Indexed text attachment
games-and-information-in-higher-categories.tex
Source for Games and Information in Higher Categories
View paper · Download original file
Lines 3379–3478 of 6559. Source text is displayed literally.
3379hidden knowledge & contract $\to$ $\theta$ $\to$ report or action & a state learned after contracting\\3380adverse selection & $\theta$ $\to$ contract & the type, known before contracting\\3381signalling & $\theta$ $\to$ signal $\to$ contract & the type; the informed party moves first\\3382screening & $\theta$ $\to$ menu $\to$ choice & the type; the uninformed party moves first\\3383\bottomrule3384\end{tabular}3385\end{center}3386 3387\noindent The first two are called \emph{moral hazard}, because the asymmetry arises after the parties are3388bound; the last three, in which the agent is informed at the outset, are the province of3389\cref{ch:adverse,ch:mechanism,ch:signalling}. As open games (\cref{ch:open}) the five are five ways of composing3390the same boxes: a move of Nature, a decision of the principal, a decision of the agent, and an observation3391channel that limits which wires may run into which box. What is constant is the shape of the principal's3392problem. She does not choose the agent's action. She chooses the agent's \emph{context}, and the action is3393whatever the agent's selection function returns in that context.3394 3395\section{The principal--agent problem}3396 3397An agent chooses an effort $e\in E$. Effort determines a distribution over observable outcomes $x\in X$:3398the \emph{technology} is a Markov kernel $f\colon E\to X$, written $f(x\mid e)$. A \term{contract} is a function3399$w\colon X\to\R$ specifying the agent's pay. The agent has utility $u(w)-c(e)$, with $u$ increasing and3400concave and $c$ the cost of effort, and an outside option worth $\overline{U}$. The principal is risk3401neutral and receives $x-w(x)$.3402 3403In the vocabulary of \cref{def:lens}, the contract composed with $u$ is a costate on the outcome boundary: a3404function assigning value to each observable result. Pulling it back along the technology gives the3405agent's decision problem,3406\[ k_w(e)=\E_{x\sim f(e)}\bigl[u(w(x))\bigr]-c(e), \]3407and the agent selects $e\in\argmax k_w$. The principal, moving first, chooses $w$ to maximize her own payoff3408at the selected effort. This is the product of selection functions of \cref{sec:selection}, with one3409addition, the agent's right to refuse:3410\begin{align*}3411\max_{w,\,e}\quad&\E_{f(e)}[x-w(x)]\\3412\text{subject to}\quad&\E_{f(e)}[u(w(x))]-c(e)\ \ge\ \overline{U} &&\text{(participation)}\\3413&e\in\argmax_{e'}\ \E_{f(e')}[u(w(x))]-c(e') &&\text{(incentive compatibility)}.3414\end{align*}3415The incentive constraint is not an extra assumption; it is the statement that the second player in a3416sequential composite best-responds.3417 3418\subsection*{The first best}3419 3420If effort were observable the contract could depend on it, incentive compatibility would be vacuous, and3421the principal would solve an insurance problem. A risk-neutral party should bear all the risk: the optimal3422contract pays a constant wage $w^*$ with $u(w^*)=\overline{U}+c(e)$ when the agreed effort is supplied3423(\emph{Borch's rule}). This is the \term{first best}. With hidden effort a constant wage induces the least3424costly effort. The agent must be exposed to the outcome in order to care about it, and exposure to risk3425is costly to a risk-averse agent. That conflict between insurance and incentives is the whole subject.3426 3427There is one case with no conflict. If the agent is risk neutral, the contract $w(x)=x-K$ makes the agent3428the residual claimant of the entire outcome. She then chooses the efficient effort, bears all the risk at3429no cost, and pays the principal the fixed fee $K$. Moral hazard is a problem only in combination with risk3430aversion, or with a limit on how much the agent can be made to pay.3431 3432\subsection*{The second best}3433 3434\begin{example}[Two efforts, two outcomes]\label{ex:pa}3435Effort is high or low. The project succeeds with probability $0.8$ under high effort and $0.4$ under low.3436The agent has $u(w)=\sqrt{w}$, $c(H)=1$, $c(L)=0$, $\overline{U}=1$; a success is worth $20$ to the3437principal and a failure $0$.3438 3439\emph{Observable effort.} To obtain $H$ the principal pays the constant wage with $\sqrt{w}=2$: $w=4$, for a3440profit of $16-4=12$. To obtain $L$ she pays $w=1$ for a profit of $8-1=7$.3441 3442\emph{Hidden effort.} Let $u_s,u_f$ be the agent's utility from pay after success and failure. Incentive3443compatibility for $H$ is $0.8u_s+0.2u_f-1\ge0.4u_s+0.6u_f$, that is $u_s-u_f\ge2.5$; participation is3444$0.8u_s+0.2u_f-1\ge1$. Both bind at the optimum: $u_f=0$, $u_s=2.5$, so $w_f=0$ and $w_s=6.25$. The expected3445wage is $5$, and the profit is $11$. The principal still prefers high effort, but pays an \emph{agency cost}3446of $1$: the expected wage exceeds the first-best wage by exactly the risk premium the agent requires for3447carrying the incentive scheme.3448\end{example}3449 3450With a continuum of efforts and outcomes, and when the agent's problem can be replaced by its first-order3451condition, the optimal contract satisfies3452\begin{equation}\label{eq:holmstrom}3453\frac{1}{u'(w(x))}=\lambda+\mu\,\frac{f_e(x\mid e)}{f(x\mid e)}3454\end{equation}3455for multipliers $\lambda,\mu>0$ \cite{Hol79}. Pay depends on the outcome only through the likelihood ratio3456$f_e/f$: the contract rewards outcomes not because they are good for the principal but because they are3457\emph{evidence} of effort. If the likelihood ratio is increasing in $x$ (the monotone likelihood ratio3458property), pay is increasing in $x$. The first-order approach is valid under further convexity conditions3459\cite{Rog85}, and can fail without them.3460 3461\section{Informativeness}\label{sec:informativeness}3462 3463Formula \eqref{eq:holmstrom} says that the contract should use exactly the information about effort3464contained in the outcome. This can be made into a theorem that involves no first-order conditions, and it3465is here that the Markov-category language of \cref{ch:belief} earns its place.3466 3467\begin{definition}\Scl[\cite{Bla51}; categorical form \cite{Fri20,FGPR23}]3468Let $f\colon E\to X$ and $g\colon E\to Y$ be Markov kernels out of the same object (two \emph{experiments} on $E$).3469$g$ is a \term{garbling} of $f$, written $g\preceq f$, if $g=h\circ f$ for some kernel $h\colon X\to Y$. A3470statistic $s\colon X\to V$ is \term{sufficient} for $f$ if $f$ is in turn a garbling of $s\circ f$.3471\end{definition}3472 3473So the informativeness order on experiments is the factorization preorder on the coslice under $E$ in the3474category of kernels, and a sufficient statistic is a deterministic map that loses nothing in this order.3475Blackwell's theorem says that $g\preceq f$ iff every Bayesian decision maker weakly prefers observing $f$ to3476observing $g$. In an agency problem the decision maker and the source of the uncertainty are different3477people, and the corresponding statement needs its own proof.3478