28. Impossible Efficient Mechanism for Bilateral Trading
28. Impossible Efficient Mechanism for Bilateral Trading
Bilateral Trade 组导读 / Bilateral Trade group overview 「双边交易」组(Ch 28–29)研究一个买家与一个卖家、双方均有私人信息时的交易机制:Ch 28 证明 Myerson–Satterthwaite (1983) 不可能性定理——不存在同时满足事后有效、预算平衡、激励相容、个人理性的交易机制;Ch 29 给出在放松事后有效(容许无效率)后的第二好机制设计。
本章导读 本章把单方私人信息的筛选/拍卖框架推广到双边私人信息:买家与卖家各自只知道自己的估值。§28.1 框架:买家估值 \(\theta_B\sim(F_B,f_B)\) 支撑 \([\underline\theta_B,\overline\theta_B]\)、卖家估值 \(\theta_S\sim(F_S,f_S)\) 支撑 \([\underline\theta_S,\overline\theta_S]\),二者支撑严格重叠;交易机制 \(\{\varphi(\hat\theta_B,\hat\theta_S),t(\hat\theta_B,\hat\theta_S)\}\)(\(\varphi\) 为交易概率、\(t\) 为买家付给卖家的转移);定义 28.1 事后预算平衡、定义 28.2 事后有效(\(\theta_B\ge\theta_S\) 才交易)、定义 28.3 激励相容 (IC)、定义 28.4 个人理性 (IR,外部选项 \(0\))。§28.2 定理 28.1 给出 IC 的充要条件(\(\bar\varphi_B\) 递增、\(\bar\varphi_S\) 递减,加上两条效用积分式),命题 28.1 用分部积分把期望效用写成虚拟剩余形式。§28.3 定理 28.2(IC 蕴含的"剩余 = 效用"恒等式 (28.15))与其逆命题定理 28.3(满足该恒等式的 \(\varphi\) 必可补上 \(t\) 使机制 IC 且 IR)。§28.4 定理 28.4:Myerson–Satterthwaite 不可能性定理——支撑严格重叠且密度严格为正时,没有任何交易机制能同时事后有效、预算平衡、IC、IR。本章无外部图。
28. Impossible Efficient Mechanism for Bilateral Trading
Bilateral Trade group overview The "Bilateral Trade" group (Ch 28–29) studies trading mechanisms between one buyer and one seller when both sides have private information: Ch 28 proves the Myerson–Satterthwaite (1983) impossibility theorem — no trading mechanism is simultaneously ex-post efficient, budget balanced, incentive compatible, and individually rational; Ch 29 develops the second-best mechanism design once ex-post efficiency is relaxed (inefficiency is allowed).
Overview This chapter extends the one-sided-private-information screening/auction framework to two-sided private information: the buyer and the seller each know only their own valuation. §28.1 the framework: buyer valuation \(\theta_B\sim(F_B,f_B)\) on support \([\underline\theta_B,\overline\theta_B]\), seller valuation \(\theta_S\sim(F_S,f_S)\) on support \([\underline\theta_S,\overline\theta_S]\), the two supports strictly overlapping; a trading mechanism \(\{\varphi(\hat\theta_B,\hat\theta_S),t(\hat\theta_B,\hat\theta_S)\}\) (\(\varphi\) the trade probability, \(t\) the transfer the buyer pays the seller); Definition 28.1 ex-post budget balance, Definition 28.2 ex-post efficiency (trade iff \(\theta_B\ge\theta_S\)), Definition 28.3 incentive compatibility (IC), Definition 28.4 individual rationality (IR, outside option \(0\)). §28.2 Theorem 28.1 gives the necessary-and-sufficient conditions for IC (\(\bar\varphi_B\) non-decreasing, \(\bar\varphi_S\) non-increasing, plus two utility-integral identities), and Proposition 28.1 uses integration by parts to write expected utilities in virtual-surplus form. §28.3 Theorem 28.2 (the "surplus = utility" identity (28.15) implied by IC) and its converse Theorem 28.3 (any \(\varphi\) satisfying that identity can be completed with a transfer \(t\) making the mechanism IC and IR). §28.4 Theorem 28.4: the Myerson–Satterthwaite impossibility theorem — when the supports strictly overlap and the densities are strictly positive, no trading mechanism can be simultaneously ex-post efficient, budget balanced, IC, and IR. No external figures.
