FRM Part I — Foundations of Risk Management

📈 Your progress: 0% (tick the ☑ boxes as you master each chapter)

风险管理的基石

1. The Building Blocks of Risk Management

学习单元 1 · 要点：风险的定义（≠ 损失大小）· 四步流程 · 预期损失 vs 非预期损失 · 七大类风险 · 风险因素交互与汇总挑战

🇨🇳 中文导览

本章解决的是「风险到底是什么、风险管理究竟在做什么」这个定义性问题，是全部内容的地基。考试里几乎全部以定性判断题出现：给你一句话让你判断对错，或给你一个情景让你归类属于哪一类风险。先记住四条硬结论：

risk ≠ size of loss风险不等于损失金额：风险是结果的不确定性。损失可以很大却完全可以预测（例如银行每年都能稳定预估的贷款违约损失），这种损失是预期的、可以定价和计提拨备的；真正需要管理的是损失的波动性，尤其是可能突然冲高的非预期损失。
四步流程 + 四个管理动作：识别 → 分析与度量 → 评估风险事件的影响 → 管理风险。最后一步是「管理」而不是「度量」；管理动作有四个选项——avoid回避 / retain保留 / mitigate缓释 / transfer转移；「规避一切已知风险」不是流程的组成部分。风险管理不会在经济整体层面消除风险——它只是把风险从一方转移给另一方（零和游戏）。
EL = PD × EAD × LGD预期信用损失：预期损失是「做生意的正常成本」；非预期损失之所以会飙升到极端水平，根源是相关性风险——不利事件成簇同时发生（借款人违约的同时抵押品价值也在下跌）。
七大类风险 + 汇总的难处：市场 / 信用 / 流动性 / 操作 / 法律与监管 / 业务与战略 / 声誉，要能「看情景归类」；风险因素之间会交互、相关性在危机中上升，所以不要指望单一数字代表企业风险——要配合预期尾部损失（ES）、情景分析与压力测试使用。

关键术语：expected loss预期损失、unexpected loss非预期损失、correlation risk相关性风险、three lines of defense三道防线、economic capital经济资本、RAROC风险调整资本回报率。

🧭 知识网位置本章是全书奠基层——先把「风险是什么、怎么管」定义清楚。本章之后 → 第 2 章 · 公司如何管理金融风险（把四步流程落成具体策略）第 3 章 · 治理与三道防线第 9–10 章 · 金融灾难与 2007–09 危机（这些概念的真实代价）

1.1 What is "risk"? And risk management vs. risk taking 风险管理 vs 风险承担

这一节的考点是两个"不等于"。第一，风险不等于潜在损失的大小：一笔损失即使金额很大，只要高度可预测（例如银行每年稳定比例的违约损失），它就是"已知成本"，可以定价、计提拨备；风险管理的重心是损失的变异性，尤其是会突然冲高的非预期损失。第二，风险管理不等于风险承担：前者是防御性的、先衡量"预期回报是否配得上预期风险"；后者是主动加风险去换取额外收益（机会主义行为）。四步流程的顺序必须记牢——识别、分析与度量、评估影响、管理风险——题目常把"度量"放在最后一步来骗你，也常把"规避所有风险"当作流程的一环来骗你。本节再加两个记忆点：风险光谱（已知 → 已知的未知 → 未知的未知）与四类管理动作（avoid / retain / mitigate / transfer）。

Risk is the uncertainty surrounding outcomes 结果的不确定性. In investing, we care most about negative surprises (unexpected losses) — not positive ones. There is a natural risk–reward tradeoff: higher expected return usually demands accepting higher risk.

⚠️ Exam trap — "risk" is NOT the same as "size of loss"

A loss can be large yet highly predictable (e.g., a bank's stable, estimable annual loan-default losses). Such predictable losses are expected and easily provisioned for. What really matters in risk management is the variability of the loss — especially an unexpected loss that could spike far above the average.

key concept Risk management = a formal, ongoing process to identify, measure, and control risk so perceived reward justifies expected risk. （防御性：先问「预期回报配不配得上预期风险」）
Risk taking = actively accepting more risk to chase more reward (the entrepreneur chasing upside). （机会主义：主动加风险换增量收益）

The Four-Step Risk Management Process 风险管理四步流程

Identify relevant risks 识别风险
Analyze & measure risks 分析与度量风险
Evaluate the potential impact of risk events 评估风险事件影响
Manage / control the risks 管理风险

Note the ordering — "identify → analyze → evaluate → manage". 考题常设陷阱：哪个不是上述步骤（答案："避免所有已知风险"）。放大的记忆点：第 4 步"管理"有四类动作——avoid（回避：砍掉产品线、退出某市场）、retain（保留：回报配得上风险就自己承担）、mitigate（缓释：降低风险的频率或幅度）、transfer（转移：用衍生品把风险转给对手方，或买保险外包给保险公司）。风险管理不能在经济整体层面消除风险——它把风险转移给另一方，是零和游戏。

Known / known unknowns / unknown unknowns — 风险的光谱

风险按「已知程度」分三段：expected（known）已知、可计量 → known unknowns 知道可能发生但难以量化（又称 Knightian uncertainty 奈特不确定性）→ unknown unknowns 真正未知（即尾部事件）。风险管理的方向是把风险尽量「拉进已知」，但无法量化的部分做不到——这正是模型之外还需要判断、情景分析和压力测试的原因。

一个宏观提醒：风险管理要对经济有益，前提是风险被足够分散到愿意且有能力的参与者手里；2007–09 危机暴露的教训之一，恰恰是风险过度集中在太少的人手里。（衍生品还会放大「随大流」行为、可能让资产负债表被高估——这些在灾难案例章节展开。）

1.2 Tools to measure & manage risk 度量与管理工具

度量工具与管控工具要分开记。定量度量：VaR（在给定置信水平下的损失门槛值，注意它只给出临界点，不告诉你尾部到底有多深）、经济资本（吸收非预期损失所需的流动性资本）、RAROC = 税后风险调整预期收益 ÷ 经济资本（分子要先扣减预期损失，算出来要高于股权成本才算可接受）。定性度量：情景分析（最好/最坏情景，同时冲击多个变量）与压力测试（一次只调一个参数；反向压力测试是先设定最坏结果再倒推成因）。管理层面靠 ERM。本节最容易丢分的点是：不要把企业层面风险压缩成单一数字——VaR 有多个版本、假设可被人通过期限和置信度调节、而且无法衡量尾部幅度，所以要用一组指标互相印证。

Type	Quantitative tools 定量	Qualitative tools 定性
Measure	Value at Risk (VaR) 在险价值Economic Capital 经济资本RAROC	Scenario analysis, stress testing 情景分析/压力测试
Manage	Enterprise Risk Management (ERM) 企业风险管理 — a firm-wide, holistic framework

Don't reduce the firm to a single number

VaR gives a loss threshold; but it has assumptions, can be "gamed" via the confidence level, and different VaR methods differ. Economic capital = capital needed to survive unexpected losses. RAROC = risk-adjusted return on capital — 分子扣减预期损失后，与股权成本比较，只有超过才可接受；四大用途：①业务比较（跨部门换算到同一把尺子）②投资分析（新业务准入）③定价策略（收益是否配得上风险）④风险管理（找出"风险没被回报覆盖"的区域）。ERM 的精神：企业级、跨部门、不搞部门孤岛，把风险规划并入战略规划；既要统计，也要判断。

1.3 Expected Loss vs. Unexpected Loss 预期损失 vs 非预期损失 ⭐

EL 与 UL 是本章最核心的一对概念，几乎必然出题。EL = PD × EAD × LGD：PD 违约概率、EAD 违约时的风险敞口、LGD 违约损失率（1 − 回收率）。EL 是"做生意的正常成本"，可以定价转嫁给客户（对高风险借款人收更高的利差），建模可靠时甚至可以当成费用而非风险。UL 是超出平均损失的那一块，几乎无法预测；它之所以会被推到极端，是因为相关性风险——坏事件成簇同时发生（经济衰退时车贷集体违约；房贷违约的同时房价下跌使回收率恶化）。解题要点：看到"可预测、可计提、可定价"选 EL；看到"同时爆发、相关性上升、超出均值"选 UL。管理抓手：把 PD 拆解成可监控的损失驱动因素（granular loss drivers，如借款人财务状况、外部环境），这是从"知道有风险"到"能动手管"的桥梁。

Expected Loss (EL) — 预期损失

The loss you expect in normal business, reliably predictable over short horizons. In banking it is modeled as:

credit EL formula — must memorise ⭐（Part II《信用风险》模块还会深入展开） EL = PD × EAD × LGD
PD = probability of default 违约概率 EAD = exposure at default 违约暴露/风险敞口 LGD = loss given default 违约损失率

EL is a predictable cost of doing business — e.g., a retailer's "bad debt expense", or a bank's provision. It is priced into the product (higher spread to riskier borrowers).

Unexpected Loss (UL) — 非预期损失

The loss above the average expected loss — genuinely hard to predict. Why do ULs rise to extreme levels? Because of correlation risk 相关性风险unfavorable events happen togetherin clusters.

💡 Real-world illustration — the auto-loan bank

Imagine a bank lending only to auto manufacturers.

In a boompeople buy cars, few defaults → loss ≈ expected. ✅
In a recessioneveryone defaults at the same time → losses cluster far above the average → this is the unexpected loss. ❌

Another correlation example: real-estate loans — borrowers default while property values (collateral) fall, so both default rate and recovery rate move against you simultaneously.

1.4 Risk & reward; conflicts of interest 风险与回报、利益冲突

风险与回报的正向关系在实务中会被三类因素搅浑：流动性差异、税收、以及随时间变化的风险容忍度——当市场整体风险容忍度高时，无风险债与风险债之间的利差会被压到异常低，真实的风险回报关系就被掩盖了（这是常见的正确选项）。补充一个值得注意的观察：对流动性差或不公开交易的资产（如私募股权），风险回报关系更难评估。利益冲突的根源是：最了解风险的人往往正是能从中获利的人（流氓交易员；为抬高短期股价和奖金而隐藏风险的管理者），而且股权激励还会加剧这一点。标准解法是三道防线：一线（前线员工与部门经理识别风险）→ 二线（风险管理体系的日常监督）→ 三线（独立的定期内审，验证前两道是否真的在运转）。

Higher potential reward = more risk. But the risk–reward link is muddied by: liquidity differences, taxes, and fluctuating risk tolerance 风险容忍度. When risk tolerance is high, riskless/risky yield spreads narrow and the true risk–reward relationship gets disguised.

Conflicts of interest — the insider danger

Those who know most about a risk are sometimes the ones who profit from it (rogue traders; managers hiding risk to boost short-term stock price and their own bonus). Even stock-based compensation can worsen this by encouraging short-term risk-hiding. The fix is three lines of defense:

First line — frontline employees & division managers recognise risk.
Second line — a robust risk-management system with daily oversight.
Third line — periodic independent internal audits.

1.5 The key classes of risk 主要风险类别 ⭐

七大类风险必须能"看情景归类"：市场风险（利率/股票/汇率/商品）、信用风险（违约/破产/评级下调/交易对手）、流动性风险（融资流动性 vs 市场/交易流动性）、操作风险（内部流程失效、人为失误或外部事件——含系统与网络、内控不足、欺诈、自然灾害、流氓交易员）、法律与监管风险（诉讼与规则变化）、业务与战略风险（收入成本波动 / 长期战略押注）、声誉风险（公众信任受损，通常是其他风险事件的结果）。信用四子类的判别：违约=不还本付息；破产=对手方停止经营且抵押品清算价值不足；评级下调=资信变差导致后续融资成本上升；交易对手=衍生品亏损方拒付，其中结算（Herstatt）风险指一方已付款而另一方未交货（得名于 1974 年德国 Herstatt 银行倒闭）。例题口径：应收 50 万只收回 40 万 → 回收率 80%、LGD 20%。

Risk class 类别	What it is 含义	Sub-types / examples
Market risk 市场风险	Loss from changing market prices/rates	Interest-rate 利率equity 股票FX 汇率commodity 大宗商品
Credit risk 信用风险	Counterparty fails to meet obligations	Default, bankruptcy, downgrade, counterparty (incl. settlement / Herstatt risk)
Liquidity risk 流动性风险	Can't fund or can't sell	Funding liquidity 融资流动性 & market/trading liquidity 市场流动性
Operational risk 操作风险	Failed internal processes, human error, external events	Fraud, rogue trading, cyber, human error, tech failure
Legal & regulatory risk 法律与监管风险	Lawsuits / regulatory rule changes	Litigation, tax/margin rule changes
Business & strategic risk 业务与战略风险	Variability in revenues/costs; long-term strategy bets	Demand shifts, failed new products, risky lending standards
Reputation risk 声誉风险	Loss of public trust / fair-dealing perception	Often the outcome of another risk; amplified by social media

💡 Credit risk detail — default vs bankruptcy vs downgrade vs counterparty

Default risk — borrower misses interest/principal (PD is key).
Bankruptcy risk — counterparty stops operating; collateral may be insufficient.
Downgrade risk — credit rating worsens → lender charges more; can cascade into default.
Counterparty risk — the losing side of a derivative refuses to pay. Includes settlement / Herstatt risk 结算风险 — one party pays but the other doesn't deliver (named after the 1974 Herstatt Bank collapse).

Example: you're owed $500k on a futures contract but only receive $400k. Recovery value = $400k (recovery rate 80%), LGD = $100k (20%).

