第6章通过效用最大化和拉格朗日乘数法介绍了消费者理论。本章抛弃特定函数形式的拐杖，从公理化基础构建理论。我们要问：什么时候偏好可以用效用函数表示？需求函数必须满足什么性质？在什么条件下竞争性市场体系能有效配置资源？
方法上的转变是从计算到证明。第二部分求解最优化问题。第三部分证明定理——确定哪些结果是稳健的，哪些依赖于特殊假设。
先修知识：第6–7章。数学先修：实分析基础（开集/闭集、连续性、不动点定理）、凸分析、矩阵代数。见附录A。
参考文献：Mas-Colell, Whinston & Green（MWG）；Debreu《价值理论》；Arrow & Debreu（1954）；Varian《微观经济分析》。
标准公理：
证明概要。固定一条射线 $\{te : t \geq 0\}$，其中 $e = (1,1,\ldots,1)$。对每个 $x$，由完备性和连续性，存在唯一的 $t(x) \geq 0$ 使得 $x \sim t(x)e$。令 $u(x) = t(x)$。传递性确保表示的一致性；连续性确保 $u$ 是连续的。
效用函数是序数的——任何单调变换 $v = g(u)$（$g' > 0$）都表示相同的偏好。基数性质（效用差异的大小）是无意义的。
考虑 $\mathbb{R}^2_+$ 上的字典序偏好：$x \succ y$ 如果 $x_1 > y_1$，或 $x_1 = y_1$ 且 $x_2 > y_2$。
完备性：满足——对任意 $x, y$，要么 $x_1 > y_1$，要么 $y_1 > x_1$，或 $x_1 = y_1$ 然后比较 $x_2, y_2$。
传递性：满足——如果 $x \succ y$ 且 $y \succ z$，则 $x \succ z$（由 $\mathbb{R}$ 上 $>$ 的传递性得出）。
连续性：不满足。考虑 $y = (1, 1)$。集合 $\{x : x \succ y\}$ 包含 $(1, 1.5)$ 但不包含 $(0.999, 100)$。"至少一样好"的集合不是闭的——在 $x_1 = 1$ 处有跳跃。
结论：不存在连续效用函数能表示字典序偏好。这说明连续性对德布鲁效用表示定理至关重要。
我们可以从观察到的选择推断偏好，而非假设偏好。
形式地：如果 $x$ 被显示偏好于 $y$（$xRy$：在 $y$ 可负担的价格下选择了 $x$），则 $y$ 不被显示偏好于 $x$。
SARP是观察到的选择与效用最大化一致的充要条件（阿弗里亚特定理）。WARP是必要的但一般不充分（尽管在两种商品时是充分的）。
一个消费者在两种价格-收入情况下的选择：
| 情况 | 价格 $(p_1, p_2)$ | 选择的组合 $(x_1, x_2)$ | 支出 |
|---|---|---|---|
| A | (1, 2) | (4, 2) | 8 |
| B | (2, 1) | (2, 4) | 8 |
检验WARP：在价格A下，消费者能否负担得起组合B？\$1(2) + 2(4) = 10 > 8$。不能。在价格B下，消费者能否负担得起组合A？\$1(4) + 1(2) = 10 > 8$。不能。WARP满足——数据与效用最大化一致。
输入最多6组观测的价格向量和选择的商品束。检查器将自动测试WARP和SARP。
| 观测 | $p_1$ | $p_2$ | $x_1$ | $x_2$ | 支出 |
|---|---|---|---|---|---|
| 1 | 8.0 | ||||
| 2 | 8.0 | ||||
| 3 | 6.0 | ||||
| 4 | — | ||||
| 5 | — | ||||
| 6 | — |
互动 11.1。输入价格-组合观测值并检验显示性偏好的一致性。WARP检查直接的成对反转；SARP检查任意长度的循环。违反情况会被高亮并附解释。
You now have the formal content of "rationality": completeness, transitivity, continuity — the axioms required for utility functions to exist — and WARP/SARP, which make rationality empirically testable from observed choices.