28.1 框架 / Framework
考虑一个买家与一个卖家就单一物品交易。卖家拥有该物品、其估值(持有价值)为 \(\theta_S\);买家的估值为 \(\theta_B\)。双方均为风险中性。估值是各自的私人信息:
- 买家估值 \(\theta_B\) 服从 cdf \(F_B\)、pdf \(f_B\),支撑 \([\underline\theta_B,\overline\theta_B]\);
- 卖家估值 \(\theta_S\) 服从 cdf \(F_S\)、pdf \(f_S\),支撑 \([\underline\theta_S,\overline\theta_S]\);
- \(\theta_B\) 与 \(\theta_S\) 独立。
我们假设两个支撑严格重叠:\(\underline\theta_S<\overline\theta_B\) 且 \(\underline\theta_B<\overline\theta_S\)。否则要么总是该交易、要么从不该交易,问题平凡。
由显示原理,只需考虑直接交易机制 \(\{\varphi(\hat\theta_B,\hat\theta_S),\,t(\hat\theta_B,\hat\theta_S)\}\):双方各报告类型 \(\hat\theta_B,\hat\theta_S\),则
- \(\varphi(\hat\theta_B,\hat\theta_S)\in[0,1]\) 为物品从卖家转移给买家的概率(交易概率);
- \(t(\hat\theta_B,\hat\theta_S)\) 为买家付给卖家的转移(货币)。
28.1 Framework
Consider one buyer and one seller trading a single object. The seller owns the object and values it (reservation value) at \(\theta_S\); the buyer values it at \(\theta_B\). Both sides are risk-neutral. The valuations are private information:
- the buyer's valuation \(\theta_B\) has cdf \(F_B\), pdf \(f_B\), support \([\underline\theta_B,\overline\theta_B]\);
- the seller's valuation \(\theta_S\) has cdf \(F_S\), pdf \(f_S\), support \([\underline\theta_S,\overline\theta_S]\);
- \(\theta_B\) and \(\theta_S\) are independent.
We assume the two supports strictly overlap: \(\underline\theta_S<\overline\theta_B\) and \(\underline\theta_B<\overline\theta_S\). Otherwise it is always or never efficient to trade and the problem is trivial.
By the revelation principle, it suffices to consider direct trading mechanisms \(\{\varphi(\hat\theta_B,\hat\theta_S),\,t(\hat\theta_B,\hat\theta_S)\}\): both sides report types \(\hat\theta_B,\hat\theta_S\), and then
- \(\varphi(\hat\theta_B,\hat\theta_S)\in[0,1]\) is the probability the object moves from seller to buyer (the trade probability);
- \(t(\hat\theta_B,\hat\theta_S)\) is the transfer (money) the buyer pays the seller.
定义 28.1(事后预算平衡 / Ex-post budget balance) 交易机制 \(\{\varphi,t\}\) 满足事后预算平衡,若买家支付的钱恰等于卖家收到的钱 \(t(\hat\theta_B,\hat\theta_S)\),没有第三方注资或销毁货币。A trading mechanism \(\{\varphi,t\}\) satisfies ex-post budget balance if the money the buyer pays equals exactly the money \(t(\hat\theta_B,\hat\theta_S)\) the seller receives, with no outside party injecting or burning money.
我们用买家与卖家各自的期望交易概率与期望转移(对对方类型取期望):
We use each side's expected trade probability and expected transfer (taking expectation over the other side's type):
$$\bar\varphi_B(\theta_B)=\mathbb{E}_{\theta_S}[\varphi(\theta_B,\theta_S)],\qquad \bar\varphi_S(\theta_S)=\mathbb{E}_{\theta_B}[\varphi(\theta_B,\theta_S)]$$
$$\bar t_B(\theta_B)=\mathbb{E}_{\theta_S}[t(\theta_B,\theta_S)],\qquad \bar t_S(\theta_S)=\mathbb{E}_{\theta_B}[t(\theta_B,\theta_S)]$$
于是买家与卖家(讲真话时)的期望效用为
The buyer's and seller's expected utilities (under truth-telling) are
$$U_B(\theta_B)=\theta_B\,\bar\varphi_B(\theta_B)-\bar t_B(\theta_B),\qquad U_S(\theta_S)=\bar t_S(\theta_S)-\theta_S\,\bar\varphi_S(\theta_S)$$
定义 28.2(事后有效 / Ex-post efficiency) 交易机制是事后有效的,若当且仅当买家估值不低于卖家估值时交易:\(\varphi(\theta_B,\theta_S)=1\) 若 \(\theta_B\ge\theta_S\),否则 \(\varphi(\theta_B,\theta_S)=0\)。A trading mechanism is ex-post efficient if trade occurs exactly when the buyer values the object at least as much as the seller: \(\varphi(\theta_B,\theta_S)=1\) if \(\theta_B\ge\theta_S\), and \(\varphi(\theta_B,\theta_S)=0\) otherwise.
