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 6179–6278 of 6559. Source text is displayed literally.

6179$(\lambda,1)$. None survives the intuitive criterion: a signal $e'\in(1,2-\mu)$ is equilibrium-dominated for6180the low type and profitable for the high type if believed.6181 6182\sol{15.2} Under full information each type equates marginal product and marginal cost:6183$\frac{\theta}{4\sqrt e}=\frac1\theta$, so $e_L=\tfrac1{16}$ and $e_H=1$, with payoffs $1.0625$ and $2.5$. In6184the Riley outcome the low type keeps $e_L=\tfrac1{16}$. The high type's education must deter imitation:6185$2+\sqrt e-e\le1.0625$, that is $\sqrt e\ge\tfrac12(1+\sqrt{4.75})$, so $e_H\approx2.53$, well above the6186efficient $1$. Her payoff is about $2.33$. Productive education does not remove the distortion; it changes6187the benchmark from which it is measured.6188 6189\sol{15.3} The boundary type $c_i$ is indifferent between the receiver's actions on the two sides,6190$\tfrac12(c_{i-1}+c_i)$ and $\tfrac12(c_i+c_{i+1})$, so her ideal point $c_i+b$ is their midpoint; this is the6191difference equation. An interval of length $l$ contributes $l\cdot\frac{l^2}{12}$ to the receiver's loss. For6192$b=\tfrac1{20}$: $N=1$ gives $\tfrac1{12}\approx0.0833$; $N=2$ has lengths $\tfrac25,\tfrac35$ and loss6193$\approx0.0233$; $N=3$ has lengths $\tfrac2{15},\tfrac5{15},\tfrac8{15}$ and loss $\approx0.0159$. In general the6194loss is $\frac1{12N^2}+\frac{b^2(N^2-1)}{3}$.6195 6196\sol{15.4} Posteriors $0$ and $0.6$ with mean $0.3$ require weight $\tfrac12$ on each. If effective drugs6197always pass and ineffective ones pass with probability $x$, then $0.3+0.7x=\tfrac12$, so $x=\tfrac27$, and the6198posterior after a pass is $0.3/0.5=0.6$. With prior $0.5$ the weight on $0.6$ is $\tfrac56$: ineffective6199drugs pass with probability $\tfrac23$, and approval has probability $\tfrac56$.6200 6201\sol{15.5} $m(\tau)=\int\mu\,d\tau(\mu)$ is linear in $\tau$, so it is affine and its fibres are convex. Take6202any experiment with two signals inducing two different posteriors and exchange the names of the signals:6203the kernel changes and $\tau$ does not.6204 6205\sol{15.6} Effort solves $a=\delta\kappa$, and the efficient level is $1$. Since $\delta\le1$ and $\kappa<1$,6206she works too little for all parameters. Working too hard requires the reputational return to accumulate6207over several future periods, as in the multi-period version of \cite{Hol99}.6208 6209\section*{Chapter 16}6210 6211\sol{16.1} Maximizing $x(1-x)^\gamma$ gives $1-x=\gamma x$, so $x=\frac1{1+\gamma}$ and player~2 receives6212$\frac{\gamma}{1+\gamma}$: one half at $\gamma=1$, one third at $\gamma=\tfrac12$, and nothing in the limit of6213extreme risk aversion.6214 6215\sol{16.2} Under the transformation the Nash product is multiplied by $\alpha_1\alpha_2$, so maximizers6216correspond. For monotonicity let $S_1$ be the convex hull of $(0,0),(1,0),(0,1),(0.75,0.75)$ and $S_2$ that6217of $(0,0),(1,0),(0,1),(1,0.7)$, with $d=0$. Then $S_1\subseteq S_2$, the solutions are $(0.75,0.75)$ and6218$(1,0.7)$, and player~2 loses from the expansion.6219 6220\sol{16.3} $x_T=1,\ 0.1,\ 0.91,\ 0.181,\ 0.8371$, and $\frac1{1+\delta}\approx0.5263$. The errors are6221$-\frac{(-\delta)^T}{1+\delta}$: they alternate in sign and shrink by the factor $0.9$ each round.6222 6223\sol{16.4} By rejecting and continuing, player~2 obtains $\delta_2(1-\delta_1x^*)=1-x^*$. If $w$ is below6224this, opting out is never chosen and the equilibrium is unchanged. If $w$ is larger, she opts out after a6225rejection, so player~1 offers exactly $w$ and keeps $1-w$, agreement being immediate. (The condition for6226her to prefer opting out to counter-offering, $w\ge\delta_2(1-\delta_1(1-w))$, reduces to the same threshold.)6227The outside option acts as a constraint on