Rational choice is now precise: complete + transitive + continuous preferences guarantee a utility function exists (Debreu's representation theorem). WARP and SARP give empirical tests — if you chose bundle $A$ when $B$ was affordable, you should never choose $B$ when $A$ is affordable at those prices. Violations of SARP are violations of rationality, full stop. The entire apparatus of welfare economics — the welfare theorems you'll prove in §11.6–11.7, the duality framework in §11.3–11.4, mechanism design in Chapter 12 — requires these axioms. Without them, utility functions don't exist, consumer surplus is undefined, and "efficiency" loses its formal meaning.
Completeness is implausible for complex choices — people genuinely don't have well-defined preferences over all possible bundles (Sen 1997). Transitivity fails systematically: preference reversals between gambles (Grether & Plott 1979) are robust and replicable across decades of experiments. Context dependence — decoy effects, framing effects, anchoring — violates the independence of irrelevant alternatives that WARP requires. The Allais paradox shows that expected utility's independence axiom fails even among trained decision theorists. These aren't occasional lapses by confused subjects; they're systematic patterns that survive incentives, experience, and high stakes. If the axioms fail, the utility function doesn't exist, and welfare analysis — which depends on maximizing a well-defined objective — loses its foundations entirely.
The mainstream response is twofold. First, at the individual level, violations are real but the stakes in most laboratory experiments are trivial — people may satisfice over small gambles but optimize over consequential decisions (mortgages, career choices, firm strategy). Second, at the market level, competition and selection may eliminate irrational agents: Alchian (1950) and Friedman (1953) argued that firms behaving as if they maximize profits survive, regardless of their actual decision process. The "as if" defense says that even if individuals aren't literally maximizing utility, markets behave as if they were — because competitive pressure weeds out consistently irrational behavior. This defense is powerful but depends on the speed and completeness of market discipline.
The axioms are best understood as a benchmark, not a description of how people actually decide. They tell you what consistency requires, and deviations from them are informative — they point to specific psychological mechanisms (loss aversion, probability weighting, framing effects, present bias) that can themselves be modeled formally. The revealed preference framework is valuable precisely because it's testable: SARP doesn't ask whether people feel rational, it checks whether their choices are consistent. The question is whether the violations that laboratory experiments document survive the aggregation and competition of real markets.
The "as if" defense works only if markets discipline irrational behavior quickly. But does arbitrage actually eliminate biases, or can noise traders survive and move prices? Come back in Chapter 19 (§19.1–19.2, §19.8), where prospect theory provides a formal alternative to expected utility, and behavioral finance tests whether biases survive in the one market — financial markets — where you'd most expect them to be eliminated. The DSSW noise trader model and limits-to-arbitrage literature give the surprising answer.
Saint-Paul argues the internal logic of behavioral economics points toward hard paternalism, not the gentle kind. If people violate the axioms systematically, who decides what "better" means?
中级The viral slogan meets the First Welfare Theorem. Some fortunes are market failures; others are surplus creation. The word "every" is where the claim breaks.
