are a system of equations and inequalities with no solution. Nothing in the argument4653refers to offers, counteroffers or deadlines: no bargaining institution whatever can do better.4654 4655\begin{example}\label{ex:CS}\Scl[\cite{CS83,MS83}]4656Let $v$ and $c$ be independent and uniform on $[0,1]$. In the sealed-bid double auction, where trade occurs4657at the average of the bids if the buyer's bid exceeds the seller's, there is an equilibrium in linear4658strategies4659\[ b(v)=\tfrac23v+\tfrac1{12},\qquad s(c)=\tfrac23c+\tfrac14, \]4660in which trade occurs iff $v\ge c+\tfrac14$. The expected gains from trade are $\tfrac9{64}$, against4661$\tfrac16$ under efficiency: a fraction $\tfrac{27}{32}$ is realized. Myerson and Satterthwaite showed that4662this is the best that any mechanism satisfying the other three conditions can do. The parties fail to4663trade when the gains are positive but small, each having shaded her bid to improve the price.4664\end{example}4665 4666The Coase theorem says that, absent transaction costs, parties bargain to an efficient outcome whatever4667the initial assignment of rights. \Cref{thm:MyS} identifies private information as a transaction cost in4668this sense, and an irreducible one. It also shows that the assignment of rights matters: if ownership is4669shared rather than assigned to one party, the participation constraints are relaxed and efficient4670dissolution can become possible \cite{CGK87}.4671 4672\section*{Notes}4673 4674The axiomatic theory is Nash's \cite{Nas50b}; the monotonic alternative is Kalai and Smorodinsky's4675\cite{KS75}. The strategic theory is due to Rubinstein \cite{Rub82}, following St\aa hl's finite-horizon4676analysis; the proof given is Shaked and Sutton's \cite{SS84b}. The relation between the two, and the role of4677outside options, are worked out in \cite{BRW86}. The impossibility theorem is \cite{MS83}; the linear4678equilibrium of the double auction is from Chatterjee and Samuelson \cite{CS83}; efficient dissolution of4679partnerships is \cite{CGK87}. A standard text is \cite{OR90}.4680 4681\section*{Exercises}4682 4683\begin{exercise}4684Verify \cref{ex:nash-risk}, and compute the Nash solution when player~2's utility is $(1-x)^\gamma$ with4685$0<\gamma\le1$. How does her share vary with $\gamma$?4686\end{exercise}4687\begin{exercise}4688Show that the Nash solution is equivariant under $(s_1,s_2)\mapsto(\alpha_1s_1+\beta_1,\alpha_2s_2+\beta_2)$, and give an4689example showing that it is not monotone with respect to expansions of $S$.4690\end{exercise}4691\begin{exercise}4692Compute the finite-horizon shares $x_T$ for $T=1,\dots,5$ and $\delta=0.9$, and verify geometric convergence4693to $\frac{1}{1+\delta}$ at rate $\delta$.4694\end{exercise}4695\begin{exercise}4696In the infinite-horizon game suppose that player~2, whenever she rejects an offer, may leave and take an4697outside option worth $w$. Find the subgame perfect equilibrium for $w<\delta_2(1-\delta_1x^*)$ and for larger4698$w$.4699\end{exercise}4700\begin{exercise}4701Verify that the strategies of \cref{ex:CS} are mutual best responses, and compute the expected gains from4702trade.4703\end{exercise}4704\begin{exercise}4705For $v$ and $c$ uniform on $[0,1]$, compute the expected budget deficit of the Groves mechanism with4706efficient trade in which the worst types get zero, and confirm that it equals $\E[(v-c)^+]=\tfrac16$.4707\end{exercise}4708%% ---- end ch/16-bargaining.tex ----4709%% ---- begin ch/17-auctions.tex ----4710\chapter{Auctions}\label{ch:auctions}4711 4712An auction is a mechanism for selling to buyers whose willingness to pay the seller does not know. It is4713the part of the subject where theory, experiment and practice are closest, and where the apparatus of4714\cref{ch:mechanism} pays its way most visibly: the central theorem of the field is one line once payments4715are seen as potentials.4716 4717\section{Formats, and two kinds of equivalence}4718 4719Four formats are standard. In the \emph{English} auction the price rises until one bidder remains. In the4720\emph{Dutch} auction the price falls until someone claims the object. In the \emph{first-price} sealed-bid4721auction the highest bidder wins and pays her bid; in the \emph{second-price} auction she pays the highest4722losing bid. Values are \emph{private} if each bidder knows her own value and it does not depend on what4723others know, and \emph{common} if the object has the same unknown value to all and bidders hold different4724estimates of it.4725 4726\begin{proposition}\label{prop:auction-equiv}\Scl[\cite{Vic61}]4727\begin{enumerate}[label=(\roman*)]4728\item The Dutch and first-price auctions have the same reduced strategic form.4729\item With private values, the English and second-price auctions each have an equilibrium in weakly4730 dominant strategies, with the same outcome.4731\end{enumerate}4732\end{proposition}4733 4734The two statements are equivalences of different types, in the sense of \cref{sec:thompson}. In a Dutch4735auction a strategy is a price at which to stop the clock, nothing is learned before the auction ends, and4736the bidder who names the highest price wins and pays it: the tree differs from that of the first-price4737auction by Thompson moves, and the two are the same point of $\EFG_{\mathrm{T}}$. This holds for any values4738and any information. The second equivalence is not an identification of games. An English auction reveals4739when others drop out, and has many more strategies than a sealed bid. With private values that information4740is useless and the outcomes coincide; with common values it is valuable and they do not. One equivalence