1.6 Risk-factor interaction & aggregation challenges 风险因素交互与汇总挑战

汇总为什么难？因为单个风险因素并不独立：一个细颗粒的违约驱动因素会沿着多米诺链条传染到信用、操作、业务和声誉风险；而相关性在危机中会上升，让分散化同时失效。技术层面还有三个障碍：期权希腊字母（delta/gamma/vega/theta）无法直接加总到企业层面；多笔衍生品敞口可能互相抵消，此时名义本金失去意义；相关性本身不稳定。因此对衍生品而言汇总更难而不是更容易。补救方向是：VaR 之外补充预期损失尾值（expected shortfall）——它给出的是"超过 VaR 门槛之后的平均损失"，能刻画尾部幅度——以及情景分析与压力测试。

Risks don't act alone

Extreme tail-risk events are most dangerous when risk-factor correlations rise (everything falls together). Triggers: structural change, behavior shifts, government intervention, innovation. Aggregation is hard because: option Greeks 期权希腊字母 (delta/gamma/vega/theta) can't simply be summed; derivative notionals can cancel; correlations are unstable.

🔑 英文要点速查 · English Essentials

risk 风险 — The uncertainty surrounding outcomes; risk management focuses on the variability of losses, not their size.
risk management 风险管理 — A formal, ongoing process to identify, measure, and control risk so that perceived reward justifies expected risk (defensive).
risk taking 风险承担 — The active acceptance of incremental risk in pursuit of incremental gains (opportunistic).
expected loss (EL) 预期损失 — Average loss expected in the normal course of business; in banking EL = PD × EAD × LGD; predictable → provisioned & priced.
unexpected loss (UL) 非预期损失 — Loss in excess of the average; pushed to extreme levels by correlation risk (unfavorable events cluster together).
economic capital 经济资本 — Liquid capital held to absorb unexpected losses.
RAROC 风险调整资本回报率 — After-tax risk-adjusted expected return ÷ economic capital; accept only if it exceeds the cost of equity.
three lines of defense 三道防线 — Frontline recognition → risk-management oversight → periodic independent audits.
Knightian uncertainty 奈特不确定性 — Known unknowns: risks recognised as possible but not quantifiable; unknown unknowns ≈ tail risk.
settlement (Herstatt) risk 结算风险 — One party pays out but the counterparty fails to deliver its side.
scenario analysis / stress testing 情景分析 / 压力测试 — Best-vs-worst-case multi-variable shocks vs one-parameter-at-a-time; reverse stress testing starts from losses and traces the exposures.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. A bank estimates that its loan portfolio will incur an average annual loss of about $50 million — an estimate that has been stable for years. Management worries about an extreme scenario in which losses far exceed this level. Which statement best reflects the concept of risk in risk management?

A. The portfolio's risk is the $50 million of expected loss.

B. What must be managed is the variability of losses around the average — especially the unexpected loss, which can spike far higher. ✅

C. Because the loss is predictable, the portfolio carries no risk that needs managing.

D. Risk is equivalent to the maximum possible size of a loss.

B 正确。考点：risk ≠ size of loss；EL vs UL
风险是结果的不确定性（variability），不是损失规模：可稳定预估的 $50M 是预期损失——可定价、可拨备的「做生意成本」；真正要管的是围绕平均值的波动性，尤其是会异常冲高的非预期损失。A、C、D 都是把风险与损失金额画等号的经典陷阱。

Q2. Which of the following is NOT one of the steps of the risk management process?

A. Identifying relevant risks.

B. Analyzing and measuring risks.

C. Adopting a policy to avoid all known risks. ✅

D. Evaluating the impact from risk events.

C 为答案（它不属于流程步骤）。考点：四步流程 + 「规避一切」陷阱
四步流程是：识别 → 分析与度量 → 评估影响 → 管理风险。「规避全部已知风险」既不可行、也不是流程的一步——真实的「管理」是在 avoid / retain / mitigate / transfer 四类动作中做取舍；承担可管理的风险本身就是获取回报的路径之一。

Q3. A bank has a $5,000,000 exposure to a corporate borrower. The estimated one-year probability of default is 2%, and the loss given default (LGD) is 40%. What is the expected loss (EL) on this exposure?

A. $40,000 ✅

B. $100,000

C. $200,000

D. $20,000

A 正确。考点：EL = PD × EAD × LGD
EL = 2% × $5,000,000 × 40% = $40,000。两个口径提醒：①若题目给的是回收率 60%，则 LGD = 1 − 60% = 40%，结果相同；②EL 只是「平均损失」，之上的不可预测部分是非预期损失（UL），由相关性风险推向极端——EL 与 UL 要分开理解。

Q4. Which statement about risk aggregation at the enterprise level is most accurate?

A. Enterprise risk should be reduced to a single number (e.g., VaR) for simplicity.

B. Expected shortfall communicates the magnitude of losses beyond the VaR threshold and complements scenario analysis and stress testing. ✅

C. Risk aggregation is easiest for derivative positions, because notional amounts can simply be added up.

D. Option Greeks (delta, gamma, vega, theta) can be summed across positions to streamline aggregation.

B 正确。考点：汇总挑战 + VaR 的缺陷
VaR 只给出一个损失门槛，还会被参数调整影响；预期尾部损失（ES）刻画门槛之外的尾部平均损失，配合情景分析与压力测试构成更完整的图景。A 错在「单一数字」思路；C、D 恰好说反——衍生品让汇总更难：多笔敞口可能互相抵消、名义本金失去意义，期权希腊字母也无法直接加总到企业层面。

企业如何管理风险

2. How Do Firms Manage Financial Risk?

学习单元 1 · 要点：四种策略（accept / avoid / mitigate / transfer）· 风险偏好与五步流程 · 风险绘图 · 对冲的优势/劣势/挑战 · 操作与财务风险 · 风险限额与衍生品工具

🇨🇳 中文导览

本章回答「企业拿到一份风险敞口后，具体有哪些处置办法、整套流程怎么走」。考试形式通常是给一个业务情景让你选策略（接受/规避/缓释/转移）或选工具；也会考对冲「理论上无效、实务中有效」的原因，以及各类风险限额的固有弱点。

四种策略看动机：接受（影响小、股东想要该敞口、或成本可直接加价转嫁给客户）；规避（风险与核心业务无关，如综合集团剥离金融板块）；缓释（风险属于本业但太大，降低频率或幅度，如加抵押、上自动化——内部动作、不动金融资产）；转移（保险或衍生品交给第三方，必然引入交易对手风险）。
风险偏好（risk appetite）＝「愿意承担」的风险水平与类型，由意愿 + 能力构成——能力（内控、监管约束）给意愿设上限。三个量要区分：capacity ≥ appetite ≥ 实际敞口，且实务中留出误差边际（如 capacity $200M → appetite $170M → 实际上限 $150M）。
先绘图、后选工具：五步流程＝设定偏好 → 风险绘图 → 把偏好操作化 → 实施 → 监控调整。绘图要细到五个维度：规模、时点、地点、口径（现值/预测值）、可轧差空间；至少覆盖前十大敞口。
对冲的"理论无效、实务有效"：MM（1958）与 CAPM（1964）在完美市场假设下认为对冲不创造价值；这些假设都不考虑财务困境成本，所以现实中不成立。最大实战挑战是沟通失败——MGRM（1993）是经典案例。

关键术语：hedging对冲、basis risk基差风险、risk mapping风险绘图、risk limits风险限额、static vs dynamic hedge静态/动态对冲、notional limits名义本金限额。

🧭 知识网位置承接第 1 章的「四步流程」，把「管理」一步展开成完整操作体系。前置 → 第 1 章 · 风险管理框架本章之后 → 第 3 章 · 治理（把偏好制度化）第 4 章 · 信用风险转移机制（transfer 的深水区）

2.1 The four risk-management strategies 四种风险管理策略

四种策略的判别不看名字，看动机与手段：股东想要这个敞口 → 接受（金矿主保留金价敞口；成本可直接加价转嫁给客户时同理）；风险与核心业务无关 → 规避（出售整个业务单元，例如 GE 在危机后剥离金融板块以保护工业主业）；风险属本业但太大 → 缓释（银行提高利率、缩短期限、加抵押；制造商自动化对冲人力成本；是内部动作、不动用金融资产）；把风险交给第三方 → 转移（保险、衍生品，必然引入交易对手风险）。最容易混的是「缓释 vs 转移」——判别标准就一条：有没有动用衍生品/保险这类金融合约。理性决策：先做最坏情景成本测算，再比较「缓释成本」与「转移成本」（教材例：网络安全风险最坏损失 $75M，该风险属于核心业务不能规避，于是在"买保险"与"添置设备缓释"之间比成本收益——最终常是混合方案）。

Strategy 策略	When to use it 何时使用	Real-world example
Accept / retain 接受/自留	Impact is small; owners want the exposure; cost can be passed through to customers	Gold miner keeping gold-price exposure；成本可加价转嫁的产品线
Avoid 规避	Risk is not a natural part of the core business	GE 在 2007–09 危机后出售金融板块，保护工业主业
Mitigate 缓释/降低	Risk is natural but too large — reduce its frequency or severity	贷款加抵押/提利差；工厂自动化；升级节能设备（内部动作）
Transfer 转移	Pay someone else to bear it（成本：保费/衍生品成本 + 对手方风险）	Insurance；derivatives（对冲）

教材例：从"最坏情景"到策略组合

某公司测算网络安全威胁最坏情景损失 $75M。该风险与核心业务相关，不能选择规避；下一步比较「购买保险的成本」与「购置新设备的缓释成本」，选成本收益最优者——教材给的方向是缓释与转移的组合（mitigation + transfer）。

2.2 Risk appetite & the five-step process 风险偏好与五步流程

风险偏好由两部分组成：意愿（willingness）——为达成目标想承担多少；能力（ability）——内控与监管约束能给意愿设上限（例：银行的杠杆率不得低于 3% 是一条硬约束）。三个量必须分清并排序：风险容量 capacity（能承受的上限）≥ 风险偏好 appetite（愿意承担的）≥ 实际执行限额（还应再留误差边际）——教材例：capacity $200M → appetite $170M → 实际敞口上限 $150M。落地路径是五步流程：① 设定风险偏好 → ② 风险绘图 → ③ 把偏好操作化（转成限额与流程）→ ④ 实施 → ⑤ 监控并调整。董事会要让目标「可执行」：对外用简版声明、对内用详版，且定性（哪些风险保留不对冲）与定量（VaR、压力测试）都可以用。天然的政治张力：债权人只想守住下行（息收封顶），股东愿意博低概率大风险换权益回报——偏好设定必须在这两者间做取舍。

Identify risk appetite 设定风险偏好
Map known risks 风险绘图
Operationalize the appetite 操作化（转为限额/流程）
Implement the plan 实施
Monitor & adjust 监控与调整

Capacity vs appetite vs actual exposure — 三个量，一条链

Risk capacity = 公司能承受的最大风险（客观上限）。Risk appetite = 董事会愿意承担的风险水平与类型（主观选择，低于 capacity）。Actual exposure = 实际敞口，还应低于偏好——因为风险估计永远有误差，要留margin for error。典型链路：capacity $200M → appetite $170M → 敞口上限 $150M。

董事会设定偏好时必须权衡的七件事（速记）

Unity：七大类风险（市场/信用/流动性/操作/法律监管/业务战略/声誉）的偏好要互相自洽；
Entrepreneurial：哪一类风险多承担一点能成为竞争优势？
Correlated layers：一个业务单元可能叠加高度相关的多重风险（海外子公司＝操作 + 汇率同步动）。
Time horizon：短期对冲与长期对冲可能互斥；为远期风险付费会压低近期利润。
Tolerance bands：给经理层留可操作的容忍区间，兼顾纪律与机会灵活性。
Reputation：偏好声明对利益相关方传递了什么信号。
Measurement：企业级用什么度量——单一指标（VaR）不可靠，需要按业务模型选择组合指标。

2.3 Risk mapping 风险绘图

风险绘图＝把「所有已知风险」系统盘点成一张带现金影响的地图，是偏好落地的第二步。颗粒度决定质量，至少要问清五个维度：规模（需要 100 万还是 1000 万磅铜？）、时点（一次性还是分十批？）、地点（一个基地还是五个国家？）、口径（按当前值还是预测值）、以及轧差空间（外币收入中有一部分要留作经营用途，天然对冲掉一部分敞口）。应同时考虑风险因子之间的相关性与「一个风险抵消另一个」的可能；覆盖范围至少前十大敞口，最好全部。绘完图，才轮到选 retain / avoid / mitigate / transfer 的组合。

💡 两个绘图示例（教材）

铜业制造公司：期货合约现成可用，但还必须回答——规模（1M vs 10M 磅）、时点（特定一天 vs 分 10 期各 10%）、地点（一个厂 vs 五个国家）——粒度不够，图纸就是废纸。
跨国销售公司：12 个国家、8 种货币；要决定口径（只算现有销售、还是含高概率预计销售），并识别轧差空间（部分外币要留存经营，不能全对冲）。

2.4 Hedging — advantages, disadvantages & challenges 对冲的利弊与挑战

本条线考「实务优势 vs 理论劣势」的对照。实务优势：降低盈利/现金流波动 → 提高举债能力、贷款条款更少；改善现金流（套保盈利、平滑税负）；向债权人/客户/员工传递稳定信号；便于经营计划、锁定高毛利；长期看常比买保险便宜。理论劣势（全部建立在完美市场假设上）：MM（1958）——投资者可自行复制对冲，公司价值不变；CAPM（1964）——只有系统性风险被定价，对冲公司特有风险没有意义。破题关键：这些理论都没有考虑财务困境与破产成本；市场不完美、分散化有交易成本，所以对冲在实务中确实增值。实战的坑要背三个：误解/漏画敞口（对冲方向或规模错）；市场动态变化导致敞口漂移；沟通失败——MGRM（1993）就是母公司不信、强平头寸、被市场反向轧两次的经典案例。