高级第6章解决了原始问题：在预算约束下最大化效用。对偶问题是最小化支出以达到目标效用水平。
解是希克斯（补偿）需求 $h(p, \bar{u})$：
间接效用函数 $V(p, m)$ 给出在价格 $p$ 和收入 $m$ 下可达到的最大效用：
$$V(p, m) = \max_{x} \; u(x) \quad \text{s.t.} \quad p \cdot x \leq m$$关键的对偶关系：
罗伊恒等式提供了从间接效用函数推导马歇尔需求的捷径：
罗伊恒等式的直觉：$p_i$ 的微小增加对福利（以 $V$ 衡量）有两个效应：（1）使商品 $i$ 更贵从而直接降低效用（分子 $\partial V/\partial p_i < 0$），（2）这一效应的大小与消费者购买商品 $i$ 的数量（$x_i$）乘以收入的边际效用（$\partial V/\partial m$）成正比。将（1）除以收入的边际效用得到商品 $i$ 的数量。
CES效用：$u(x_1, x_2) = (x_1^\rho + x_2^\rho)^{1/\rho}$，$\rho < 1$，$\rho \neq 0$。
支出函数为：$e(p, \bar{u}) = \bar{u} \cdot (p_1^r + p_2^r)^{1/r}$，其中 $r = \rho/(\rho - 1)$。
希克斯需求（谢帕德引理）：$h_i = \bar{u} \cdot p_i^{r-1} / (p_1^r + p_2^r)^{(r-1)/r}$。
当 $\rho \to 0$（替代弹性 $\sigma = 1/(1-\rho) \to 1$）时，收敛到柯布-道格拉斯的情形。
柯布-道格拉斯效用 $u = x_1^{0.5} x_2^{0.5}$，收入 $m = 10$。滑动 $p_1$ 观察预算线切点、马歇尔需求和支出函数三种表述如何编码相同的信息。
互动 11.2。同一消费者的三个视角。左：无差异曲线与预算线相切（原始问题）。中：商品1的马歇尔需求关于 $p_1$ 的函数。右：达到当前效用水平所需的支出函数 $e(p_1, p_2, \bar{u})$。三者编码相同的偏好。
第6章的斯勒茨基方程（公式6.7）推广为矩阵。定义斯勒茨基（替代）矩阵，其元素为：
如果需求由效用最大化产生，斯勒茨基矩阵必须是：
这些是可检验的限制——如果观察到的需求违反它们，则不可能由理性消费者最大化良好效用函数产生。
柯布-道格拉斯需求：$x_1 = am/p_1$，$x_2 = (1-a)m/p_2$。
$S_{12} = \partial x_1/\partial p_2 + x_2 \cdot \partial x_1/\partial m = 0 + [(1-a)m/p_2] \cdot [a/p_1] = a(1-a)m/(p_1 p_2)$
$S_{21} = \partial x_2/\partial p_1 + x_1 \cdot \partial x_2/\partial m = 0 + [am/p_1] \cdot [(1-a)/p_2] = a(1-a)m/(p_1 p_2)$
$S_{12} = S_{21}$ ✓
调整商品1的价格，观察马歇尔需求、希克斯（补偿）需求和收入效应如何响应。使用柯布-道格拉斯效用 $u(x_1,x_2)=x_1^a x_2^{1-a}$，$a=0.6$，$p_2=3$，$m=120$。
图 11.2.左：商品空间中的斯勒茨基分解。原始组合（蓝色）、补偿组合（橙色，在新价格下位于原无差异曲线上）和新组合（绿色）。替代效应从蓝色移至橙色；收入效应从橙色移至绿色。右：$S_{11}$ 和 $S_{12}$ 随 $p_1$ 变化的斯勒茨基矩阵项，确认半负定性（$S_{11} \leq 0$）和对称性。
考虑一个有 $I$ 个消费者和 $L$ 种商品的经济。消费者 $i$ 拥有禀赋 $\omega_i \in \mathbb{R}^L_+$ 和偏好 $\succsim_i$。
在价格 $p$ 下，消费者 $i$ 的财富为 $m_i = p \cdot \omega_i$。她的需求为 $x_i(p, m_i)$。
总超额需求：
均衡要求 $z(p^*) = 0$。
含义：（1）如果 $L - 1$ 个市场出清，第 $L$ 个自动出清。（2）只有相对价格重要——我们可以将一种价格标准化为1（计价物）。
证明策略（概要）。将价格标准化到单位单纯形 $\Delta$。定义价格调整映射 $f: \Delta \to \Delta$，提高超额需求商品的价格。由布劳威尔不动点定理，$f$ 有不动点 $p^*$。在不动点处，$z(p^*) = 0$——所有市场出清。
对于2消费者、2商品经济，埃奇沃思盒提供完整的可视化。盒的尺寸等于总禀赋。消费者1的原点在左下角，消费者2的在右上角。盒中每一点都是一个可行配置。
两个具有柯布-道格拉斯偏好的消费者。拖动禀赋点，探索瓦尔拉斯均衡、契约曲线和核心如何变化。