定义 28.3(激励相容 / Incentive compatibility, IC) 机制 \(\{\varphi,t\}\) 是激励相容的,若双方均如实报告其类型为最优。即对买家所有 \(\theta_B,\hat\theta_B\):\(U_B(\theta_B)\ge\theta_B\,\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)\);对卖家所有 \(\theta_S,\hat\theta_S\):\(U_S(\theta_S)\ge\bar t_S(\hat\theta_S)-\theta_S\,\bar\varphi_S(\hat\theta_S)\)。A mechanism \(\{\varphi,t\}\) is incentive compatible if truthful reporting is optimal for both sides. That is, for the buyer all \(\theta_B,\hat\theta_B\): \(U_B(\theta_B)\ge\theta_B\,\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)\); for the seller all \(\theta_S,\hat\theta_S\): \(U_S(\theta_S)\ge\bar t_S(\hat\theta_S)-\theta_S\,\bar\varphi_S(\hat\theta_S)\).
定义 28.4(个人理性 / Individual rationality, IR) 机制 \(\{\varphi,t\}\) 是个人理性的(事后/事中意义、外部选项为 \(0\)),若每个类型参与的期望效用非负:\(U_B(\theta_B)\ge0\) 对所有 \(\theta_B\),\(U_S(\theta_S)\ge0\) 对所有 \(\theta_S\)。卖家不参与则保留物品得 \(0\),买家不参与得 \(0\)。A mechanism \(\{\varphi,t\}\) is individually rational (interim sense, outside option \(0\)) if every type's expected utility from participating is non-negative: \(U_B(\theta_B)\ge0\) for all \(\theta_B\) and \(U_S(\theta_S)\ge0\) for all \(\theta_S\). A non-participating seller keeps the object and gets \(0\); a non-participating buyer gets \(0\).
注 28.1(有效 = 最大社会剩余 / Remark 28.1: efficiency = maximal social surplus) 事后有效等价于最大化社会剩余 \(\mathbb{E}[\varphi(\theta_B,\theta_S)(\theta_B-\theta_S)]\):当 \(\theta_B\ge\theta_S\) 交易产生正剩余 \(\theta_B-\theta_S\),故应交易;当 \(\theta_B<\theta_S\) 交易产生负剩余,故不应交易。转移 \(t\) 只在双方间再分配剩余,不影响总剩余。Ex-post efficiency is equivalent to maximizing social surplus \(\mathbb{E}[\varphi(\theta_B,\theta_S)(\theta_B-\theta_S)]\): when \(\theta_B\ge\theta_S\) trade creates positive surplus \(\theta_B-\theta_S\) so trade should occur; when \(\theta_B<\theta_S\) trade creates negative surplus so it should not. The transfer \(t\) only redistributes surplus between the two parties without affecting total surplus.
28.2 激励相容的刻画 / Characterizing incentive compatibility
与单方筛选/拍卖一样,IC 可由"单调性 + 效用积分式"完全刻画,只是现在买卖两边都有约束。
28.2 Characterizing incentive compatibility
As in one-sided screening/auctions, IC is fully characterized by "monotonicity + a utility-integral identity", except now there are constraints on both the buyer and the seller sides.
定理 28.1(IC 的充要条件 / N&S conditions for IC) 交易机制 \(\{\varphi,t\}\) 激励相容,当且仅当:(1) 单调性——\(\bar\varphi_B(\cdot)\) 关于 \(\theta_B\) 不减、\(\bar\varphi_S(\cdot)\) 关于 \(\theta_S\) 不增;(2) 效用积分式——\(U_B(\theta_B)=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)\,dx\) 且 \(U_S(\theta_S)=U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)\,dx\)。A trading mechanism \(\{\varphi,t\}\) is incentive compatible if and only if: (1) monotonicity — \(\bar\varphi_B(\cdot)\) is non-decreasing in \(\theta_B\) and \(\bar\varphi_S(\cdot)\) is non-increasing in \(\theta_S\); (2) utility-integral identities — \(U_B(\theta_B)=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)\,dx\) and \(U_S(\theta_S)=U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)\,dx\).
注 28.2(等价的转移条件 / Remark 28.2: equivalent transfer conditions) 把 \(U_B,U_S\) 的定义代入 (2),可得对 \(\bar t_B,\bar t_S\) 的等价条件(即正文 (28.1)、(28.2)):买家的期望转移 \(\bar t_B(\theta_B)=\theta_B\bar\varphi_B(\theta_B)-U_B(\underline\theta_B)-\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\),卖家的期望转移 \(\bar t_S(\theta_S)=\theta_S\bar\varphi_S(\theta_S)+U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)dx\)。直觉与单方机制相同:信息租金由边际类型的"分配规则斜率"积分决定。Substituting the definitions of \(U_B,U_S\) into (2) gives the equivalent conditions on \(\bar t_B,\bar t_S\) (the text's (28.1), (28.2)): the buyer's expected transfer \(\bar t_B(\theta_B)=\theta_B\bar\varphi_B(\theta_B)-U_B(\underline\theta_B)-\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\), and the seller's expected transfer \(\bar t_S(\theta_S)=\theta_S\bar\varphi_S(\theta_S)+U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)dx\). The intuition is the same as in one-sided mechanisms: information rents are determined by integrating the slope of the allocation rule over marginal types.
