Indexed text attachment
games-and-information-in-higher-categories.tex
Source for Games and Information in Higher Categories
View paper · Download original file
Lines 5999–6098 of 6559. Source text is displayed literally.
5999\sol{10.5} States: $(0,0)$ in state $a$, and $(n,n-1)$, $(n,n)$ for $n\ge1$ in state $b$. Player~1's cells6000are $\{(0,0)\}$ and $\{(n,n-1),(n,n)\}$; player~2's are $\{(n,n),(n+1,n)\}$ for $n\ge0$. The graph is the6001path $(0,0)-(1,0)-(1,1)-(2,1)-\cdots$, with edges alternately cells of~2 and of~1. $\K_1E=E$, everything6002but $(0,0)$; $\K_2\K_1E$ removes $(1,0)$; $\K_1\K_2\K_1E$ removes $(1,1)$ as well. Each further operator6003removes one more state, and the intersection is empty.6004 6005\sol{10.6} Monotonicity is that of probability. If $\Prob(E)\ge p$ and $\Prob(F)\ge p$ then6006$\Prob(E\cap F)\ge2p-1$. If a cell is contained in $E$ its probability of $E$ is $1$, so6007$\K_iE\subseteq\mathsf{B}^1_iE$, with equality when posteriors have full support. As for the last question:6008only loosely. The family is antitone in $p$; it has a counit only at grade $1$ and only up to null events;6009each $\mathsf{B}^p_i$ is idempotent, since $\mathsf{B}^p_iE$ is a union of cells; and conjunction degrades6010grades by $p\odot q=\max\{0,p+q-1\}$. It is a lax monoidal family graded by the unit interval with that6011product, comonadic at the top grade.6012 6013\section*{Chapter 11}6014 6015\sol{11.1} With effort observable she prefers $H$ iff $0.8v-4\ge0.4v-1$, that is $v\ge7.5$. With hidden6016effort she prefers $L$ iff $0.8v-5<0.4v-1$, that is $v<10$. The range is $7.5\le v<10$.6017 6018\sol{11.2} Incentive compatibility is $0.4(w_s-w_f)\ge1$. With $w\ge0$ the cheapest contract is $w_f=0$,6019$w_s=2.5$, expected wage $2$. The agent's expected utility is $2-1=1>0$: limited liability leaves her a6020rent of $1$, which a negative $w_f$ would otherwise extract.6021 6022\sol{11.3} The likelihood ratio of low to high effort for the pair is the product of the ratios for the6023outcome and for the signal, and the latter is $\tfrac59$ at $y=1$ and $5$ at $y=0$; so the pair is more6024informative than the outcome. Under $H$ the four results (success/failure, $y$) have probabilities6025$0.72,0.08,0.18,0.02$, and under $L$ $0.2,0.2,0.3,0.3$. Minimizing $\sum p_Hu_z^2$ subject to the two6026binding constraints and $u_z\ge0$ gives $u_z=\max\{0,\ a+m(1-\ell_z)\}$ with $\ell_z$ the likelihood ratio, and6027the solution has $u=0$ at (failure, $y=0$) and6028\[ u\approx(2.378,\ 0.663,\ 1.306,\ 0),\qquad w\approx(5.65,\ 0.44,\ 1.71,\ 0), \]6029with expected wage $\approx4.41$. The agency cost falls from $1$ to about $0.41$.6030 6031\sol{11.4} Reflexivity: $h=\id$. Transitivity: garblings compose. For any deterministic $s$, $s\circ f$ is6032by definition a garbling of $f$. Sufficiency adds $f\preceq s\circ f$, so $s$ is sufficient iff $f$ and6033$s\circ f$ are equivalent in the preorder.6034 6035\sol{11.5} The surplus $\frac\beta c-\frac{\beta^2}{2c}-\frac12r\sigma^2\beta^2$ has first-order condition6036$1-\beta(1+rc\sigma^2)=0$. At the optimum it equals $\frac{\beta^*}{2c}$. With observable effort $e=\frac1c$ and6037full insurance the surplus is $\frac1{2c}$. The agency cost is6038$\frac1{2c}\bigl(1-\beta^*\bigr)=\frac{r\sigma^2}{2(1+rc\sigma^2)}$.6039 6040\sol{11.6} See \cref{fig:sol-pa}. The payment box receives the contract $w$ and the outcome $x$. With6041observable effort it would also receive a copy of $e$; that wire, dotted in the figure, is what is missing.6042 6043\begin{figure}[ht]\centering6044\begin{tikzpicture}[x=1cm,y=1cm]6045 \node[ogbox] (P) at (0,0) {$\mathcal{P}$};6046 \node[ogbox] (A) at (2.4,0) {$\mathcal{A}$};6047 \node[ogbox] (N) at (4.8,0) {$f$};6048 \node[ogbox] (Pay) at (7.4,1.6) {pay};6049 \node[ogbox,minimum height=3.6cm] (U) at (9.8,0.4) {$u$};6050 \node[dec] (cw) at (1.2,0) {}; \node[dec] (ce) at (3.6,0) {}; \node[dec] (cx) at (6.1,0) {};6051 \draw[thick,fwd] (P) -- node[above=1pt,lab,black]{$w$} (cw); \draw[fw] (cw) -- (A);6052 \draw[thick,fwd] (A) -- node[above=1pt,lab,black]{$e$} (ce); \draw[fw] (ce) -- (N);6053 \draw[thick,fwd] (N) -- node[above=1pt,lab,black]{$x$} (cx); \draw[fw] (cx) -- (U.west|-0,0);6054 \draw[fw] (cw) |- (Pay.west|-0,1.9);6055 \draw[fw] (cx) |- (Pay.west|-0,1.3);6056 \draw[densely dotted,very thick,gray] (ce) |- (Pay.west|-0,1.6);6057 \draw[fw] (Pay) -- node[above=1pt,lab,black]{$t$} (U.west|-0,1.6);6058 \draw[fw] (ce) |- (U.west|-0,-1.0);6059\end{tikzpicture}6060\caption{The principal--agent problem with hidden action: the principal chooses a contract, the agent an6061effort, Nature the outcome. Backward wires are omitted. The dotted wire, a copy of $e$ into the payment6062rule, is present only when effort is observable.}\label{fig:sol-pa}6063\end{figure}6064 6065\section*{Chapter 12}6066 6067\sol{12.1} A non-shirker has $V_N=w-e+\delta[(1-s)V_N+sV_U]$, so $V_N-V_U=\frac{w-e-\overline u}{1-\delta(1-s)}$.6068Shirking once raises the probability of losing the job from $s$ to $s+(1-s)q$, so the condition is6069$\delta(1-s)q\,(V_N-V_U)\ge e$, that is6070\[ w\ \ge\ \overline u+e+\frac{e\,[1-\delta(1-s)]}{\delta(1-s)\,q}. \]6071The last term is $\frac eq\bigl(\frac{1}{\delta(1-s)}-1\bigr)$, increasing in $s$: a job that may end anyway is6072a weaker hostage.6073 6074\sol{12.2} The difference of two independent uniforms on $[-a,a]$ has a triangular density with6075$g(0)=\frac1{2a}$. First-best effort is $e=1$, so the spread is $W_1-W_2=2a$. Participation binds when6076$\tfrac12(W_1+W_2)-\tfrac12=0$: $W_1=\tfrac12+a$, $W_2=\tfrac12-a$. One must check that $e=1$ is a global best6077response. For $e<1$ the marginal net benefit is $(1-e)\bigl(1-\frac1{2a}\bigr)$, which is positive only if6078$a>\tfrac12$. With less noise than that, the first-order condition picks out a minimum and no symmetric6079pure-strategy equilibrium implements the first best.6080 6081\sol{12.3} Efficient efforts are $e_i=1$ with surplus $1$. Under equal sharing each maximizes6082$\tfrac12(e_1+e_2)-\tfrac12e_i^2$, so $e_i=\tfrac12$ and the surplus is $\tfrac34$, a loss of $\tfrac14$. Under the scheme paying6083$b_i=1$ iff $x\ge2$: if the other works $1$, working $1$ yields $\tfrac12$, and any lower effort yields at most6084$0$. So $(1,1)$ is an equilibrium. So is $(0,0)$: reaching the target alone costs $2$ and pays $1$. So is6085every $(e_1,2-e_1)$ with $2-\sqrt2\le e_1\le\sqrt2$. The scheme makes efficiency an equilibrium, not the6086equilibrium.6087 6088\sol{12.4} The agent's certainty equivalent is $\alpha+\beta^{\top}e-\tfrac12e^{\top}Ce-\tfrac r2\beta^{\top}\Sigma\beta$,6089so $e=C^{-1}\beta$, and total surplus is $b^{\top}C^{-1}\beta-\tfrac12\beta^{\top}C^{-1}\beta-\tfrac r2\beta^{\top}\Sigma\beta$.6090The first-order condition $C^{-1}(b-\beta)=r\Sigma\beta$ gives $\beta=(I+rC\Sigma)^{-1}b$. With $\beta_2=0$,6091$e=\beta_1C^{-1}(1,0)^{\top}$ and the surplus in $\beta_1$ is quadratic with maximizer6092$\frac{b_1C_{22}-b_2C_{12}}{C_{22}+r\sigma_1^2\det C}$, which is the stated formula. In the example6093$\beta_1^*=\frac{1-2\kappa}{2-\kappa^2}$: a flat wage at $\kappa=\tfrac12$, and a penalty on measured output6094for $\tfrac12<\kappa<1$.6095 6096\sol{12.5} Let the contracts be an incentive scheme $w_0$ and full insurance $w_1$. At the renegotiation6097stage effort is sunk and both parties prefer $w_1$ to $w_0$, while nothing is preferred by both to $w_1$.6098So $\rho(w_0)=\rho(w_1)=w_1$ and $\rho\rho=\rho$. The principle: the outcome of signing $w$ and then