the split and not as a disagreement point.6228 6229\sol{16.5} Against $s(c)=\tfrac23c+\tfrac14$ a bid $b$ trades with costs below $\bar c=\tfrac32(b-\tfrac14)$. The6230derivative of the buyer's payoff is $\tfrac32(v-b)-\tfrac12\bar c$, which vanishes at $b=\tfrac23v+\tfrac1{12}$.6231Symmetrically the seller's first-order condition against $b(v)$ gives $s=\tfrac23c+\tfrac14$. Trade occurs iff6232$v-c\ge\tfrac14$. The difference $z=v-c$ has density $1-z$ on $[0,1]$, so the gains are6233$\int_{1/4}^1z(1-z)\,dz=\tfrac9{64}$.6234 6235\sol{16.6} With efficient trade and each trader paid the other's report, the buyer pays $c$ and the seller6236receives $v$ whenever $v>c$; a buyer of value $0$ and a seller of cost $1$ never trade and get zero. The6237deficit is $v-c$ per trade, with expectation $\int_0^1z(1-z)\,dz=\tfrac16$.6238 6239\section*{Chapter 17}6240 6241\sol{17.1} A bidder of value $v$ bidding $b(z)$ earns $(v-b(z))z^{n-1}$; optimality at $z=v$ gives6242$(b\,v^{n-1})'=(n-1)v^{n-1}$, so $b=\frac{n-1}{n}v$. By revenue equivalence $m(v)=\int_0^vs\,d(s^{n-1})=\frac{n-1}{n}v^n$6243and $b=m/v^{n-1}$. Revenue is $n\int_0^1m=\frac{n-1}{n+1}$.6244 6245\sol{17.2} Revenue is $r$ when exactly one value exceeds $r$, and the lower value when both do:6246$2r^2(1-r)+\int_r^12s(1-s)\,ds=\tfrac13+r^2-\tfrac43r^3$, maximal at $r=\tfrac12$ with value $\tfrac5{12}$.6247 6248\sol{17.3} $m(v)=\int_0^vs\,d(s^2)=\tfrac23v^3$. In a third-price auction the winner pays the lowest of the6249other two bids, whose expected value given that both are below $v$ is $b(\tfrac v3)$ for linear $b=kv$. So6250$v^2\cdot\tfrac{kv}3=\tfrac23v^3$ and $k=2$: bidders bid twice their values.6251 6252\sol{17.4} If the opponent bids $2x_j$, winning yields $x_i+x_j-2x_j=x_i-x_j$, positive exactly when6253$x_i>x_j$; bidding $2x_i$ wins in exactly those cases, whatever $x_j$ is. So it is optimal even with6254knowledge of $x_j$. Revenue is $\E[2\min]=\tfrac23$. In the first-price auction with $b(x)=x$ revenue is6255$\E[\max]=\tfrac23$. They agree: with independent signals revenue equivalence holds even though values are6256interdependent.6257 6258\sol{17.5} For a product density both sides of the affiliation inequality are6259$\prod_ig_i(z_i)g_i(z_i')$. If $(x,y)$ are affiliated then for $x'>x$ the ratio $g(y\mid x')/g(y\mid x)$ is6260increasing in $y$, so the conditional distribution at $x'$ dominates that at $x$ in the first-order6261stochastic sense, and expectations of increasing functions are ordered.6262 6263\sol{17.6} In a Dutch auction nothing is observed before the auction ends, so a bidder's successive6264decisions ``not yet'' can be coalesced into one choice of a stopping price, and the bidders' choices6265interchanged: the reduced strategic form is that of the first-price auction, for any structure of values.6266In an English auction with three or more bidders, exits are observed, information sets distinguish6267histories with different exit prices, and a strategy is a function of them; the reduced strategic form is6268strictly larger than that of the sealed bid. With two bidders the only observable event ends the auction,6269and the two formats are again the same game.6270 6271\section*{Chapter 18}6272 6273\sol{18.1} In units of $(a-c)^2/b$: the collusive profit is $\frac1{4n}$; the best deviation, a best6274response to the others' $\frac{a-c}{2nb}$, earns $\frac{(n+1)^2}{16n^2}$; the Cournot profit is $\frac1{(n+1)^2}$.6275Grim trigger works iff $\delta\ge\frac{\pi^d-\pi^c}{\pi^d-\pi^N}=\frac{(n+1)^2}{n^2+6n+1}$. This is $\tfrac9{17}$6276at $n=2$, above the Bertrand threshold $\tfrac12$, and $\tfrac47$ at $n=3$, below $\tfrac23$; for all $n\ge3$ it6277is below $1-\tfrac1n$. Price competition makes deviation more tempting and punishment harsher, and for more6278than two firms the first effect dominates.