Advantages of hedging ✅

Lower cost of capital（降低债务/权益成本）→ 更高的 debt capacity：现金流越稳，借款人给的条款与限制越少。
Cash-flow enhancement：套保头寸可能盈利；平滑后的应税收入也可能降低税负（直接影响现金流）。
Signaling：稳定的业绩向贷款人、客户、供应商、员工传递实力信号，往往体现在股价上。
Business planning 更容易；可以把好毛利「锁定」；对管理层来说还有声誉与薪酬的附加效应。
Often cheaper than insurance：但要比较多年保费总额与估计损失。

Disadvantages of hedging（理论口径）❌

Modigliani–Miller (1958)：完美市场（无税、无交易成本）下，公司对冲与投资者自建对冲等价 → 公司价值不变。假设不现实，论点薄弱。
CAPM / Sharpe (1964)：完美市场下只有 systematic risk（贝塔）被定价，公司特有（idiosyncratic）风险可用分散化免费消除 → 对冲它没有意义。同样不现实。
「对冲是零和游戏」的说法也不成立：它假设衍生品定价完全反映所有风险因子，而现实里衍生品定价远比股票/债券复杂——未必准。

Disadvantages of hedging（实务口径）及挑战 ⚠️

无形成本：管理层注意力从主业转移；合规成本（披露、审计、监控）。
衍生品的杠杆与复杂性可能带来意料之外的新敞口；也可能泄露不愿公开的经营信息（披露要求）。
挑战一：低估或画错敞口 → 名义规模过大/过小、该保的没保。
挑战二：市场在动，敞口随商品/汇率/利率变化，对冲策略必须同样动态；管不过来时，错误的对冲会放大亏损。
挑战三：沟通失败——MGRM 案例（见下）。
挑战四：缺技能/研究/时间 → 可外包给可信的第三方，但随之要管理该方的风险。

💡 MGRM（1993）——沟通失败的名场面

Metallgesellschaft 的美国子公司 MGRM 签约向客户供应 1.5 亿桶油品（跨度 10 年以上），用滚动短期期货 + OTC 互换对冲。油价变动引发巨额保证金追缴，母公司不信这套策略、下令平掉全部对冲头寸、锁死大亏；随后油市反转，未对冲的敞口再亏第二波。若沟通充分、母公司理解策略，这两波损失本可避免。

建立内部风险文化的四条（教材）

定期沟通风险目标与「即将触碰限额」的预警信号；
培训关键岗位，统一对风险管理目标的理解；
让关键岗位明白触碰限额的后果；
董事会能一口说出公司的前十大风险。

2.5 Hedging operational & financial risks 对冲操作与财务风险

先把两类风险分清：操作风险是生产经营与销售层面（利润表）——典型是投入品价格风险（买远期/期货锁定采购成本）与汇率风险（收入套保：货币看跌期权锁定下限、远期锁定确定汇率；资产负债表套保：盯海外净货币资产，可用外币负债做自然对冲）。财务风险是资产负债表层面——典型是利率风险（利率互换/swaption 把浮动换固定、降低借款成本）。两个补充考点：套保成本过高时企业会故意保留部分敞口；策略分静态对冲（一次性匹配、盯住基差、后续不动）与动态对冲（敞口随时间变化、需频繁再平衡，交易成本与监控成本更高）。再加三个实施细节：期限要匹配（绩效评估的口径也要一致）；会计上若非完全对冲会有损益入表；衍生品税务处理复杂、各国不同，是隐性成本。

Risk hedged	Typical tool	Intuition 直觉
Price / commodity risk 价格/大宗商品风险	Futures, forwards（锁定采购或售价）	把输入成本或产出价格提前固定
FX risk — revenue 汇率风险（收入端）	Currency put options（保底）、forwards（锁定汇率）	保护外币销售回款的本币价值
FX risk — balance sheet 汇率风险（资产负债表）	Forwards；外币负债（natural offset）	对冲海外净货币资产的汇兑波动
Interest-rate risk 利率风险	Interest-rate swaps / swaptions	浮动换固定（或反向），控制净敞口

2.6 Risk-management tools: limits & derivatives 风险限额与衍生品工具

先「定剂量」（rightsizing）：把风险管理当成成本中心（最小化负面冲击）还是利润中心（对净利有正贡献），决定投入规模与限额体系。八类限额每类都要能说出固有弱点（下表，考频高）。工具层面：远期（OTC、完全定制、现金或实物交割）、期货（场内、标准化、清算所降低对手方风险）、互换（OTC、交换经济头寸）、看涨/看跌期权（权利非义务）、奇异期权（如亚式期权按均价结算）、swaption（未来某日进入互换的权利）。场内外之别是一条主线：场内流动性好、成本低、无对手方风险，但标准化可能"配不上"需求 → 基差风险；OTC 完全定制，但流动性差、更贵、含对手方与结算风险。

Limit 限额	Purpose	Potential weakness 固有弱点 ⭐
Stop-loss 止损	设定损失阈值，阻止亏损继续扩大	只管"已实现损失"，不解决未来潜在损失
Notional 名义本金	限定名义敞口规模	名义额与实际风险可能关系不大
Risk-specific 专项	盯一类具体风险（如流动性、汇率）	难在集团层面汇总；依赖特定技能人才
Maturity / gap 期限/缺口	限制单期到期规模，平滑到期压力	不解决价格风险
Concentration 集中度	限制对手方/产品集中度	不直接处理相关性——不集中也可能同涨同跌
Greek 希腊字母	期权敏感度（delta/gamma/theta/vega）限额	易受模型风险与估计误差影响
VaR 在险价值	聚合的风险门槛	不给门槛之外的幅度；模型风险；参数可调
Stress / scenario 压力测试	特定压力点与多压力点组合	依赖对内部敞口的深度理解；难确认是否全覆盖

💡 教材例：航空公司与喷射燃油

航空业竞争的残酷之处：无法把燃油波动直接转嫁给乘客，所以必须要稳定。但市场没有喷射燃油的场内产品，只能用原油类产品近似 → 基差风险（还需管理原油-航油价差）；OTC 产品可以对准航油，但要承担对手方风险。Delta 曾另辟蹊径——买下一座炼油厂（纵向一体化）控制燃油成本，代价是把炼油业的风险全部接了进来。风险管理没有免费午餐，只有取舍。

交换 vs 场内、远期 vs 期货（一句话各一条）

Forward：OTC 定制，远期交割；Futures：场内标准化 + 清算所（消除对手方风险），流动性/成本占优。
定制与流动性天然矛盾：越贴合你的敞口（OTC），越难转手、越贵、越担对手方；越标准化（场内），越便宜好卖，但可能配不上 → basis risk。

🔑 英文要点速查 · English Essentials

accept / avoid / mitigate / transfer 接受/规避/缓释/转移 — The four strategies; retain when owners want the exposure or costs pass through; transfer (insurance/derivatives) introduces counterparty risk.
risk appetite 风险偏好 — The level and types of risk a firm is willing to retain, shaped by willingness and bounded by ability (controls, regulation).
risk capacity 风险容量 — The maximum risk the firm can bear; appetite is set below it, and actual exposure below that (margin for error).
risk limits 风险限额 — Quantitative ceilings (stop-loss, notional, VaR, concentration...) that operationalise risk appetite; each has a known weakness.
risk mapping 风险绘图 — Inventory of known risks by magnitude, timing, location, measurement basis and netting potential; at least top-10 exposures.
basis risk 基差风险 — The hedge is imperfect: the instrument used doesn't move exactly with the exposure (e.g., crude oil vs jet fuel).
static hedging 静态对冲 — One-off matching of a position with a hedging vehicle; minimal maintenance.
dynamic hedging 动态对冲 — Rebalancing as the underlying position changes; more faithful but costlier and monitoring-heavy.
Modigliani–Miller MM 定理 — Under perfect markets hedging doesn't change firm value (investors replicate it themselves); real frictions (taxes, distress costs) break the argument.
signaling 信号效应 — Stable operations signal strength to lenders, customers, suppliers and employees.
rightsizing 定剂量 — Deciding the appropriate scale of the risk-management program (cost center vs profit center).

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. A gold-mining company sells futures contracts on gold to lock in the price of a large part of its expected production. Which high-level risk-management strategy does this BEST represent?

A. Retain / accept the risk.

B. Avoid the risk.

C. Mitigate the risk.

D. Transfer the risk. ✅

D 正确。考点：四策略判别（缓释 vs 转移）
动用衍生品 = 把价格风险交给第三方 = transfer。判别线只有一条：缓释是内部动作（加抵押、自动化、内控），转移一定动用金融合约（衍生品或保险）。A 反了——金矿主保留敞口才叫 accept（股东想要金价敞口）；B、C 均为干扰项。

Q2. Which statement about a firm's risk appetite is MOST accurate?

A. It equals the maximum risk the firm is able to bear.

B. It reflects what the firm is willing to retain, is set below risk capacity, and may be expressed qualitatively and/or quantitatively. ✅

C. It is set by risk managers independently of the board of directors.

D. It must be expressed as a single quantitative number (e.g., VaR).

B 正确。考点：capacity / appetite / actual 三量关系
A 混淆了 capacity（能力上限）与 appetite（愿意承担的量）：appetite ≤ capacity，实际敞口还要再低一截（留 margin for error，课堂例：200M → 170M → 150M）。C 错——偏好由董事会/高管设定。D 错——定性、定量（VaR、压力测试）都可以，且单一数字并不可靠。

Q3. Under the perfect-capital-market assumptions of Modigliani–Miller (1958) and the CAPM (1964), which conclusion about corporate hedging is correct — and why can it fail in practice?

A. Hedging always destroys value, because derivatives are zero-sum.

B. Hedging cannot create value under those assumptions, because investors can replicate it (MM) and only systematic risk is priced (CAPM) — yet the assumptions ignore taxes, transaction costs and the costs of financial distress, which is why hedging can add value in practice. ✅

C. Firms should hedge all risks, since both systematic and unsystematic risk are priced.

D. Hedging is irrelevant under those assumptions because derivatives pricing always reflects every risk factor.

B 正确。考点：对冲"理论无效、实务有效"
MM：完美市场里投资者可自建对冲 → 公司对冲不改变价值；CAPM：只有系统性风险被定价 → 对冲特有风险"没价值"。但这两个框架都没有考虑财务困境与破产成本，且真实市场有税、有交易成本——这正是实务中对冲能增值的原因。C 反了（特有风险不被定价）；D 的说法本身就是被教材点名的错误假设（衍生品定价并不总能反映全部风险因子）。

Q4. Which of the following correctly pairs a risk limit with its main recognized weakness?

A. Stop-loss limit — it also constrains future losses.

B. Concentration limit — it does not directly address correlation risk; exposures can move together without being concentrated. ✅

C. Maturity/gap limit — it addresses price risk but not maturity smoothing.

D. Notional limit — notional exposure always tracks true risk closely.

B 正确。考点：八类限额的固有弱点
集中度限额只管"集中"，而相关性风险不需要集中也能发生——不同敞口在压力情景下可能同向恶化（教材原话：outcomes may be correlated even if they are not concentrated）。A 反了：止损限额只管已实现损失、不解决未来损失；C 反了：期限/缺口限额平滑到期压力、不解决价格风险；D 反了：名义本金与真实风险常常关系不大。

风险管理的公司治理

3. The Governance of Risk Management

学习单元 1 · 要点：董事会职责 · 危机后监管（SOX / Basel I–III / Dodd-Frank）· 三大委员会 · 风险偏好与激励 · 职能部门分工 · 审计委员会独立性

🇨🇳 中文导览

本章讲「谁对风险负责」。几乎全是定性题，最常考的是角色归属与独立性冲突：谁来设定风险偏好、谁执行缓释与转移、谁不能进审计委员会、CRO 向谁汇报、奖金怎么设计才不诱导冒险。

董事会是企业风险的最终责任人：清晰表述企业级风险偏好、决定各类风险保留/规避/缓释/转移、设立向 CEO 汇报但可直达董事会的 CRO、设立风险委员会、把薪酬委员会与风险偏好挂钩、维持独立的审计委员会；CEO 与董事长应分设（MF Global 就是反面教材）。
危机后的四条教训：利益相关方优先级由董事会裁定（债权人要下行保护、股东愿博高收益）；董事会必须主动参与并懂风险框架；风险偏好必须转成限额并传达；薪酬不得激励过度冒险——递延 + 追回条款，可配"奖金债券"（如 UBS：资本充足率跌破 7.5% 即没收）。
危机后监管三件套：SOX（2002）——CEO/CFO 亲自签署财报真实性；Basel 路线——I（1988，信用风险，资本 ≥ 风险加权资产 8%）→ II（2006）→ III（危机后 + 2016 FRTB）；Dodd-Frank（2010）——七大要素，含沃尔克规则（禁止银行自营交易）。
审计委员会的红线：内部审计师向它汇报；CFO 不得担任主席或成员（自己审自己）；它与管理层保持独立但须频繁沟通。

关键术语：corporate governance公司治理、agency risk代理风险、CRO首席风险官、clawback奖金追回、bonus bond奖金债券、audit committee审计委员会。

