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 3881–3980 of 6559. Source text is displayed literally.

3881competitive market maker sets the ask at the expected value conditional on a buy order and the bid at the3882expected value conditional on a sale:3883\[ \text{ask}=\Prob(v=1\mid\text{buy})=\frac{1+\mu}{2},\qquad\text{bid}=\frac{1-\mu}{2}. \]3884The spread equals $\mu$. It is not a cost of doing business but a transfer: the market maker loses to the3885informed and recovers the loss from the uninformed, and the spread is the price of the adverse selection3886in the order flow.3887\end{example}3888 3889\begin{example}[Kyle's lambda]\label{ex:kyle}\Scl[\cite{Kyl85}]3890An asset has value $v\sim N(p_0,\Sigma_0)$. An insider who knows $v$ submits an order $x$; noise traders submit3891$u\sim N(0,\sigma_u^2)$; the market maker sees the total $y=x+u$ and sets the price $p=\E[v\mid y]$. Look for a3892linear equilibrium $x=\beta(v-p_0)$, $p=p_0+\lambda y$. The insider maximizes3893$\E[(v-p)x]=(v-p_0-\lambda x)x$, so $\beta=\frac{1}{2\lambda}$. The market maker's rule is a regression,3894$\lambda=\frac{\beta\Sigma_0}{\beta^2\Sigma_0+\sigma_u^2}$. Solving,3895\[ \lambda=\frac{\sqrt{\Sigma_0}}{2\sigma_u},\qquad\beta=\frac{\sigma_u}{\sqrt{\Sigma_0}},\qquad\Var(v\mid p)=\frac{\Sigma_0}{2}. \]3896The insider trades less aggressively the more her trades move the price; the price moves more the more3897informative the order flow; and in equilibrium exactly half of her information is impounded in the price,3898whatever the amount of noise.3899\end{example}3900 3901Both examples have the shape described in \cref{ch:belief}: a forward kernel from the hidden state to an3902observable (the order), and a backward map from the observable to a belief about the state (the price),3903which is its Bayesian inverse. The informed trader chooses the forward kernel taking the backward map as3904given; the market maker's backward map must be the inverse of the forward kernel actually chosen. A3905rational-expectations equilibrium is a fixed point of this pair---a Bayesian lens that is consistent with3906the best response it induces. The same structure, with a manager's unobserved action in place of the3907insider's order, is the signal-jamming model of \cref{ch:signalling}.3908 3909\section*{Notes}3910 3911The market for lemons is Akerlof's \cite{Ake70}; the lattice structure of its equilibria and their Pareto3912ranking are discussed in \cite{Wil80,MWG95}. The insurance model is due to Rothschild and Stiglitz3913\cite{RS76}; the reactive equilibrium concepts are from \cite{Wil77}, mixed-strategy existence from3914\cite{DM86}, and the price-theoretic equilibrium from Azevedo and Gottlieb \cite{AG17}. The sequential3915trade model of the bid--ask spread is Glosten and Milgrom's \cite{GM85}; the batch-auction model is Kyle's3916\cite{Kyl85}. The source text \cite[ch.~9]{Ras06} develops lemons markets through several variants with3917different distributions of quality and tastes; \cref{ex:lemons,ex:lemons2} are in that spirit with3918different numbers.3919 3920\section*{Exercises}3921 3922\begin{exercise}3923Let $\theta$ be uniform on $[0,1]$, let sellers value a car at $\theta$ and buyers at $\theta+k$ with $k>0$.3924Find all equilibrium prices as a function of $k$ and the fraction of cars traded.3925\end{exercise}3926\begin{exercise}3927Construct a distribution of quality for which $\Phi$ has three fixed points, and verify that they are3928Pareto-ranked.3929\end{exercise}3930\begin{exercise}3931In \cref{ex:RS}, verify $\alpha^*$, compute both types' expected utilities at the separating pair and at3932full insurance at the pooled fair premium when $\lambda=0.1$, and confirm that the pooling contract would3933attract both.3934\end{exercise}3935\begin{exercise}3936In \cref{ex:GM} let the asset be worth $1$ with prior probability $\pi$. Compute the bid and the ask, and3937the market maker's posterior after two consecutive buy orders.3938\end{exercise}3939\begin{exercise}3940In \cref{ex:kyle}, compute the insider's expected profit and the expected loss of the noise traders, and3941show that they are equal. How do they depend on $\sigma_u$?3942\end{exercise}3943%% ---- end ch/13-adverse.tex ----3944%% ---- begin ch/14-mechanism.tex ----3945\chapter{Mechanism Design}\label{ch:mechanism}3946 3947So far the game has been given and the equilibrium sought. Mechanism design reverses the problem: an3948outcome is wanted, as a function of information the designer does not have, and the question is whether3949some game has it as an equilibrium. The answer is organized by one principle, which reduces the search3950over all games to a system of inequalities, and by the structure of that system, which turns out to be a3951piece of enriched category theory.3952 3953\section{The revelation principle}3954 3955An agent has a type $\theta\in\Theta$, known only to her, and utility $U(y,\theta)$ over outcomes $y\in Y$. A3956\term{mechanism} is a set $M$ of messages with an outcome function $g\colon M\to Y$. A strategy is a map3957$\sigma\colon\Theta\to M$; it is a best response if $U(g\sigma(\theta),\theta)\ge U(g(m),\theta)$ for all3958$\theta$ and $m$. The mechanism then \term{implements} the social choice function $f=g\circ\sigma\colon\Theta\to Y$.3959 3960Fix $f$ and consider the category $\mathsf{Impl}(f)$ whose objects are triples $(M,g,\sigma)$ with $\sigma$ a best3961response in $(M,g)$ and $g\sigma=f$, and whose morphisms $(M,g,\sigma)\to(M',g',\sigma')$ are maps3962$h\colon M\to M'$ with $g'h=g$ and $h\sigma=\sigma'$.3963 3964\begin{theorem}[Revelation principle]\label{thm:revelation}\Srf[\cite{Gib73,Mye79}]3965\begin{enumerate}[label=(\roman*)]3966\item (Best responses pull back.) Let $\sigma$ be a best response in $(M,g)$ and let $h\colon M'\to M$ be any map3967  such that $\sigma=h\sigma'$ for some $\sigma'\colon\Theta\to M'$. Then $\sigma'$ is a best response in $(M',gh)$.3968\item (Direct mechanisms are initial.) If $\mathsf{Impl}(f)$ is nonempty, then $(\Theta,f,\id_\Theta)$ is an3969  object of it, and is initial.3970\end{enumerate}3971Hence $f$ is implementable by some mechanism if and only if it is \term{incentive compatible}:3972\[ U(f(\theta),\theta)\ \ge\ U(f(\theta'),\theta)\qquad\text{for all }\theta,\theta'. \]3973\end{theorem}3974 3975\begin{proof}3976(i) Every deviation in $M'$ maps under $h$ to a deviation in $M$:3977$U(gh(m'),\theta)\le U(g\sigma(\theta),\theta)=U(gh\sigma'(\theta),\theta)$.3978(ii) Given $(M,g,\sigma)$, apply (i) with $M'=\Theta$, $h=\sigma$, $\sigma'=\id$: truth-telling is a best response in3979the direct mechanism $(\Theta,g\sigma)=(\Theta,f)$. The map $\sigma$ is a morphism $(\Theta,f,\id)\to(M,g,\sigma)$, and3980any morphism $k$ from the direct mechanism satisfies $k=k\circ\id=\sigma$.