图 11.1（互动）。埃奇沃思盒。橙色点是禀赋。绿色点是瓦尔拉斯均衡。红色曲线是契约曲线（所有帕累托有效配置）。阴影核区域显示两个消费者都优于禀赋的配置。预算线通过禀赋点，斜率为 $-p_x/p_y$。
消费者1：$u_1 = x_1^{1/2}y_1^{1/2}$，禀赋 $(4, 0)$。消费者2：$u_2 = x_2^{1/2}y_2^{1/2}$，禀赋 $(0, 4)$。
市场出清给出 $p_x = p_y$，均衡配置为 $x_1^* = y_1^* = 2$，$x_2^* = y_2^* = 2$。
每个消费者用一半禀赋交换另一种商品，最终拥有等量的两种商品。
证明。我们用反证法。假设在价格 $p^*$ 下的瓦尔拉斯均衡配置 $x^*$ 不是帕累托最优的。则存在一个可行配置 $x'$，使所有人至少一样好而某人严格更好。
第1步。对于严格更好的消费者 $j$：由于 $x_j^*$ 是效用最大化的且 $x_j'$ 严格优于它，$x_j'$ 必定是不可负担的：$p^* \cdot x_j' > p^* \cdot \omega_j$。
第2步。对每个消费者 $i$：由局部非饱和性，$p^* \cdot x_i' \geq p^* \cdot \omega_i$。
第3步。求和：$\sum_i p^* \cdot x_i' > \sum_i p^* \cdot \omega_i$。
第4步。但可行性要求 $\sum_i x_i' = \sum_i \omega_i$，给出 $\sum_i p^* \cdot x_i' = \sum_i p^* \cdot \omega_i$。矛盾。$\square$
该证明只使用了局部非饱和性和预算耗尽。它不需要凸性、可微性或任何特定函数形式。这种一般性是该定理强大的原因。
解释。第一福利定理是亚当·斯密"看不见的手"的正式表述。竞争市场产生的配置，任何重新安排都无法在不使某人变差的情况下改善。但假设条件（完全市场、价格接受、无外部性、无公共品、完全信息）恰好界定了看不见的手失效的情形。
消费者1：$u_1 = x_1^{1/2}y_1^{1/2}$，禀赋 $(4, 0)$。消费者2：$u_2 = x_2^{1/2}y_2^{1/2}$，禀赋 $(0, 4)$。
由例11.4，均衡为 $x_1^* = y_1^* = x_2^* = y_2^* = 2$，$p_x = p_y$。
检验帕累托最优性：在均衡处，$MRS_1 = y_1/x_1 = 1$ 且 $MRS_2 = y_2/x_2 = 1$。由于 $MRS_1 = MRS_2 = p_x/p_y$，无差异曲线相切——配置在契约曲线上。
验证不存在帕累托改进：任何给消费者1更多商品 $x$ 的重新配置（比如 $x_1 = 3$）要求 $x_2 = 1$。则 $u_1 = \sqrt{3 \cdot y_1}$ 且 $u_2 = \sqrt{1 \cdot y_2}$，其中 $y_1 + y_2 = 4$。要使消费者1获益（$u_1 > \sqrt{4} = 2$），需要 \$1y_1 > 4$，即 $y_1 > 4/3$，则 $y_2 < 8/3$，给出 $u_2 = \sqrt{8/3} < 2 = u_2^*$。消费者2境况变差。不存在帕累托改进。
瓦尔拉斯均衡位于契约曲线上（帕累托有效）。切换"帕累托改进？"以验证：在均衡处，两个消费者都能获益的透镜形区域为空。在禀赋点处则不是。
互动 11.3。在均衡视图（不存在帕累托改进）和禀赋视图（阴影透镜显示互利交易）之间切换。均衡在契约曲线上的位置直观地证明了效率。
Dan Riffle, AOC's former policy aide, turned this line into a social media mantra — shared millions of times, printed on T-shirts, chanted at rallies. The claim is stark: billionaires don't exist because they created extraordinary value. They exist because the system is broken — tax loopholes, monopoly power, rigged rules. The First Welfare Theorem you just proved gives you the tools to test this precisely: does extreme wealth concentration represent the market working correctly (and we just dislike the endowment), or the market failing (and efficiency is not achieved)?