🧭 知识网位置第 2 章的「偏好与流程」在本章被制度化为角色与委员会；治理失败的代价在第 9–10 章案例里集中爆发。前置 → 第 2 章 · 风险偏好与五步流程本章之后 → 第 9 章 · 金融灾难（治理失灵的真实代价） · Part II · 操作风险治理

3.1 Regulation & governance after the 2007–09 crisis

这一节要背三套清单。其一监管演进：SOX（2002）——CEO 与 CFO 必须个人签署保证向 SEC 报送文件的准确性、必须披露内控缺陷与舞弊、报告流程与内控须年度审计、审计委员会成员姓名须公开且具备会计与财报理解能力（欧洲走的是"遵守或解释"的自愿模式）；Basel 路线——I/II/III 的时间与抓手（含 Basel III 五项宏观审慎叠加与 2016 FRTB）；Dodd-Frank（2010）——七大要素：强化美联储对 SIFI 的监管、终结"大而不能倒"、生前遗嘱（living will）、衍生品透明度、沃尔克规则（禁自营交易）、消费者金融保护局、双层级压力测试（DFAST / CCAR）。其二是银行业的四条治理教训：利益相关方优先级、董事会主动监督、风险偏好明确传达并转成限额、薪酬不得激励不当冒险。其三隐含在所有条文背后的一句话：治理失败是危机的共同底色——招股书式的条文记不住没关系，但要记住"每个条文治的是哪种失败"。常见陷阱：把"CEO 与董事长兼任"或"由 CRO 决定薪酬"当成正确做法。

Corporate governance 公司治理 = the processes established to run a business — shareholders, senior management, and the board of directors. Scandals (Enron 2001, WorldCom 2002, Global Crossing 2002, Parmalat 2003) produced the U.S. Sarbanes-Oxley Act (SOX, 2002):

CEO & CFO must personally certify the accuracy of SEC filings（并须保证披露完整呈现公司状况）。
一定的内控要求（如董事会与审计委员会构成）必须满足；内控缺陷（含已发现的舞弊）必须及时、准确披露。
报告流程与内控须年度审计。
审计委员会成员姓名须公开披露，且须能理解会计原则、读懂财务报表、具备审计经验。

Europe chose a different path: a voluntary "comply-or-explain" model（若不遵守最佳实践，说明理由即可）。

Crisis lessons for the banking industry

Stakeholder priority — depositors, borrowers, regulators, employees, bondholders, shareholders 的需求常常互相竞争，董事会必须裁定优先级。
Board composition — 危机没有给出"独立董事 vs 内部董事"谁更有效的清晰证据（传统逻辑被外部力量压倒）；但仍要求董事称职且独立。
Board oversight — 董事会必须主动参与风险管理，并接受教育以理解风险框架与治理基础设施的关联。
Risk appetite — 董事会清晰表述并传达风险偏好，把它翻译成企业级限额体系。
Compensation — 董事会控制薪酬机制，避免诱导不当冒险：递延奖金 + 追回条款是标准工具。

Basel accords 巴塞尔协议 — know the timeline

Version	Year	Focus
Basel I	1988	信用风险；最低资本 = 风险加权资产的 8%（拉美债务危机之后）
Basel II	2006	把交易与借贷活动纳入资本充足框架；增加监管与披露标准
Basel III	危机后	直接回应 2007–09 危机；Tier 1 资本 = 普通股 + 留存收益；LCR（覆盖 30 天现金需求）与 NSFR（一年稳定资金）；加宏观审慎叠加
Basel III + FRTB	2016	FRTB（交易账簿基础审查）：扩大市场风险敞口的覆盖，聚焦交易台引入的衍生品/期货/复杂资产风险

Basel III macroprudential overlay — 5 elements

Leverage ratio 杠杆率 — 一级资本 ÷ 总合并资产，最低要求 3%（是一条下限）。
Countercyclical capital buffer 逆周期资本缓冲。
Total loss-absorbing capacity (TLAC) — 全球系统重要性银行的最低损失吸收能力标准。
推动中央清算尽可能多的交易 → 降低对手方风险。
改进风险建模与压力测试，更好捕捉尾部风险。

Dodd-Frank Act (2010) — 七要素 ⭐

背景两句话：1933 年《格拉斯-斯蒂格尔法》隔离商业银行与投行业务（保护存款人）；1999 年《GLBA 法》拆除隔离、允许组建金融服务控股公司（FSHC）——随后的危机把投行风险传回了银行体系（贝尔斯登、美林被并购，雷曼破产）。危机后的 Dodd-Frank 正是对这套混业风险的回应：

强化美联储：对资产 > $50B 的系统重要性金融机构（SIFI）行使监管。
终结"大而不能倒"：建立有序清算机制（orderly liquidation authority）。
Resolution plan：SIFI 须提交"生前遗嘱"（living will），概述危机处置安排。
衍生品市场：提高透明度、降低对手方风险（与 Basel III 的中央清算呼应）。
沃尔克规则（Volcker Rule）：禁止银行自营交易（用银行自有资金投机），部分恢复格拉斯-斯蒂格尔的隔离精神。
消费者保护：设立 CFPB（消费者金融保护局）。
压力测试：自上而下的宏观冲击测试，覆盖信用/流动性/市场/操作等多类风险，并纳入流动性规划——资产 > $10B 参加 DFAST，> $50B 参加 CCAR。

BCBS 2015 治理最佳实践（十二项压缩清单）

董事会最终负责监督风险偏好、战略目标与治理框架的落实；
全体董事须称职（有知识有技能）；
董事会为自身运作制定政策；
高管层按董事会批准的战略执行日常经营；
集团（conglomerate）母公司董事会对全部成员单位负最终监督责任；
设置独立的风险管理职能，向董事会汇报、由 CRO 日常领导；
风险识别/监控/控制：董事会督导风险绘图，并决定保留/规避/缓释/转移；
风险沟通：把偏好与流程传达到全公司各层级；
合规：董事会最终负责监督合规风险；
内部审计：定期审计并向董事会报告进展；
薪酬：让管理层对风险决策承担财务责任；
披露：向利益相关方充分披露风险管理流程。

3.2 Best practices: board & committees

董事会的构成与分工是高频考点。最佳实践：成员以独立董事为主、具备本行业基础知识（不足者须先培训）；目标已从"服务股东"扩展为"兼顾全部利益相关方"；CEO 与董事长分设；设立 CRO 帮助董事会理解风险绘图。核心张力是代理风险（agency risk）——所有者与经营者分离带来的利益不一致；股权式激励若只挂钩短期股价，会诱导管理者隐藏风险换取奖金。三大委员会要能对号入座：风险顾问董事（独立、懂本行业风险，给董事会而非管理层提供专业意见，列席风险与审计委员会，可视为董事会与管理层之间的联络人）≠ 设偏好；风险委员会（董事会的子集，设定风险偏好、独立监控、审批高层级风险决策）≠ 管薪酬；薪酬委员会（独立于管理层，避免短期利润挂钩，用递延/追回/奖金债券）。

Board of directors — ultimate risk owner

董事会的审查清单：风险管理政策、定期风险报告、风险偏好及其对战略的影响、内控、财务报表与披露、关联交易、内外部审计报告、本行业治理最佳实践、竞争者与行业的风控做法。

Risk advisory director 风险顾问董事 — an independent board member with industry-specific risk expertise who advises the board（尤其当多数董事来自其他行业时）；列席风险与审计委员会会议，与管理层定期会面，可视为董事会与管理层之间的 liaison。
Risk management committee 风险委员会 — 董事会的子集，设定风险偏好、独立监控风险、审批高层级风险决策（如大额授信额度）；与内外部审计师保持联系以确保合规。
Compensation committee 薪酬委员会 — 独立于管理层，审批关键高管薪酬；避免短期利润/收入挂钩（易被操纵），改用递延支付、追回条款、奖金债券（UBS 例：资本充足率跌破 7.5% 即没收）。

💡 MF Global（2010）——董事会被"兼任"架空的反面教材

Jon Corzine 同时担任 MF Global 的 CEO 与董事长。他无视 CRO 的警告，大举押注欧洲主权债（正值希腊债务危机蔓延），最终酿成约 $1.6B 损失并破产。因为 CEO 与董事长是同一人，董事会没有任何力量可以对 CEO 的决策形成制衡——这正是教材要求两职分设的原因。

3.3 Risk appetite, strategy & incentives

风险偏好必须与经营战略一致：想做高收益贷款，就配套信用风险限额；想平滑经营，就用期货对冲操作与汇率风险。监督链条：董事会（经风险委员会）设定企业级偏好 → CRO 负责日常监督，可在董事会设定的容忍区间内批准临时超限，并以书面形式记入每日超限例外报告、提交风险委员会知悉；CRO 向 CEO 汇报，但职能上是董事会与管理层之间的桥梁（并坐在由 CEO、CFO、司库、CCO 与各业务单元负责人组成的风险高管团队里）。薪酬是本节题眼：只奖短期盈利、不为亏损回溯的奖金＝给管理者一份看涨期权式的收益结构（涨了拿钱、跌了不赔），会诱导隐藏风险。G20 危机后五项改革：取消多年期奖金保证、递延+追回使薪酬对称、激励上限（通常为工资 100%，股东批准可至 200%）、强化披露、保持薪酬委员会独立。

Risk appetite must be consistent with business strategy. The board (via risk committee) sets enterprise-level appetite → the CRO supervises day-to-day risk, may approve temporary limit breaches within board tolerance bands, reports to the CEO, and liaises with the board.

The dangerous compensation asymmetry (call-option payoff)

只奖短期利润、不做追回的奖金，给了管理者一个类看涨期权的收益结构——享受上涨、回避下跌。G20 危机后改革：取消多年期奖金保证；用递延 + 追回让薪酬对称；限定激励性薪酬上限（工资的 100%，股东批准可至 200%）；强化披露；保持薪酬委员会独立。

限额的必要设计（题眼）

限额的设计标准：正常业务不应触发限额，且流程中要内置 margin for error。一线员工处理超限请求（例外）；CRO 必须把例外书面化（daily risk limit exception report）并呈报风险委员会——"临时批准"可以做，但"不记录"不行。

3.4 Interdependence of functional units 职能部门相互依存

四个单元的分工必须能对号入座，这是最典型的送分/送命题：高管层 + 风险委员会设定风险偏好、设计与监督政策、按限额评估绩效；业务单元执行风险政策并及时识别例外；财务与运营实际执行风险缓释与转移交易（分析工具、参与风险与经营计划）；风险管理办公室（CRO）监控限额与内控、管理流程、向高管层与风险委员会汇报。若题目问"谁去市场上真正把对冲做出来"，答案是财务与运营；"谁监督"才是风险办公室。一线经理与员工是政策落地的关键，例外必须书面化。

Unit	Role
Senior management + risk committee	设定风险偏好、设计与监督政策、按限额评估绩效
Business units	执行风险政策；及时识别例外
Finance & operations	实际执行风险缓释与转移交易
Risk management office (CRO)	监控限额/内控、管理流程、向高管层与风险委员会沟通

3.5 Audit committee 审计委员会 ⭐

审计委员会是董事会的下设委员会，职责早已超出"财报准确性"：它还要监督风险管理流程是否被切实执行、政策是否足够。内部审计师向审计委员会汇报（而不是向 CFO 汇报），负责监控风险程序、验证合规、对 VaR 等风险指标的有效性发表意见、验证定价模型及其假设（波动率、相关性）。两个必考点：① 成员必须独立于所审计的日常业务，且具备会计与财报知识（GAAP/IFRS）；② CFO 不能担任主席也不能担任成员——被审计对象去审自己，典型的利益冲突，几乎每年都考。正确表述是"独立但与管理层保持频繁沟通"。

Ensures accuracy of financial statements & regulatory reporting; also oversees the risk-management process.
Internal auditors report to the audit committee — 监控风险程序、跟踪系统进展、验证合规、对风险指标（如 VaR）与定价模型假设（波动率、相关性）发表意见。
Must be independent of day-to-day risk implementation; members need financial knowledge (GAAP/IFRS)。独立但须与管理层协作、频繁沟通，确保问题被解决。

Exam trap — who's NOT on the audit committee?

The CFO must NOT chair (or sit on) the audit committee — that's a conflict of interest（他准备的财报正是被审计对象）。Internal auditors report to the audit committee, not the CFO.

🔑 英文要点速查 · English Essentials

corporate governance 公司治理 — The processes established to run a business: shareholders, senior management, board of directors.
agency risk 代理风险 — Risk from the separation of owners and operators; managers may take short-term risks for personal gain.
risk advisory director 风险顾问董事 — Independent director with industry-specific risk expertise; advises the board, attends risk/audit committees, liaises with management.
risk management committee 风险委员会 — Board subset that sets risk appetite and independently monitors risk; approves high-level risk decisions.
compensation committee 薪酬委员会 — Independent of management; aligns pay with long-term risk via deferral, clawbacks, bonus bonds.
clawback provision 奖金追回条款 — Managers repay bonuses when long-term results disappoint.
bonus bond 奖金债券 — Pays only if thresholds are met (UBS: forfeited if capital ratio falls below 7.5%).
CRO 首席风险官 — Day-to-day risk supervision; reports to the CEO, full access to the board; may approve temporary limit breaches within tolerance bands.
audit committee 审计委员会 — Independent of management; internal auditors report to it; CFO cannot sit on it; oversees statements, controls AND the risk process.
Sarbanes-Oxley Act (SOX, 2002) 萨班斯法 — CEO/CFO certify filings; controls audited annually; audit committee members financially literate.
Basel I/II/III 巴塞尔协议 — 1988 credit risk (8% of RWA) → 2006 trading/lending → post-crisis Tier 1, LCR/NSFR, macroprudential overlay; FRTB added 2016.
Dodd-Frank Act (2010) 多德-弗兰克法 — SIFI oversight, orderly liquidation, living wills, derivatives transparency, Volcker Rule (no proprietary trading), CFPB, DFAST/CCAR stress tests.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. A bank's board is revising its risk governance after a regulatory review. Which action is MOST consistent with best practices for board-level risk governance?