证明 / Proof 必要性。 设机制 IC。固定买家、考虑两类型 \(\theta_B,\theta_B'\)。讲真话不低于谎报为 \(\theta_B'\):Necessity. Suppose the mechanism is IC. Fix the buyer and consider two types \(\theta_B,\theta_B'\). Truth-telling weakly dominates misreporting \(\theta_B'\):
$$U_B(\theta_B)\ge\theta_B\bar\varphi_B(\theta_B')-\bar t_B(\theta_B')=U_B(\theta_B')+(\theta_B-\theta_B')\bar\varphi_B(\theta_B')$$
对称地交换角色:\(U_B(\theta_B')\ge U_B(\theta_B)+(\theta_B'-\theta_B)\bar\varphi_B(\theta_B)\)。两式相加整理得Swapping roles symmetrically: \(U_B(\theta_B')\ge U_B(\theta_B)+(\theta_B'-\theta_B)\bar\varphi_B(\theta_B)\). Adding and rearranging the two inequalities gives
$$(\theta_B-\theta_B')\big(\bar\varphi_B(\theta_B)-\bar\varphi_B(\theta_B')\big)\ge0$$
即 \(\bar\varphi_B\) 不减。由 \(U_B(\theta_B)=\max_{\hat\theta_B}\{\theta_B\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)\}\) 是仿射函数族的上包络,\(U_B\) 凸且几乎处处可导,由包络定理 \(U_B'(\theta_B)=\bar\varphi_B(\theta_B)\),积分得 \(U_B(\theta_B)=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\)。i.e. \(\bar\varphi_B\) is non-decreasing. Since \(U_B(\theta_B)=\max_{\hat\theta_B}\{\theta_B\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)\}\) is the upper envelope of a family of affine functions, \(U_B\) is convex and differentiable a.e.; by the envelope theorem \(U_B'(\theta_B)=\bar\varphi_B(\theta_B)\), and integrating gives \(U_B(\theta_B)=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\).
卖家对称:因卖家估值越高越不愿交易,谎报不优条件给出 \(\bar\varphi_S\) 不增、\(U_S'(\theta_S)=-\bar\varphi_S(\theta_S)\),从上端 \(\overline\theta_S\) 向下积分得 \(U_S(\theta_S)=U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)dx\)。The seller is symmetric: since a higher-valued seller is less willing to trade, the no-misreporting conditions give \(\bar\varphi_S\) non-increasing and \(U_S'(\theta_S)=-\bar\varphi_S(\theta_S)\); integrating downward from the top \(\overline\theta_S\) gives \(U_S(\theta_S)=U_S(\overline\theta_S)+\int_{\theta_S}^{\overline\theta_S}\bar\varphi_S(x)dx\).
充分性。 设 (1)(2) 成立。对买家,需验证对任意 \(\theta_B,\hat\theta_B\) 有 \(U_B(\theta_B)\ge\theta_B\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)=U_B(\hat\theta_B)+(\theta_B-\hat\theta_B)\bar\varphi_B(\hat\theta_B)\)。由 (2),\(U_B(\theta_B)-U_B(\hat\theta_B)=\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(x)dx\)。于是只需Sufficiency. Suppose (1)(2) hold. For the buyer we must verify that for all \(\theta_B,\hat\theta_B\), \(U_B(\theta_B)\ge\theta_B\bar\varphi_B(\hat\theta_B)-\bar t_B(\hat\theta_B)=U_B(\hat\theta_B)+(\theta_B-\hat\theta_B)\bar\varphi_B(\hat\theta_B)\). By (2), \(U_B(\theta_B)-U_B(\hat\theta_B)=\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(x)dx\). So it suffices that
$$\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(x)\,dx\ge(\theta_B-\hat\theta_B)\bar\varphi_B(\hat\theta_B)$$
当 \(\theta_B>\hat\theta_B\),由 \(\bar\varphi_B\) 不减,\(\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(x)dx\ge\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(\hat\theta_B)dx=(\theta_B-\hat\theta_B)\bar\varphi_B(\hat\theta_B)\);当 \(\theta_B<\hat\theta_B\) 同理(两侧符号同时翻转)成立。故买家 IC 成立。卖家由 \(\bar\varphi_S\) 不增对称验证。\(\blacksquare\)When \(\theta_B>\hat\theta_B\), since \(\bar\varphi_B\) is non-decreasing, \(\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(x)dx\ge\int_{\hat\theta_B}^{\theta_B}\bar\varphi_B(\hat\theta_B)dx=(\theta_B-\hat\theta_B)\bar\varphi_B(\hat\theta_B)\); when \(\theta_B<\hat\theta_B\) the same holds (both sides flip sign together). So buyer IC holds. The seller follows symmetrically from \(\bar\varphi_S\) non-increasing. \(\blacksquare\)
下一步把期望效用对类型积分,用分部积分写成只含 \(\varphi\) 的"虚拟剩余"形式。
Next we integrate the expected utilities over types and use integration by parts to write them in a "virtual surplus" form involving only \(\varphi\).