高级解释。第二福利定理说效率与公平是可分离的问题。社会可以通过两个步骤选择任何帕累托有效分配：
然后市场将产生一个既有效（由第一福利定理保证）又达到期望分配的竞争性均衡。
为什么对政策重要。不要为了实现公平而扭曲市场（那会牺牲效率）。相反，使用一次性转移进行再分配，然后让市场运作。右派含义：让市场自由运行。左派含义：想怎么再分配就怎么再分配。两者可以同时实现——在理论上。
为什么在实践中失败。一次性转移需要政府不掌握的关于个人类型的信息。现实世界的再分配使用扭曲性税收（所得税、资本利得税、财富税），这些改变激励并产生无谓损失。这一信息问题是机制设计（第11章）和最优税收（第16章）的主题。
在大型经济中，核配置集（没有联盟能改善的配置集）收缩到瓦尔拉斯均衡配置集。这就是核等价定理——竞争性均衡是经受住所有可能联盟竞争的唯一结果。
我们将玛雅的柠檬水市场建模为2消费者、2商品的埃奇沃思盒交换经济。
设定：玛雅和亚历克斯。两种商品：柠檬水（$L$）和饼干（$C$）。玛雅初始拥有45份柠檬水和0份饼干。亚历克斯初始拥有0份柠檬水和40份饼干。
偏好：$u_M = L_M^{0.5}C_M^{0.5}$，$u_A = L_A^{0.3}C_A^{0.7}$。
市场出清给出 $p_L/p_C = 8/15 \approx 0.533$。
均衡：玛雅：$(L_M, C_M) = (22.5, 12)$。亚历克斯：$(L_A, C_A) = (22.5, 28)$。
根据第一福利定理，该配置是帕累托最优的。
Arrow-Debreu（1954）：存在性证明。肯尼斯·阿罗和杰拉德·德布鲁证明了在弱假设（凸偏好、无外部性）下竞争性均衡存在。利用角谷静夫不动点定理，他们证明了存在一组使所有市场同时出清的价格——在《国富论》两个世纪之后正式化了亚当·斯密的"看不见的手"。
这一数学成就非常了不起：将问题归结为证明某个对应（超额需求作为价格的函数）满足不动点的条件。结果只需要局部非饱和性和凸性——不需要可微性或特定函数形式。
德布鲁的《价值理论》（1959）将这一框架提炼为严格的公理系统，使他获得了1983年诺贝尔奖。阿罗已于1972年因其对一般均衡和社会选择的广泛贡献获得诺贝尔奖。他们的存在性证明仍然是福利经济学以及本章证明的两个福利定理的数学基础。
You now have the formal welfare theorems — the definitive statement of when and why competitive markets produce efficient outcomes, and the precise conditions under which any efficient outcome can be decentralized.
The First Welfare Theorem delivers the strongest possible efficiency result: if preferences are locally nonsatiated and markets are complete and competitive, then every Walrasian equilibrium is Pareto optimal. You saw the proof — it works by contradiction, exploiting the fact that any Pareto improvement would require someone to afford a bundle they couldn't at equilibrium prices. The Second Welfare Theorem completes the picture: under convexity, any Pareto optimal allocation can be achieved as a competitive equilibrium after appropriate lump-sum redistribution of endowments. Together, these theorems say that the market mechanism is both sufficient for efficiency (First WT) and flexible enough to achieve any efficient outcome society desires (Second WT). The price system simultaneously solves the information problem (no planner needed) and the coordination problem (all markets clear).