A. Combine the CEO and chairperson roles so that decisions can be made faster.

B. Establish a CRO who reports to the CEO but retains full access to the board. ✅

C. Let the risk management office set the enterprise-level risk appetite, subject to board ratification.

D. Link the compensation committee's bonus pools primarily to annual revenue growth.

B 正确。考点：董事会六项职责 + 独立性
CRO 向 CEO 汇报、但可直达董事会，是教材列出的标准做法之一。A 错——CEO 与董事长应分设（MF Global 案例的教训）；C 错——风险偏好由董事会（经风险委员会）设定，不是风险办公室"设定后追认"；D 错——薪酬不能只挂短期收入，应递延 + 追回、与长期风险一致。

Q2. Which statement about the audit committee is MOST accurate?

A. The CFO should chair it, since no one understands the financial statements better.

B. Its remit covers the firm's risk-management process — internal auditors report to it, it opines on risk metrics and validates pricing-model assumptions. ✅

C. It should operate strictly independently of management and avoid contact with executives.

D. Industry expertise fully substitutes for financial knowledge among its members.

B 正确。考点：审计委员会议程范围
审计委员会的职责早已超出财报准确性：监督风险流程、验证合规、对 VaR 等指标发表意见、验证定价模型与假设（波动率、相关性）都归它。A 错——CFO 不得进入审计委员会（利益冲突）；C 错——应"独立但频繁沟通"；D 错——成员须具备会计与财报知识，行业经验不能替代。

Q3. A bank's business unit wants to exceed an approved credit-risk limit to book a profitable loan. Per the governance process, which outcome is most appropriate?

A. The business unit may proceed so long as the expected profit exceeds the limit's threshold.

B. The CRO may approve a temporary breach within the board's tolerance bands, documented in the daily limit-exception report for the risk committee. ✅

C. Any breach requires prior approval from the full board, with no delegated authority.

D. Breaches need no documentation if they are later cured within the same quarter.

B 正确。考点：CRO 的超限处置权与书面化要求
CRO 的日常职责之一就是在董事会容忍区间内批准临时超限，并确保例外以书面形式记录（每日超限例外报告）呈报风险委员会。A 错——限额不是利润门槛，盈亏不能改变限额纪律；C 错——全部上行到董事会会把决策瘫痪（这才是 CRO 授权的意义）；D 错——"事后弥补"不能免除记录与报告义务。

Q4. Which of the following elements belongs to the Dodd-Frank Act (2010) rather than to Basel III or SOX?

A. Minimum capital of 8% of risk-weighted assets.

B. A rule prohibiting banks from engaging in proprietary trading. ✅

C. A liquidity coverage ratio funding 30 days of cash needs.

D. A "comply-or-explain" corporate-governance regime.

B 正确。考点：监管三件套的归属（SOX / Basel / Dodd-Frank）
禁自营交易的是 Dodd-Frank 的沃尔克规则（部分恢复格拉斯-斯蒂格尔式的隔离）。A 是 Basel I（1988）；C 是 Basel III 的 LCR；D 是欧洲的治理模式（自愿遵守或解释），都不是 Dodd-Frank。

信用风险转移机制

4. Credit Risk Transfer Mechanisms

学习单元 1 · 要点：CDS / CDO / CLO 的分工 · 传统缓释六件套 · 危机链条与监管回应（Volcker / Section 15G）· 证券化四步 · OTD 激励

🇨🇳 中文导览

本章讲银行如何把信用风险「搬出」资产负债表。考试重点是三类信用衍生品的分工、传统缓释手段的判别（尤其是哪些不属于信用风险转移）、以及证券化的四步流程与两种业务模式的激励差异。

CDS信用违约互换：买方按季付保费、卖方在参考实体违约时赔付。优点——促进创新（买方敢做更高风险的事）、给卖方现金流、以及对信用风险做纯价格发现（债券价格里还混着利率风险）；缺点——监管薄弱带来的对手方风险、以及买方的「虚假安全感」。
CDO债务抵押证券：把贷款重组分档出售，顺序为超优先档（常为 AAA，先受偿、收益最低）、中间档（mezzanine）、股权档/毒性垃圾（后受偿、先承担损失、收益最高）；由其他 CDO 的档位再打包即 CDO²，其复杂性主要是为了方便销售，而非改善风险。CLO 以经过严格承销的银行贷款为底层，危机中表现远好于偏房贷的 CDO。
传统缓释：第三方保险/担保、逐日盯市、敞口净额结算、要求抵押品（注意错向风险：抵押品价值与借款人违约同向恶化）、终止条款、再指定、银团贷款（牵头行通常自留约 20%；分 firm commitment（兜底）与 best-efforts（尽力）两种）。赎回条款（call feature）只对冲利率风险，不转移信用风险——经典干扰项。
证券化四步：设立 SPV → SPV 借款向银行收购贷款 → 按优先级/评级分档 → 向投资者出售。买入持有模式激励贷款质量；发起—分销（OTD）模式激励贷款规模，并因表外处理而降低资本占用。

关键术语：tranche分档/层级、SPV特殊目的载体、risk retention风险留存、exposure netting敞口净额结算、syndication银团贷款、wrong-way risk错向风险。

🧭 知识网位置第 2 章「转移」策略的深水区：把信用风险单独剥离、定价、再分配。前置 → 第 2 章 · 四策略与对冲本章之后 → 第 9 章 · 金融灾难（信用衍生品教训） · Part II · 信用衍生品

4.1 Credit derivatives: the off-balance-sheet toolbox

信用衍生品是表外工具，作用是把信用风险单独剥离出来转移（Greenspan 曾指出美国银行体系正是靠这些工具扛过了 2001–02 的放缓）。CDS 要记住三优两缺：优——促进创新（买方受保护、敢支持更高风险项目）、为卖方提供可跨行业跨地域分散的现金流、提供纯粹的信用风险价格发现（债券价格里混着利率风险，这一点是 CDS 的独特优势）；缺——危机前监管薄弱导致对手方风险（卖方可能赔不起）、以及「虚假安全感」使买方去支持本不该支持的高风险发行人。CDO 的档位顺序必考：超优先档（先受偿、最后承担损失、收益最低、常被评为 AAA）→ 中间档 → 股权档（最后受偿、先承担损失、收益最高，俗称"毒性垃圾"）。CLO 与 CDO 结构相同但底层是经严格承销的银行贷款，因此危机后仍受欢迎。

Credit derivatives 信用衍生品 are off-balance-sheet instruments that isolate and transfer credit risk (the risk a borrower defaults). The big three: CDS, CDO, CLO.

Credit Default Swap (CDS) 信用违约互换 ⭐

Like an insurance policy on a bond — the buyer pays regular (quarterly) premiums; the seller pays out if the reference issuer defaults.

Advantages ✅	Disadvantages ❌
Spurs innovation (buyers shelter credit risk → fund riskier ventures)	Historically weak regulation → counterparty risk (seller may not pay)
Cash-flow source for sellers (diversify across CDS)	False sense of security → buyers fund riskier issuers
Pure price discovery of credit risk (bonds mix in interest-rate risk)	—

Collateralized Debt Obligation (CDO) 债务抵押证券/担保债务凭证

Repackaged loans (heavily mortgage-biased → MBS; mortgage-only = CMO) split into tranches 分档 by seniority/rating:

Super-senior tranche (often AAA): paid first, safest, lowest yield.
Mezzanine tranches — middle risk/return.
Equity/junior tranche (a.k.a. "toxic waste"): paid last, highest yield.

CDO-squared = a CDO containing repackaged pieces of other CDOs — added complexity for marketing, not better risk.

Collateralized Loan Obligation (CLO)

Like a CDO but backed by bank loans subject to rigorous underwriting. CLOs weathered the GFC better than mortgage-heavy CDOs and stayed popular afterwards.

4.2 Traditional credit-risk mitigation (non-derivative)

本节最常考的是「以下哪项不能转移信用风险」。能转移/缓释的：第三方保险或担保（单一借款人投保时技术上称为 guarantee，如市政债保险）、敞口净额结算、逐日盯市（定期重估并立即划付，主要与场内衍生品配合，避免到期一次性大额支付风险）、要求抵押品、终止条款（触发事件如评级下调、财务指标不达标）、再指定（触发后自动转给第三方）、以及银团贷款。银团的两个细节：牵头行一般自留约 20%。firm commitment 模式 = 牵头行向发行人兜底全额贷款（找不到参团行就自己全扛）；best-efforts 模式 = 尽力而为，找不到就少借——发行人可能拿不满。注意银团不能定向转移某类特定风险（保留的部分风险仍在账上，需配合信用衍生品）。不能转移信用风险的典型是赎回条款：它让发行人提前赎债，对冲的是利率风险。抵押品那条还要记「错向风险」——抵押品价值与借款人违约由同一因素驱动而同时恶化（油企以原油为抵押，油价跌时既经营困难又抵押贬值）。

Mechanisms to transfer/mitigate credit risk

Third-party insurance / guarantee 第三方担保 — covers a single borrower (technically a guarantee) or a group; e.g., municipal bond insurers.
Marking-to-market 盯市 — revalue and settle gains/losses daily; mainly with exchange-traded derivatives.
Exposure netting 敞口轧差/净额结算 — offset gains against losses with the same counterparty.
Collateral 抵押品 — watch wrong-way risk: collateral value falls together with the borrower's ability to pay.
Termination / reassignment clauses 终止/再指定条款 — trigger events (downgrade, missed financial metrics) end the position or move it to a third party.
Loan syndication 银团贷款 — lead bank retains ~20%; firm commitment vs best efforts.

Exam note

A call feature protects the issuer from interest-rate risk, not credit risk — a classic distractor.

4.3 Credit derivatives in the GFC & post-crisis changes

危机链条要能复述：2007 年 CDS 名义本金膨胀到约 45 万亿美元，超过美国股票、国债与按揭余额之和（其中一个推手是投资者对并不持有的资产买 CDS 来押注违约），而提供流动性的对手方过少 → 系统性集中度风险；雷曼破产时，其 6000 亿美元存续债务中约 4000 亿由 CDS 覆盖，卖方（如 AIG 与 Citadel）因所有合约同时到期而濒临倒闭——这本质是一次相关性事件；同时可调利率按揭（ARM）进入重定价，美联储 2004–2006 年加息叠加房价下跌，违约潮沿 MBS → CDO → CDO² 传导。结论必须记住：问题在于信用衍生品被滥用，而不是它们存在本身。监管回应：Dodd-Frank 的沃尔克规则禁止吸存银行自营交易及投资 CDS，并由 CFTC 监管全部互换；SEC Section 15G（2014）要求证券化产品发起人至少自留 5% 的信用风险，且不得转移或缓释这部分敞口（迫使发起人对产品负责）。

The chain reaction (2008)

CDS notional hit $45 trillion in 2007 — more than U.S. equities + Treasuries + mortgages combined.
Too few liquidity providers were counterparties → systemic concentration risk.
Lehman failed: ~$400bn of its $600bn debt was covered by CDS; sellers (e.g., AIG and Citadel) nearly collapsed when all CDS came due at once (a correlation event).
Adjustable-rate mortgages hit resets while the Fed hiked rates (2004–06) and home prices fell → mass defaults rippled through MBS → CDO → CDO-squared.

Key lesson: the problem was misuse/abuse of credit derivatives, not their existence.

Post-crisis regulatory changes

Dodd-Frank (2010) — the Volcker rule bans deposit-taking banks from proprietary trading & investing in CDSs; CFTC now regulates all swaps.
SEC Section 15G (2014) — originators of securitized products must retain ≥5% of credit risk on balance sheet (cannot transfer/mitigate it) → aligns incentives (risk retention 风险留存).

4.4 Securitization, SPVs & bank business models

证券化的四步必须按顺序背：① 设立 SPV（表外法律实体，实际上是发起行的"半隐藏"子公司，对投资者不透明）→ ② SPV 借入资金向一家或多家银行收购贷款，形成 CMO/CDO/CLO → ③ 按优先级或评级把资产池分档形成风险层 → ④ 把各档卖给二级市场投资者。两种业务模式的激励差异是题眼：买入持有把贷款留在账上赚利息，因此关心贷款质量；发起—分销（OTD）只为证券化卖出而放贷、赚的是手续费，因此关心贷款规模——带来道德风险、激励错配与不透明。OTD 的吸引力：降低短期盈利波动、优化资本（表外处理减轻巴塞尔资本占用）、把信用与利率风险分散给更多参与者、投资者获得更多品种、借款人也更容易更便宜地融资。风险点是：投资者一旦撤离，银行手上积压的贷款库存就会把信用风险重新砸回自己的资产负债表；有些银行还会把留存的产品装进高杠杆表外载体 SIV（结构投资载体，赚短长利差）——危机中随底层违约率飙升而反噬。另可留意里程碑年份：首只 MBS（1970）、首只结构化 CMO（1983，Freddie Mac）、首只车贷 ABS（1985，Chrysler）、首只 CDO 与信用卡 ABS（1987）、CLO（90 年代初）。

The 4-step securitization process 证券化流程

Create a Special Purpose Vehicle (SPV) 特殊目的载体 — an off-balance-sheet legal entity, a "semi-hidden" subsidiary.
SPV borrows funds to buy loans from one or more banks to form structured products (CMO/CDO/CLO).
Tranche the loans by seniority/rating into risk layers.
Sell the tranches to investors in secondary markets.