命题 28.1(期望效用的虚拟剩余表示 / Virtual-surplus form of expected utilities) 在 IC 下,把定理 28.1 的效用积分式对 \(\theta_B\)(resp. \(\theta_S\))取期望并分部积分,得Under IC, taking the expectation of Theorem 28.1's utility-integral identities over \(\theta_B\) (resp. \(\theta_S\)) and integrating by parts gives
$$\mathbb{E}[U_B(\theta_B)]=U_B(\underline\theta_B)+\mathbb{E}\!\left[\varphi(\theta_B,\theta_S)\,\frac{1-F_B(\theta_B)}{f_B(\theta_B)}\right]$$
$$\mathbb{E}[U_S(\theta_S)]=U_S(\overline\theta_S)+\mathbb{E}\!\left[\varphi(\theta_B,\theta_S)\,\frac{F_S(\theta_S)}{f_S(\theta_S)}\right]$$
证明 / Proof 对买家:\(\mathbb{E}[U_B]=\int_{\underline\theta_B}^{\overline\theta_B}U_B(\theta_B)f_B(\theta_B)d\theta_B=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\overline\theta_B}\Big(\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\Big)f_B(\theta_B)d\theta_B\)。交换积分次序(或分部积分,\(\int u\,dF\) 取 \(u=\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B\)):For the buyer: \(\mathbb{E}[U_B]=\int_{\underline\theta_B}^{\overline\theta_B}U_B(\theta_B)f_B(\theta_B)d\theta_B=U_B(\underline\theta_B)+\int_{\underline\theta_B}^{\overline\theta_B}\Big(\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\Big)f_B(\theta_B)d\theta_B\). Swapping the order of integration (or integrating by parts, \(\int u\,dF\) with \(u=\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B\)):
$$\int_{\underline\theta_B}^{\overline\theta_B}\!\Big(\int_{\underline\theta_B}^{\theta_B}\bar\varphi_B(x)dx\Big)f_B(\theta_B)d\theta_B=\int_{\underline\theta_B}^{\overline\theta_B}\bar\varphi_B(x)\big(1-F_B(x)\big)dx=\mathbb{E}\!\left[\bar\varphi_B(\theta_B)\frac{1-F_B(\theta_B)}{f_B(\theta_B)}\right]$$
再由 \(\bar\varphi_B(\theta_B)=\mathbb{E}_{\theta_S}[\varphi(\theta_B,\theta_S)]\) 把内层期望放回,即得第一式。卖家对称:从 \(\overline\theta_S\) 向下积分、权重变为 \(F_S(\theta_S)\),得第二式。\(\blacksquare\)Then putting the inner expectation back via \(\bar\varphi_B(\theta_B)=\mathbb{E}_{\theta_S}[\varphi(\theta_B,\theta_S)]\) yields the first identity. The seller is symmetric: integrating downward from \(\overline\theta_S\), the weight becomes \(F_S(\theta_S)\), giving the second identity. \(\blacksquare\)
28.3 剩余 = 效用恒等式 / The "surplus = utility" identity
将命题 28.1 的两式相加,并利用预算平衡(双方转移相消)便得到一个关键恒等式:在 IC 机制下,总期望社会剩余必须恰好等于双方最低类型的期望效用之和。
28.3 The "surplus = utility" identity
Adding the two identities of Proposition 28.1 and using budget balance (the two transfers cancel) yields a key identity: under any IC mechanism, the total expected social surplus must equal exactly the sum of the boundary types' expected utilities.