The conditions of the First Welfare Theorem are exacting, and every one of them fails in important real-world markets. Complete markets require a market for every good, every state of the world, every date — this fails massively (you cannot buy insurance against most life risks, future markets are thin, contingent claims are incomplete). Price-taking fails in any market with significant firms (tech, pharma, airlines). No externalities fails for climate, pollution, network effects, and knowledge spillovers. Greenwald and Stiglitz (1986) proved the devastating result: whenever markets are incomplete — which is always — competitive equilibria are generically constrained-inefficient. That is, there exist interventions using only the same information and instruments available to markets that are Pareto improving. The theorem doesn't say markets are bad; it says the conditions for the First Welfare Theorem are a knife-edge that reality never hits.
The profession's relationship with the welfare theorems matured considerably after Greenwald-Stiglitz. The theorems are now understood not as claims that markets work, but as a diagnostic framework: they identify exactly which conditions must hold for efficiency, and deviations from those conditions point precisely to where intervention might help. The Second Welfare Theorem's promise — that you can separate efficiency from equity — is formally correct but practically hollow. Lump-sum transfers require the government to know each individual's type (ability, preferences, endowment) without distorting behavior. Any feasible transfer instrument (income tax, wealth tax, means-tested benefits) changes incentives and creates deadweight loss. This is the Mirrlees (1971) insight: optimal taxation is a constrained problem precisely because the Second Welfare Theorem's instrument doesn't exist.
The welfare theorems are the most important results in economics — not because they prove markets work, but because they identify exactly when and why markets work or fail. Understanding the theorems is prerequisite to intelligent intervention: every market failure is a specific violation of a specific condition. The First Welfare Theorem is a conditional claim, and the conditions rarely hold in full — but they hold approximately in enough settings to explain why markets coordinate as well as they do. The Second Welfare Theorem is theoretically beautiful and practically cruel: it tells you equity and efficiency are separable, then makes the separation instrument informationally infeasible. Real policy lives in the second-best world where every redistribution creates distortions.
If markets fail when the welfare theorem conditions aren't met, is there a systematic way to design better institutions? The welfare theorems tell you when markets work but not what to do when they don't. Come back in Chapter 12 (§12.1–12.5), where mechanism design asks exactly this question. The revelation principle, VCG mechanisms, and matching markets show that economic theory can engineer efficient outcomes — sometimes outperforming both unregulated markets and blunt government intervention. That's the final stop on this question.
The viral slogan meets the First Welfare Theorem. Some fortunes are market failures; others are surplus creation. The word "every" is where the claim breaks.
高级Sanders' viral rallying cry meets Arrow's 1963 paper. The moral force is real — but declaring a right doesn't solve the allocation problem.
中级| 标签 | 公式 | 描述 |
|---|---|---|
| 公式 11.1 | $e(p, \bar{u}) = \min p \cdot x$ s.t. $u(x) \geq \bar{u}$ | 支出最小化 |
| 公式 11.2 | $h_i = \partial e / \partial p_i$ | 谢帕德引理 |
| 公式 11.3–11.4 | $e(p, V(p,m)) = m$; $V(p, e(p,\bar{u})) = \bar{u}$ | 对偶恒等式 |
| 公式 11.5 | $h(p, \bar{u}) = x(p, e(p, \bar{u}))$ | 希克斯需求 = 补偿收入下的马歇尔需求 |
| 公式 11.6 | $x_i = -(\partial V/\partial p_i)/(\partial V/\partial m)$ | 罗伊恒等式 |
| 公式 11.7 | $S_{ij} = \partial h_i/\partial p_j = \partial x_i/\partial p_j + x_j \partial x_i/\partial m$ | 斯勒茨基矩阵元素 |
| 公式 11.8 | $z(p) = \sum_i x_i(p) - \sum_i \omega_i$ | 总超额需求 |