Model	Mechanism	Incentive
Buy-and-hold 买入持有	Source loans, keep on the books, earn interest	Loan quality (you keep credit risk)
Originate-to-Distribute (OTD) 发起-分销	Source loans solely to securitize & sell; earn fees not interest	Loan volume, not quality ⚠️

OTD's off-balance-sheet treatment also eased Basel capital requirements — a driver of its popularity. When investors fled CDOs, banks were left holding loan inventory they couldn't sell (see Lehman, Bear Stearns).

History to know: first MBS 1970; first structured CMO 1983 (Freddie Mac); first auto-loan ABS 1985 (Chrysler); first CDO & credit-card ABS 1987; CLOs early 1990s.

🔑 英文要点速查 · English Essentials

credit default swap (CDS) 信用违约互换 — Insurance-like: quarterly premiums; seller pays on the reference issuer's default. Pure price discovery of credit risk.
collateralized debt obligation (CDO) 债务抵押证券 — Repackaged loans sold as tranches; mortgage-biased products were crisis epicenters.
collateralized loan obligation (CLO) 贷款抵押证券 — Like a CDO but backed by rigorously underwritten bank loans; fared better in the GFC.
tranche / waterfall 分档/瀑布 — Super-senior (paid first, AAA) → mezzanine → equity ("toxic waste"; first to absorb losses, highest yield).
special purpose vehicle (SPV) 特殊目的载体 — Off-balance-sheet entity that buys loans and issues tranches; a "semi-hidden" subsidiary.
wrong-way risk 错向风险 — Collateral value deteriorates exactly when the borrower becomes likely to default (oil collateral vs an energy firm).
exposure netting 敞口净额结算 — Offsetting gains and losses with the same counterparty.
syndication 银团贷款 — Lead bank keeps ~20%, spreads the rest; firm commitment (lead guarantees the full amount) vs best efforts (no guarantee).
call feature 赎回条款 — Lets the issuer retire bonds early; addresses interest-rate risk, not credit risk.
originate-to-distribute (OTD) 发起—分销模式 — Source-to-securitize; fees not interest; rewards loan volume → moral hazard, misaligned incentives, opaqueness.
buy-and-hold 买入持有模式 — Keep loans on the books; rewards loan quality.
Section 15G (SEC, 2014) 15G 条款 — Originators must retain ≥5% of credit risk, unhedged, on balance sheet (risk retention).

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. In a CDO's cash-flow waterfall, which tranche is paid first (and therefore absorbs losses last)?

A. The equity (junior) tranche.

B. The mezzanine tranche.

C. The super-senior tranche. ✅

D. The tranche with the lowest credit rating.

C 正确。考点：CDO 分档瀑布顺序
瀑布顺序：超优先档先受偿、最后承担损失（常评 AAA、收益率最低）→ 中间档 → 股权档最后受偿、最先吸收损失（收益率最高，俗称 toxic waste）。D 恰好说反：评级最低的档最先承担损失、最后拿钱。

Q2. Which of the following does NOT transfer or mitigate a lender's credit risk?

A. Exposure netting with the borrower.

B. A termination clause triggered by the borrower's downgrade.

C. Writing a call feature into the bond the borrower issues. ✅

D. Loan syndication.

C 正确（它不转移信用风险）。考点：传统缓释手段判别 + 赎回条款陷阱
call feature 让发行人可以提前赎回债券，针对的是利率风险（利率下行时再融资）——与信用风险无关，这是本章最经典的干扰项。A（净额结算）、B（终止条款）、D（银团）都是教材列明的信用风险缓释/转移机制。

Q3. A lead bank arranges a $500 million syndicated loan on a best-efforts basis. If it cannot find enough participating banks, what happens?

A. The lead bank must fund the entire $500 million itself.

B. The borrower may receive less loan proceeds than desired. ✅

C. The loan automatically converts into a CDO structure.

D. The lead bank retains only its usual 20% and the remainder is government-guaranteed.

B 正确。考点：firm commitment vs best-efforts
best-efforts＝牵头行尽力组团，但不对全额兜底——组不满，发行人拿不到全部资金。A 描述的是 firm commitment（牵头行兜底全额，卖不掉的部分自己扛）；C、D 都是教材里不存在的机制（银团也不会自动变 CDO，政府担保更非标准操作）。

Q4. Which statement about credit derivatives and the 2007–2009 crisis is MOST accurate?

A. The mere existence of CDS contracts caused the crisis.

B. After the crisis, originators must retain 5% of securitization credit risk but are free to hedge it away.

C. The crisis stemmed from misuse and concentration of credit risk transfer, and post-crisis rules responded with the Volcker rule and Section 15G risk retention. ✅

D. CDS notional value peaked at roughly $45 billion.

C 正确。考点：危机的"滥用而非存在"论 + 监管回应
教材结论原话：问题是信用衍生品被滥用（misuse/abuse）而非存在本身。A 与 D 均为事实错误（名义本金峰值约 45 万亿美元，不是 45 亿）；B 错在后半句——Section 15G 要求发起人自留 ≥5% 且不得转移或缓释该部分敞口，正是为了对齐激励。

现代投资组合理论与 CAPM ⭐（Part I 最高频计算点之一）

5. Modern Portfolio Theory (MPT) & the Capital Asset Pricing Model (CAPM)

学习单元 2 · 要点：MPT 与有效前沿 · CML vs SML · β 的两种算法 · CAPM 与高估/低估判断 · 六大绩效指标（Sharpe / Treynor / Jensen α / 跟踪误差 / 信息比率 / Sortino）

🇨🇳 中文导览

本章是整个模块计算量最大的一章，必须做到「公式能默写、指标会选、图形不混淆」。三条主线：

MPT：收益正态、投资者理性且厌恶风险、市场完美 → 投资者只在均值与方差之间权衡。组合收益是各资产收益的加权平均，组合方差取决于相关系数 ρ：ρ = +1 完全没有分散化效果，ρ < 1 方差下降，ρ = −1 可构造零方差组合。充分分散后只剩系统性风险，只有它才有回报补偿。有效前沿上每一点都是"给定风险下收益最高"；其最左端是全局最小方差组合；前沿下方为无效率、上方不可达。
CML vs SML：CML 横轴是总风险 σ、只适用于有效组合、斜率是市场夏普比率；SML 横轴是系统性风险 β、适用于任意单个证券或组合、斜率是市场风险溢价 MRP。给单个证券定要求回报率、判断高估低估，一律用 SML（即 CAPM）。
绩效指标：夏普 = (R_P−R_F)/σ_P（总风险，适用任何组合）；特雷诺 = (R_P−R_F)/β_P（系统性风险，适用充分分散组合）；詹森 α = R_P − [R_F + (E(R_M)−R_F)β_P]；信息比率 = (R_P−R_B)/跟踪误差；索提诺 = (R_P−R_MIN)/下行标准差。

关键术语：efficient frontier有效前沿、market portfolio市场组合、beta贝塔/系统性风险系数、market risk premium市场风险溢价、Jensen's alpha詹森阿尔法。

🧭 知识网位置第 1 章「风险与回报」的定量化：把「承担系统性风险获得补偿」变成了可计算的一章。前置 → 第 1 章 · risk & reward 本章之后 → 第 6 章 · APT 与多因子（对 CAPM 的扩展） · Part I 定量 · 线性回归（β 的估计方法） · Part II · 因子理论

5.1 Modern Portfolio Theory & the efficient frontier

马科维茨的三条假设是：收益服从正态分布（因此只需均值与方差，忽略偏度峰度）、投资者理性且厌恶风险（同等风险下选预期收益最高的）、资本市场完美（无税无佣金、信息免费、完全竞争）。由此得到分散化的量化结论：组合收益是加权平均，但组合方差取决于相关系数——ρ = +1 时方差就是各方差的加权平均、毫无分散化收益；ρ < 1 时方差低于加权平均且越低越好；ρ = −1 时可以构造出零方差组合（合成无风险资产）。持有足够多的资产可消除特质性/非系统性风险，只剩系统性风险，而只有系统性风险才获得回报补偿（因为分散化几乎免费）。有效前沿 = 每一风险水平上收益最高的组合集合；其下方无效率、上方不可达，最左端是全局最小方差组合；投资者按相对风险厌恶在前沿上选点（更保守的选低风险端点附近，更激进的往高风险端走）。

Markowitz's MPT assumptions

Returns are normally distributed → investors only care about mean & variance (mean-variance framework).
Investors are rational & risk-averse — at equal risk they pick the highest expected return.
Capital markets are perfect — no taxes/commissions, free information, perfect competition.

Diversification is the whole game

Portfolio return = weighted average of asset returns. Portfolio variance depends on correlations ρ between assets:

ρ = +1 → no diversification benefit (variance = weighted average).
ρ < 1 → diversification works, variance falls.
ρ = −1 → can build a zero-variance (synthetic risk-free) portfolio.

Holding enough assets eliminates idiosyncratic (company-specific) risk 特质性风险/非系统性风险 — leaving only systematic (market) risk 系统性风险，and that is the only risk you're compensated for.

Efficient frontier — intuition The set of portfolios that maximise return for each level of total risk.
• Below the frontier = inefficient (dominated). · Above = unattainable.
• Global minimum variance portfolio = leftmost point of the efficient frontier.

5.2 The Capital Market Line (CML)

引入无风险资产（通常以国库券为代表）后，投资者把它与某个有效组合搭配，得到一条从 R_F 出发、与有效前沿相切的直线；在同质预期假设下切点唯一，该切点组合就是市场组合（包含全部风险资产、按市值加权，实务上用标普 500 等指数替代），这条线即 CML。公式是 E(R_P) = R_F + [(E(R_M) − R_F)/σ_M] × σ_P，斜率等于市场夏普比率；M 点左侧是"以 R_F 出借"（部分资金买国债），右侧是"以 R_F 借入"（加杠杆放大市场组合敞口）。与 SML 的对照必须背熟：CML 用总风险、只对有效组合成立、用于衡量总风险；SML 用 β、对任何证券或组合都成立、用于定价与判断高估低估。给单只股票定要求回报率要用 SML，不是 CML。

Add a risk-free asset (e.g., T-bill). Combining R_F with a tangency efficient portfolio gives the CML — a line tangent to the efficient frontier from R_F. With homogeneous expectations, there's ONE tangency point = the market portfolio 市场组合 (all risky assets, market-cap weighted; proxied by e.g. the S&P 500).

Capital Market Line ⭐ E(R_P) = R_F + [ (E(R_M) − R_F) / σ_M ] · σ_P
Slope = the Sharpe ratio of the market. Left of M = lending at R_F; right of M = borrowing (leverage).

CML vs SML — the classic exam distinction

	CML 资本市场线	SML 证券市场线
X-axis	Total risk (σ)	Systematic risk (beta)
Applies to	Efficient portfolios only	Any individual security or portfolio
Slope	Sharpe ratio of market	Market risk premium (MRP) = E(R_M) − R_F
Use	Total-risk measurement	Pricing/required return (CAPM); detect mispricing

判例口径：说"CML 可给单个证券定要求回报率"的是错的——那要用 SML；说"CML 假定投资者持有市场组合 + 无风险资产"的是对的。

5.3 Beta — systematic risk 贝塔/系统性风险系数 ⭐

β 衡量的是资产收益对市场收益的敏感度，有两个等价写法：β_i = Cov(i,M) / σ_M² = ρ_i,M × (σ_i / σ_M)。三个必记性质：市场自身的 β 恒等于 1（因为 Cov(M,M) = σ_M²）；β > 1 为周期型（如奢侈品，扩张期跑赢）；β < 1 为防御型（如公用事业，衰退期相对抗跌）。实务中 β 用超额收益对市场超额收益的最小二乘回归估计，回归直线的斜率就是 β。做题时注意两种给法：给 ρ 与两个标准差就用 ρ×(σ_i/σ_M)；给协方差就用 Cov/σ_M²（分母是市场方差，别忘了平方）。例题口径：σ_M = 20%、σ_i = 30%、ρ = 0.8 → β = 1.2；Cov = 0.048 → β = 0.048 / 0.04 = 1.2。

Beta (two equivalent forms) — MEMORISE ⭐ β_i = Cov(i, M) / σ_M² = ρ_i,M (σ_i / σ_M)

• β_market = 1 by definition. β > 1 → cyclical (e.g., luxury goods); β < 1 → defensive (e.g., utilities). In practice beta = slope of regressing (excess) stock returns on (excess) market returns.