定理 28.2(IC + 预算平衡蕴含的恒等式 / Identity implied by IC + budget balance) 若交易机制 \(\{\varphi,t\}\) 激励相容且事后预算平衡,则If a trading mechanism \(\{\varphi,t\}\) is incentive compatible and ex-post budget balanced, then
$$\mathbb{E}\!\left[\varphi(\theta_B,\theta_S)\Big(\big(\theta_B-\tfrac{1-F_B(\theta_B)}{f_B(\theta_B)}\big)-\big(\theta_S+\tfrac{F_S(\theta_S)}{f_S(\theta_S)}\big)\Big)\right]=U_B(\underline\theta_B)+U_S(\overline\theta_S)\tag{28.15}$$
证明 / Proof 预算平衡下买家付的等于卖家收的,故总期望剩余等于双方期望效用之和:\(\mathbb{E}[\varphi(\theta_B-\theta_S)]=\mathbb{E}[U_B(\theta_B)]+\mathbb{E}[U_S(\theta_S)]\)(左侧 \(=\mathbb{E}[\theta_B\bar\varphi_B-\bar t_B]+\mathbb{E}[\bar t_S-\theta_S\bar\varphi_S]\),而 \(\mathbb{E}[\bar t_B]=\mathbb{E}[t]=\mathbb{E}[\bar t_S]\) 相消)。代入命题 28.1 的两式:Under budget balance what the buyer pays equals what the seller receives, so the total expected surplus equals the sum of expected utilities: \(\mathbb{E}[\varphi(\theta_B-\theta_S)]=\mathbb{E}[U_B(\theta_B)]+\mathbb{E}[U_S(\theta_S)]\) (the LHS \(=\mathbb{E}[\theta_B\bar\varphi_B-\bar t_B]+\mathbb{E}[\bar t_S-\theta_S\bar\varphi_S]\), and \(\mathbb{E}[\bar t_B]=\mathbb{E}[t]=\mathbb{E}[\bar t_S]\) cancel). Substituting the two identities of Proposition 28.1:
$$\mathbb{E}[\varphi(\theta_B-\theta_S)]=U_B(\underline\theta_B)+\mathbb{E}\!\left[\varphi\tfrac{1-F_B}{f_B}\right]+U_S(\overline\theta_S)+\mathbb{E}\!\left[\varphi\tfrac{F_S}{f_S}\right]$$
将两个期望项移到左边、合并到 \(\varphi\) 内即得 (28.15)。\(\blacksquare\)Moving the two expectation terms to the left and combining them inside \(\varphi\) yields (28.15). \(\blacksquare\)
解读 / Interpretation 左侧 \(\theta_B-\frac{1-F_B}{f_B}\) 是买家的虚拟估值(边际收入),\(\theta_S+\frac{F_S}{f_S}\) 是卖家的虚拟成本(边际成本)。恒等式说:IC + 预算平衡机制下,期望"虚拟剩余"被双方边界类型的租金完全占用。要 IR(即 \(U_B(\underline\theta_B),U_S(\overline\theta_S)\ge0\)),就要求左侧虚拟剩余 \(\ge0\)。On the left, \(\theta_B-\frac{1-F_B}{f_B}\) is the buyer's virtual valuation (marginal revenue) and \(\theta_S+\frac{F_S}{f_S}\) is the seller's virtual cost (marginal cost). The identity says: under an IC + budget-balanced mechanism, expected "virtual surplus" is entirely absorbed by the boundary types' rents. IR (i.e. \(U_B(\underline\theta_B),U_S(\overline\theta_S)\ge0\)) then requires the left-hand virtual surplus to be \(\ge0\).
定理 28.2 的逆命题表明:(28.15)(弱化为 IR 所需的不等式)几乎就是 IC + 预算平衡 + IR 可实现的全部约束——只要分配规则单调且满足该不等式,总能补上合适的转移 \(t\)。
The converse of Theorem 28.2 shows that (28.15) (weakened to the inequality required by IR) is essentially the only constraint for IC + budget balance + IR feasibility — as long as the allocation rule is monotone and satisfies that inequality, one can always supply an appropriate transfer \(t\).
定理 28.3(逆命题:可实现性 / Converse: implementability) 设分配规则 \(\varphi(\theta_B,\theta_S)\) 满足 \(\bar\varphi_B(\cdot)\) 不减、\(\bar\varphi_S(\cdot)\) 不增,且Suppose an allocation rule \(\varphi(\theta_B,\theta_S)\) has \(\bar\varphi_B(\cdot)\) non-decreasing, \(\bar\varphi_S(\cdot)\) non-increasing, and satisfies
$$\mathbb{E}\!\left[\varphi(\theta_B,\theta_S)\Big(\big(\theta_B-\tfrac{1-F_B(\theta_B)}{f_B(\theta_B)}\big)-\big(\theta_S+\tfrac{F_S(\theta_S)}{f_S(\theta_S)}\big)\Big)\right]\ge0\tag{28.16}$$
则存在转移规则 \(t(\theta_B,\theta_S)\),使机制 \(\{\varphi,t\}\) 同时激励相容、事后预算平衡且个人理性。then there exists a transfer rule \(t(\theta_B,\theta_S)\) such that the mechanism \(\{\varphi,t\}\) is simultaneously incentive compatible, ex-post budget balanced, and individually rational.
证明(构造转移 / constructing the transfer) 取 \(U_B(\underline\theta_B)=U_S(\overline\theta_S)=\frac12\,\mathbb{E}[\varphi((\theta_B-\frac{1-F_B}{f_B})-(\theta_S+\frac{F_S}{f_S}))]\ge0\)(由 (28.16) 非负,故 IR 在边界类型处成立;单调性 + 积分式保证所有类型 \(U_B,U_S\ge0\))。给定单调 \(\varphi\) 与这两个边界效用,由注 28.2 可定出期望转移 \(\bar t_B(\theta_B),\bar t_S(\theta_S)\) 使 IC 成立。最后只需找一个事后转移 \(t(\theta_B,\theta_S)\) 同时满足两个边际条件 \(\mathbb{E}_{\theta_S}[t]=\bar t_B(\theta_B)\)、\(\mathbb{E}_{\theta_B}[t]=\bar t_S(\theta_S)\)。一种可行构造为Set \(U_B(\underline\theta_B)=U_S(\overline\theta_S)=\frac12\,\mathbb{E}[\varphi((\theta_B-\frac{1-F_B}{f_B})-(\theta_S+\frac{F_S}{f_S}))]\ge0\) (non-negative by (28.16), so IR holds at the boundary types; monotonicity + the integral identities then ensure \(U_B,U_S\ge0\) for all types). Given the monotone \(\varphi\) and these two boundary utilities, Remark 28.2 pins down the expected transfers \(\bar t_B(\theta_B),\bar t_S(\theta_S)\) that make IC hold. It remains to find an ex-post transfer \(t(\theta_B,\theta_S)\) satisfying the two marginal conditions \(\mathbb{E}_{\theta_S}[t]=\bar t_B(\theta_B)\) and \(\mathbb{E}_{\theta_B}[t]=\bar t_S(\theta_S)\). One feasible construction is
$$t(\theta_B,\theta_S)=\bar t_B(\theta_B)+\bar t_S(\theta_S)-\mathbb{E}[t],\qquad \mathbb{E}[t]=\mathbb{E}[\bar t_B(\theta_B)]=\mathbb{E}[\bar t_S(\theta_S)]$$
直接验证 \(\mathbb{E}_{\theta_S}[t]=\bar t_B(\theta_B)\)、\(\mathbb{E}_{\theta_B}[t]=\bar t_S(\theta_S)\) 成立(用 \(\mathbb{E}[\bar t_B]=\mathbb{E}[\bar t_S]=\mathbb{E}[t]\)),且 \(t\) 事后即为买家所付、卖家所收(预算平衡)。故 \(\{\varphi,t\}\) 满足全部要求。\(\blacksquare\)One directly verifies \(\mathbb{E}_{\theta_S}[t]=\bar t_B(\theta_B)\) and \(\mathbb{E}_{\theta_B}[t]=\bar t_S(\theta_S)\) (using \(\mathbb{E}[\bar t_B]=\mathbb{E}[\bar t_S]=\mathbb{E}[t]\)), and \(t\) is ex post both what the buyer pays and what the seller receives (budget balance). Hence \(\{\varphi,t\}\) meets all requirements. \(\blacksquare\)
28.4 Myerson–Satterthwaite 不可能性定理 / The impossibility theorem
现在把事后有效的分配规则代入恒等式 (28.15)。若机制还要预算平衡且 IR,则需 \(U_B(\underline\theta_B)+U_S(\overline\theta_S)\ge0\);但代入后左侧严格为负,矛盾。
28.4 The Myerson–Satterthwaite impossibility theorem
Now plug the ex-post efficient allocation rule into the identity (28.15). If the mechanism is also budget balanced and IR, we need \(U_B(\underline\theta_B)+U_S(\overline\theta_S)\ge0\); but after substitution the left-hand side is strictly negative — a contradiction.
定理 28.4(Myerson–Satterthwaite 1983 不可能性定理 / Impossibility theorem) 设买家与卖家的支撑有非空交集(\(\overline\theta_S\ge\underline\theta_B\) 且 \(\underline\theta_S<\overline\theta_B\))、且密度在各自支撑上严格为正。则不存在任何交易机制能同时满足:事后有效、事后预算平衡、激励相容、个人理性。Suppose the buyer's and seller's supports have a non-empty intersection (\(\overline\theta_S\ge\underline\theta_B\) and \(\underline\theta_S<\overline\theta_B\)) and the densities are strictly positive on their supports. Then no trading mechanism can be simultaneously ex-post efficient, ex-post budget balanced, incentive compatible, and individually rational.