Worked example (from the notes)

Market σ = 20%. (a) Asset σ = 30%, ρ = 0.8 → β = 0.8 × (0.30/0.20) = 1.2. (b) Cov = 0.048 → β = 0.048 / 0.20² = 0.048/0.04 = 1.2。

5.4 The CAPM & the Security Market Line 资本资产定价模型

CAPM 的核心结论是：预期收益只取决于 β（公司特有风险已被分散掉），且是 β 的线性函数。推导套路的两个支点：① 只有系统性风险被定价；② 收益与 β 呈线性关系。SML 就是这条直线的图形：截距为 R_F（β = 0 处），斜率为市场风险溢价 MRP = E(R_M) − R_F（由两点确定：(0, R_F) 与 (1, R_M)）。公式 E(R_i) = R_F + [E(R_M) − R_F] × β_i 算出来的是要求回报率/最低门槛收益率（hurdle rate）：门槛定高了会错过好项目，定低了会买入被高估的资产。判价规则：分析师预测收益 > 要求回报 → 被低估 → 落在 SML 上方 → 买入；反之被高估 → 落在 SML 下方 → 回避或做空。假设条件要能识别（常见正确选项）：信息免费、无摩擦（无税无佣金）、资产可无限细分、完全竞争（价格接受者）、只按预期收益与方差决策、可按 R_F 无限借入与贷出、同质预期（对未来收益/方差/协方差的预测一致、单期）。注意：CAPM 假设无任何税收与交易成本，"只有资本利得税""交易成本恒定"都是错误选项。

CAPM assumptions

Information free & available; frictionless markets (no taxes/commissions); assets infinitely divisible; perfect competition (price takers); decisions based only on expected return & variance; unlimited borrow/lend at R_F; homogeneous expectations (same forecasts for one period).

CAPM — THE equation ⭐⭐ E(R_i) = R_F + [E(R_M) − R_F] · β_i = R_F + MRP · β_i
The SML is the graphical picture of the CAPM: intercept = R_F, slope = MRP.

💡 Applying CAPM — Sky-Air

MRP = 5%, R_F = 4%, β = 1.5 → E(R) = 0.04 + 1.5(0.05) = 11.5% (the hurdle rate 最低要求回报率).

If you forecast return > 11.5% → stock is undervalued → plots above the SML → buy.
If forecast < 11.5% → stock is overvalued → plots below the SML → don't buy / short.

5.5 Risk-adjusted performance measures 风险调整后绩效指标 ⭐

六个指标的分子和分母要分毫不差。夏普与特雷诺分子相同（组合收益 − R_F），区别只在分母：夏普除以总风险 σ（适用任何组合，最通用），特雷诺除以系统性风险 β（假设已充分分散）。由此得到经典结论：特雷诺高于市场、夏普低于市场 ⇒ 组合分散不足——它背着额外的特质性风险，把夏普的分母撑大了。詹森 α = 实际收益 − CAPM 要求收益，正值代表跑赢（被低估），且只适合在相同系统性风险水平下横向比较；特雷诺与詹森同向（特雷诺优于市场 ⇔ α 为正），但两个指标给出的排名可能不同。信息比率分子是相对基准的超额收益（不是相对 R_F，R_F 是干扰项），分母是跟踪误差（组合收益与基准收益之差的标准差）。索提诺分子用最低可接受收益 R_MIN（不是 R_F），分母是下行标准差（只统计低于 R_MIN 的波动，高于 R_MIN 的不算"风险"）。

Measure	Formula	Risk metric	Interpretation
Sharpe (SPI)	(R_P−R_F) / σ_P	Total risk (σ)	Return per unit of total risk; use for any portfolio
Treynor (TPI)	(R_P−R_F) / β_P	Systematic (β)	Return per unit of market risk; for well-diversified portfolios
Jensen's alpha	R_P − [R_F + (E(R_M)−R_F)β_P]	β (via CAPM)	Actual return above the CAPM required return; +α = undervalued
Tracking error (TE)	σ(R_P − R_B)	Std dev of active return	Variability of return vs benchmark
Information ratio (IR)	E(R_P−R_B) / TE	Active risk	Active return per unit of active risk
Sortino	(R_P−R_MIN) / downside dev.	Downside deviation	Only penalises downside volatility

Sharpe & Treynor share the SAME numerator (excess return over R_F); only the denominator differs. Jensen & Treynor assume diversification (use β). If Treynor > market but Sharpe < market → portfolio is NOT well-diversified (it carries extra idiosyncratic σ that penalises Sharpe; a classic question).

💡 Worked example — all three classic measures

Portfolio: E(R_P) = 14%, σ_P = 25%, β_P = 1.1. Market: E(R_M) = 12.5%, σ_M = 20.2%. R_F = 2.6%.

Sharpe_P = (0.14−0.026)/0.25 = 0.456; Sharpe_M = (0.125−0.026)/0.202 = 0.49 → portfolio underperforms on total risk.
Treynor_P = 0.114/1.1 = 0.1036; Treynor_M = 0.099 → portfolio outperforms on market risk.
Jensen α_P = 0.14 − [0.026 + 0.099(1.1)] = 0.14 − 0.1349 = 0.0051 (+0.51%).

💡 Worked example — IR & Sortino

Active portfolio E(R)=15% vs FTSE 100 = 12%, tracking error = 9%, R_MIN = 4%, downside dev = 7%.

IR = (0.15−0.12)/0.09 = 0.33. · Sortino = (0.15−0.04)/0.07 = 1.57.

🔑 英文要点速查 · English Essentials

modern portfolio theory (MPT) 现代投资组合理论 — Mean-variance framework: normal returns, rational risk-averse investors, perfect markets.
efficient frontier 有效前沿 — Max return for each risk level; below = inefficient, above = unattainable.
global minimum variance portfolio 全局最小方差组合 — Leftmost point of the efficient frontier.
market portfolio 市场组合 — All risky assets, market-cap weighted; the tangency portfolio under homogeneous expectations.
capital market line (CML) 资本市场线 — R_F + [(E(R_M)−R_F)/σ_M]·σ_P; x-axis = total risk; efficient portfolios only; slope = market Sharpe.
security market line (SML) 证券市场线 — Graphical CAPM; x-axis = β; applies to all securities; slope = MRP.
beta 贝塔 — β = Cov(i,M)/σ_M² = ρ·(σ_i/σ_M); market β = 1; >1 cyclical, <1 defensive.
market risk premium (MRP) 市场风险溢价 — E(R_M) − R_F; the slope of the SML.
hurdle rate 最低要求回报率 — The CAPM required return; forecast above it → undervalued (plots above SML).
Sharpe index 夏普指数 — Excess return ÷ total risk (σ); applicable to any portfolio.
Treynor index 特雷诺指数 — Excess return ÷ β; for well-diversified portfolios. T > market but S < market ⇒ poor diversification.
Jensen's alpha 詹森阿尔法 — R_P − CAPM required return; compare at the same β.
information ratio 信息比率 — Active return (vs benchmark) ÷ tracking error (active risk).
Sortino ratio 索提诺比率 — (R_P − R_MIN) ÷ downside deviation; penalises only downside variability.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文；本章为最高频计算点，共 5 题——每道都要亲手算一遍。

Q1. Which of the following is MOST likely an assumption of the capital asset pricing model (CAPM)?

A. Investors pay taxes only on capital gains.

B. Market participants can lend and borrow unlimited amounts at the risk-free rate. ✅

C. An individual investor's trading activity can move market prices.

D. Investors price higher moments of return (skewness and kurtosis).

B 正确。考点：CAPM 假设清单
CAPM 的假设包括：信息免费、无税无佣金（任何税都不存在——A 错）、资产可无限细分、完全竞争的价格接受者（C 反了）、只按预期收益与方差决策（不理会偏度峰度——D 反了）、可按 R_F 无限借贷（B 对）、同质预期。

Q2. The risk-free rate is 3%, the market risk premium is 7%, and stock J's beta is 1.4. An analyst forecasts J's return at 11%. J is most likely:

A. Undervalued, with a required return of 12.8%.

B. Overvalued, with a required return of 12.8%. ✅

C. Overvalued, with a required return of 9.8%.

D. Fairly valued, with a required return of 11.0%.

B 正确。考点：CAPM 计算 + 高估/低估判断
要求回报 = 3% + 1.4×7% = 12.8%。预测收益 11% 低于要求回报 ⇒ 股票被高估（落在 SML 下方）。C 是漏乘 β 的算法（3+7=10→9.8 系 3%+1.4×7% 误算），D 把预测值当成了要求值——估值判断的前提是"要求值 vs 预测值"的比较。

Q3. The market's standard deviation is 25%. Stock K has a standard deviation of 40% and a correlation with the market of 0.5. Stock K's beta is closest to:

A. 0.5

B. 0.8 ✅

C. 1.25

D. 2.0

B 正确。考点：β 的两种算法
β = ρ × σ_K/σ_M = 0.5 × (40%/25%) = 0.8。校验另一口径：Cov = ρ·σ_K·σ_M = 0.5×0.4×0.25 = 0.05；β = 0.05/0.25² = 0.05/0.0625 = 0.8 ✓。A 漏乘了 σ 比值、D 把相关与 β 混为一谈。

Q4. Portfolio Z shows a Treynor measure above the market's, but a Sharpe measure below the market's. Z is most likely:

A. Not well diversified — it carries idiosyncratic risk that inflates its σ. ✅

B. Generating a negative CAPM alpha.

C. Using leverage to amplify the market portfolio.

D. Superior to the market on both total-risk and market-risk bases.

A 正确。考点：Sharpe vs Treynor 的分散化诊断
特雷诺（除以 β）高、夏普（除以 σ）低，说明分母里的特质性风险把 σ 撑大了——组合分散不足，而其系统性风险表现其实不错。B 与结论矛盾（特雷诺优于市场 ⇒ α 为正）；C 无关——杠杆不改变风险调整后的排序；D 明显错误（夏普低于市场）。

Q5. A portfolio has an expected return of 13%, beta of 1.2. The market's expected return is 10% and the risk-free rate is 3%. The portfolio's Jensen's alpha is:

A. −1.6%

B. +0.4%

C. +1.6% ✅

D. +3.0%

C 正确。考点：Jensen α 计算
要求回报 = 3% + 1.2×(10%−3%) = 3% + 8.4% = 11.4%；α = 13% − 11.4% = +1.6%。D 是直接拿 13%−10% 的错误算法（α 不是相对市场收益，而是相对 CAPM 要求回报）。

套利定价理论与多因子模型

6. The Arbitrage Pricing Theory & Multifactor Models

学习单元 2 · 要点：APT 的三条假设 · 因子意外值与因子 β · APT vs CAPM · 单/多因子计算 · 多因子对冲与模型风险 · Fama-French 三因子（及四/五因子扩展）

🇨🇳 中文导览

本章把 CAPM 的「单因子」推广成「多因子」。考试形式很固定：① 给你预期收益、因子 β 与因子意外值，让你算实际/预期收益；② 让你构造对冲组合把某个因子的 β 归零；③ 比较 APT 与 CAPM。

APT（Ross，1976）：R_i = E(R_i) + β₁F₁ + β₂F₂ + … + β_kF_k + e_i。这里的 F 是因子的意外值（实际值 − 预期值），β 是因子敏感度，e_i 是特质性误差且期望为 0。APT 只有三条假设：参与者追求利润最大化、市场无摩擦（无税无成本、可卖空）、不存在套利机会（一旦出现会被迅速抹平）。
APT vs CAPM：CAPM 单因子（市场）、假设多、来自均衡；APT 不指定用哪些因子（这是灵活之处也是缺点）、假设更少、来自无套利。经典四因子（Chen, Roll & Ross）：长短利率差、预期与非预期通胀、工业生产、低风险与高风险公司债利差。
多因子对冲：用因子敏感度相同的资产反向持仓即可把该因子 β 归零（例如做多 P1、做空 P2 把 GDP 因子对冲掉），但会保留甚至引入其他因子敞口；且因子与 β 会漂移，需要再平衡，因此必然存在模型风险。
Fama-French 三因子：E(R_i) = R_F + β_i,MRP_M + β_i,SMBF_SMB + β_i,HMLF_HML + e_i。SMB = 小减大（做多小公司、做空大公司）；HML = 高减低（做多高账面市值比、做空低账面市值比）。Carhart（1997）加动量成四因子；Fama-French（2015）加 RMW（盈利）与 CMA（投资保守度）成五因子。

关键术语：factor surprise因子意外值、factor beta / factor loading因子敏感度、idiosyncratic risk特质性风险、arbitrage套利。

🧭 知识网位置 CAPM 的「多因子升级版」：从单一 β 走向因子动物园。前置 → 第 5 章 · CAPM（单因子）本章之后 → 第 8 章 · ERM（体系化视角） · Part II · 因子理论 Factor Theory · Part I 定量 · 线性回归

6.1 Arbitrage Pricing Theory (APT) 套利定价理论

APT 由 Stephen Ross 于 1976 年提出，本质是一个多因子模型，把系统性风险拆得更细：因子可以是指数（标普 500、债券指数、商品指数），也可以是宏观变量（GDP、利率、工业生产、就业）。公式里最关键的一点：F 是因子的"意外值"（实际 − 预期），不是因子水平本身——所以如果实际值等于预期值，这一项为 0，不产生额外收益。β 是因子敏感度（也称因子载荷），e_i 的期望为 0，代表模型解释不了的公司特有风险，可以通过分散化消除。与 CAPM 对比的五条结论：因子数量多 vs 单一、多个因子 β vs 单一 β、更灵活（分析师自选因子）vs 固定、假设更少更简单 vs 假设多、来源于无套利 vs 来源于均衡。APT 不告诉你该用哪些因子，这既是优点也是缺点（因子与 β 需定期重估——这正是模型风险的一部分）。

Proposed by Stephen Ross (1976). A multifactor model: expected return is a linear function of several risk factors (indices like S&P 500, bond/commodity indices + macro variables like GDP, interest rates, production, employment).

APT actual-return equation ⭐ R_i = E(R_i) + β₁F₁ + β₂F₂ + … + β_kF_k + e_i
F_k = factor surprise (deviation from expectation); β_k = factor sensitivity; e_i = idiosyncratic error (E[e]=0).