证明 / Proof 设存在这样的机制。事后有效即 \(\varphi(\theta_B,\theta_S)=\mathbf 1\{\theta_B\ge\theta_S\}\)。代入 (28.15) 的左侧,先对 \(\theta_S\) 在 \([\underline\theta_S,\theta_B]\) 上积分(仅此区间交易)。买家虚拟项部分:Suppose such a mechanism exists. Ex-post efficiency means \(\varphi(\theta_B,\theta_S)=\mathbf 1\{\theta_B\ge\theta_S\}\). Substitute into the LHS of (28.15) and integrate over \(\theta_S\in[\underline\theta_S,\theta_B]\) (trade occurs only there). For the buyer's virtual part:
$$\mathbb{E}\!\left[\mathbf 1\{\theta_B\ge\theta_S\}\Big(\theta_B-\tfrac{1-F_B(\theta_B)}{f_B(\theta_B)}\Big)\right]=\int\Big(\theta_B-\tfrac{1-F_B(\theta_B)}{f_B(\theta_B)}\Big)F_S(\theta_B)f_B(\theta_B)\,d\theta_B$$
卖家虚拟项部分对称。把两项合并、对交易区域 \(\{\theta_B\ge\theta_S\}\) 重排并分部积分,经标准化简(Myerson–Satterthwaite 1983),(28.15) 的左侧化为The seller's virtual part is symmetric. Combining the two terms, rearranging over the trading region \(\{\theta_B\ge\theta_S\}\) and integrating by parts, standard simplification (Myerson–Satterthwaite 1983) reduces the LHS of (28.15) to
$$\mathbb{E}\!\left[\mathbf 1\{\theta_B\ge\theta_S\}\Big((\theta_B-\tfrac{1-F_B}{f_B})-(\theta_S+\tfrac{F_S}{f_S})\Big)\right]=-\int_{\underline\theta_B}^{\overline\theta_S}\big(1-F_B(\theta)\big)F_S(\theta)\,d\theta$$
在交集内部 \((\underline\theta_B,\overline\theta_S)\),由密度严格为正有 \(0
直觉 / Intuition 要事后有效就必须在所有 \(\theta_B\ge\theta_S\) 时交易,但双方都有动机谎报以攫取信息租金:买家想压低报价、卖家想抬高报价。补偿这两类信息租金所需的钱,超过了机制内部可再分配的剩余——除非有外部补贴打破预算平衡,否则有效交易、自愿参与、预算自足三者不可兼得。这是双边私人信息下交易必然存在无效率(错失部分本应发生的交易)的根本原因。Ex-post efficiency requires trade whenever \(\theta_B\ge\theta_S\), but both sides have an incentive to misreport to capture information rents: the buyer wants to understate, the seller to overstate. The money needed to compensate both kinds of information rent exceeds the surplus that can be redistributed within the mechanism — so unless an outside subsidy breaks budget balance, efficient trade, voluntary participation, and budget self-sufficiency cannot all hold at once. This is the fundamental reason why trade under two-sided private information is necessarily inefficient (some socially desirable trades are missed).
注 28.3(事后 IR vs 事前合约 / Remark 28.3: ex-post IR vs ex-ante contracting) 不可能性定理是相对事后 IR约束而言的——即双方在各自已私下得知类型后仍自愿参与。若考虑事前交易机制(例如双方在了解各自类型之前就签订交易合约),则可以实现第一好(社会福利/有效率意义上的最优),此时事前 IR约束不会被事后有效的交易机制破坏。The impossibility theorem is stated relative to the ex-post IR constraint — i.e. both agents still participate voluntarily after each has privately learned their type. If instead we consider an ex-ante trading mechanism (e.g. the two sides sign a trading contract before learning their types), then the first best (the optimum in terms of social welfare or efficiency) can be realized, and in that case the ex-ante IR constraint won't be violated by an ex-post efficient trading mechanism.
参考文献 / References
- Myerson, R., & Satterthwaite, M. (1983). Efficient Mechanisms for Bilateral Trading. Journal of Economic Theory, 29(2), 265-281.(双边交易不可能性定理)
- 本章把 [[monopolistic-screening]](Ch 20)、[[optimal-auction]](Ch 27)的单方私人信息机制推广到双边私人信息;第二好机制设计见下一章 [[second-best-bilateral-trade]](Ch 29)。
References
- Myerson, R., & Satterthwaite, M. (1983). Efficient Mechanisms for Bilateral Trading. Journal of Economic Theory, 29(2), 265-281. (the bilateral-trade impossibility theorem)
- This chapter extends the one-sided-private-information mechanisms of [[monopolistic-screening]] (Ch 20) and [[optimal-auction]] (Ch 27) to two-sided private information; the second-best mechanism design is in the next chapter [[second-best-bilateral-trade]] (Ch 29).