APT assumptions (fewer than CAPM)

Investors maximise profit.
Markets are frictionless (no taxes/costs, shorting allowed).
No arbitrage opportunities — any discovered are quickly exploited away.

APT vs CAPM — the comparison table

	CAPM	APT
# of factors	One (market / S&P 500)	Multiple (unspecified by the model)
Risk exposure	Single beta	Multiple factor betas
Flexibility	Fixed	More flexible — analyst chooses factors
Assumptions	Many (perfect markets, homogeneous expectations…)	Fewer / simpler
Source of the model	Equilibrium (MPT)	Arbitrage / no-arbitrage

APT is flexible but doesn't tell you WHICH factors to use — a downside in practice (factors & betas must be periodically re-estimated).

Chen, Roll & Ross's classic 4 factors: (1) term spread (short vs long rates), (2) expected vs unexpected inflation, (3) industrial production, (4) low-risk vs high-risk corporate bond spread.

6.2 Inputs to a multifactor model 多因子模型的输入

多因子模型的三类输入：股票的预期收益、一组因子及其取值（可以是宏观变量，也可以是公司属性如市盈率、收入趋势）、以及每个因子对应的β（因子敏感度/因子载荷）。误差项 e_i 捕捉模型未纳入但仍与收益相关的因素、随机性、以及罢工、自然灾害、关税变化等公司特有事件，其期望值（默认值）为 0。核心直觉是因子 β 会放大因子的运动：例如 β_GDP = 2.0，市场一致预期 GDP 为 3.2% 而实际公布 2.2%，意外值为 −1%，则该股票预期下跌 2.0 × (−1%) = −2%。计算时务必用"实际 − 预期"得到意外值，并保留正负号。

💡 GDP factor illustration

Suppose β_GDP = 2.0, consensus GDP = 3.2%, actual = 2.2% → surprise = −1%. Expected stock fall = β × surprise = 2.0 × (−1%) = −2%. A factor beta amplifies the factor's movement.

6.3 Calculating expected returns — single vs multifactor 计算预期收益 ⭐

这一节就是套公式，但有两个易错点。第一，先算因子意外值：单因子例子里预期 GDP 3.2%、实际 2.6%，意外值 = −0.6%，于是 R = 0.10 + 2.0 × (−0.006) = 8.8%。第二，多因子是各因子贡献的加总：再加消费者信心（β = 1.5，预期 1.0%、实际 0.75% → 意外 −0.25%），R = 0.10 − 0.012 − 0.00375 = 8.425% ≈ 8.43%。判断模型好坏的标准是预测值与实际值的差距：本例实际 8.25%，多因子的 8.43% 比单因子的 8.8% 更接近，说明多因子捕捉到了更多的系统性风险，剩下的残差就是可分散的特质性风险。解题思路：因子选得越多、预测越贴近实际，但也越容易过拟合，因此原则是"在能覆盖被定价的系统性风险的前提下，因子尽量少"。

💡 HealthCare Inc. (HCI) — worked

Single-factor (GDP): E(R)=10%, β_GDP=2.0, expected GDP 3.2%, actual 2.6% → surprise = −0.6%.

single-factor R_HCI = 0.10 + 2.0(−0.006) = 8.8%

Multifactor (add consumer sentiment): β_CS=1.5, expected CS 1.0%, actual 0.75% → surprise = −0.25%.

multifactor R_HCI = 0.10 + 2.0(−0.006) + 1.5(−0.0025) = 0.10 − 0.012 − 0.00375 = 8.425% ≈ 8.43%

Actual return was 8.25%. The multifactor (8.43%) is closer to actual than single-factor (8.8%) — i.e., it captures more systematic risk. The residual (actual − predicted) is idiosyncratic risk (e_i), which diversifies away.

6.4 Hedging multiple factors 构建多因子对冲组合 ⭐

多因子对冲的做法是：找一个与被对冲资产在某因子上敏感度相同的资产或组合，做反向持仓，该因子的 β 就被减到 0。例子中做多 P1、做空 P2，GDP 因子 β = 0.50 − 0.50 = 0，被完全对冲；但组合仍保留 +0.30 的消费者信心敞口，并因做空而新增 −0.25 的就业敞口。同理可做多 P1、做空 P3：这次中和的是消费者信心（0.30 − 0.30 = 0），保留 GDP、新增制造（ISM）敞口。想同时中和两个因子，可以构造一个只含这两个因子敞口的"因子组合"（例如 50% 某 GDP 敞口衍生品 + 30% 某 CS 敞口衍生品 + 20% 无风险资产），再与 P1 反向组合。这套工艺还能用来"隔离套利"：若 P1 预期收益 12%、对冲组合 10%，做多 P1 / 做空组合即可锁定 2% 的（模型计算出的）套利空间。这就是多因子对冲的实质困难：你很难只对冲一个因子而不动其他因子；且因子与 β 会漂移、再平衡过频（成本）或过疏（跟踪误差）都有代价、危机中模型假设可能失效——必然伴随模型风险。另两条必记：APT 的结论对充分分散的组合成立，允许个别证券偏离，但几乎所有证券都偏离时套利机会就会出现；组合充分分散的标准是成分资产来自不同资产类别、相关性低。

💡 Worked — long Portfolio 1, short Portfolio 2

Factor	P1 sensitivity	P2 sensitivity
GDP surprise	0.50	0.50
Consumer sentiment	0.30	—
Unemployment	—	0.25

Long P1 + short P2 → GDP surprise beta = 0.50 − 0.50 = 0 (hedged away). Remaining: +0.30 consumer sentiment, and −0.25 unemployment (because shorted). To hedge one factor you find assets offsetting exposure, but you retain/introduce others — that's the challenge of multifactor hedging.

Challenge of multifactor hedging

Truly hedged portfolios must be well diversified. APT holds for well-diversified portfolios even if individual securities violate it — but if too many securities in the portfolio violate it, arbitrage becomes available. Factors and betas also drift, so hedges must be rebalanced and re-estimated frequently.

6.5 The Fama–French three-factor model Fama-French 三因子模型 ⭐

Fama-French 解决了 APT「不告诉你用哪些因子」的问题，明确给出三因子：市场风险溢价 RP_M（就是 CAPM 那一项）、SMB（Small Minus Big，做多小市值、做空大市值，因为历史上小公司收益更高、被认为风险更高）、HML（High Minus Low，做多高账面市值比、做空低账面市值比，即低估值公司）——注意账面市值比与市净率互为倒数。公式是 E(R_i) = R_F + β_i,MRP_M + β_i,SMBF_SMB + β_i,HMLF_HML + e_i。例题口径：0.0275 + 0.85×0.085 + 1.65×0.025 + (−0.25)×0.0175 = 13.66%；与实际收益的偏离即为 alpha，来源可能是 e_i，也可能是还缺别的因子。扩展版本要记年份与内容：Carhart（1997）加动量 → 四因子；Fama-French（2015）加 RMW（盈利强减弱）与 CMA（投资保守减激进）→ 五因子。

Fama-French 3-factor ⭐ E(R_i) = R_F + β_i,MRP_M + β_i,SMBF_SMB + β_i,HMLF_HML + e_i

Market factor (RP_M) — the CAPM market risk premium.
SMB = "Small Minus Big" 小减大 — a hedge long small firms / short big firms (small firms earn more historically).
HML = "High Minus Low" 高减低 — long high book-to-market / short low book-to-market value firms.

Extensions: Carhart (1997) added a momentum factor → four-factor. Fama–French (2015) proposed RMW (profitability: robust minus weak) and CMA (conservative minus aggressive) → five-factor.

💡 Fama-French worked example

β_M=0.85, β_SMB=1.65, β_HML=−0.25; RP_M=8.5%, F_SMB=2.5%, F_HML=1.75%, R_F=2.75%.

calculate E(R_i) = 0.0275 + 0.85(0.085) + 1.65(0.025) + (−0.25)(0.0175)
= 0.0275 + 0.07225 + 0.04125 − 0.004375 = 13.66%

Any deviation from 13.66% = alpha (from e_i or missing factors).

🔑 英文要点速查 · English Essentials

arbitrage pricing theory (APT) 套利定价理论 — Ross (1976); multifactor model: R_i = E(R_i) + β₁F₁ + … + β_kF_k + e_i.
factor surprise 因子意外值 — F = actual − expected; a zero surprise contributes nothing.
factor beta / loading 因子敏感度/载荷 — Amplifies the factor's movement (β = 2.0 → double the surprise, sign included).
arbitrage 套利 — APT's premise: no arbitrage opportunities persist; any found are exploited away quickly.
APT assumptions APT 假设 — Profit-maximising investors; frictionless markets (shorting allowed); no arbitrage. Fewer than CAPM.
idiosyncratic error (e_i) 特质性误差 — Unexplained firm-specific return; E[e] = 0; diversifies away.
Chen, Roll & Ross factors 经典四因子 — Term spread; expected vs unexpected inflation; industrial production; credit spread.
SMB (Small Minus Big) 小减大 — Long small caps / short large caps; small firms earn more historically.
HML (High Minus Low) 高减低 — Long high book-to-market / short low; book-to-market is the inverse of price-to-book.
Carhart momentum Carhart 动量因子 — Added 1997 → four-factor model.
RMW & CMA 盈利/投资因子 — Fama-French (2015): robust-minus-weak profitability; conservative-minus-aggressive investment → five-factor.
multifactor hedging & model risk 多因子对冲与模型风险 — Hedging one factor retains/introduces others; betas drift → rebalance (cost vs tracking error); crisis breaks assumptions.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. Which of the following is NOT an assumption of the Arbitrage Pricing Theory?

A. Market participants seek to maximise profits.

B. Markets are frictionless — no taxes or transaction costs, and short selling is available.

C. Expected returns are determined by exposure to a single market factor. ✅

D. There are no arbitrage opportunities; any that emerge are quickly exploited away.

C 正确（它不是 APT 的假设）。考点：APT 假设 vs CAPM 假设
"预期收益只由单一市场因子决定"是 CAPM 的世界观；APT 恰恰相反——它用多个因子、且不预先指定用哪些。A、B、D 正是 APT 的三条（很少的）假设。

Q2. A stock has a GDP-surprise factor beta of 1.8 and an expected return of 9%. Consensus GDP growth was 2.5%, but the actual figure came in at 1.9%. Using the single-factor model (setting the error term to zero), the predicted return is closest to:

A. 7.9% ✅

B. 9.0%

C. 10.1%

D. 12.4%

A 正确。考点：因子意外值计算（先算 surprise，再乘 β）
surprise = 1.9% − 2.5% = −0.6%；贡献 = 1.8 × (−0.6%) = −1.08%；R = 9% − 1.08% ≈ 7.9%。B 是漏算因子贡献；C、D 都把符号做反（实际 GDP 低于预期应是负贡献）——"实际 − 预期"的顺序和正负号是本题唯一的坑。

Q3. Portfolio 1 has sensitivities: β_GDP = 0.60, β_CS = 0.40. Portfolio 2 has: β_GDP = 0.60, β_JOBS = 0.30. An investor goes long P1 and short P2. Which description of the resulting factor exposures is correct?

A. All exposures are eliminated.

B. GDP exposure is hedged to zero; the portfolio retains +0.40 consumer-sentiment exposure and gains −0.30 jobs exposure. ✅

C. GDP exposure doubles; other exposures are unchanged.

D. Only the jobs exposure remains, at +0.30.

B 正确。考点：多因子对冲的"对冲掉一个、保留/引入其他"
GDP：0.60 − 0.60 = 0（对冲掉）；CS：0.40 保留；JOBS：因为做空 P2 而新增 −0.30。A 错——多因子对冲很难把所有敞口归零；C 算反；D 把留存与新增弄混。这就是教材强调的"对冲一个因子的代价是动到其它因子"。

Q4. Which of the following factors is NOT included in the Fama–French three-factor model?

A. The market risk premium.

B. SMB (small minus big).

C. HML (high minus low).

D. Momentum. ✅

D 正确（动量不在三因子里）。考点：三因子 vs 四因子 vs 五因子
Fama-French 三因子 = 市场 + SMB + HML；动量是 Carhart（1997）追加的第四因子；RMW（盈利）与 CMA（投资）是 Fama-French（2015）的五因子扩展。题目问"不在三因子里"，答案就是动量。

有效数据聚合与风险报告（BCBS 239）

7. Principles for Effective Data Aggregation & Risk Reporting

学习单元 2 · 要点：风险数据聚合的定义与四类收益 · 模型风险四类（输入/估计/估值/对冲）· BCBS 239 原则地图（治理 1–2 / 聚合 3–6 / 报告 7–11）·「不得牺牲一条原则换另一条」

🇨🇳 中文导览

本章是全套资料里少见的纯定性章节，围绕巴塞尔委员会关于风险数据聚合与风险报告的14 条原则（BCBS 239）展开（本章详讲 1–11 条）。命题方式几乎固定：给你一段银行的做法，问它符合或违反哪条原则。

定义：风险数据聚合 = 按银行的风险报告需求来定义、收集与加工风险数据，使银行能衡量自己相对于风险容忍度/偏好的表现，过程包括拆解、分类与合并。
四类收益：更早预见问题（整体而非孤立地看风险）、压力期更容易找到恢复路径（如合适的并购方）、提升可处置性（resolvability）（监管当局能拿到聚合数据处置问题银行，对 G-SIB 尤为关键）、以及改善战略决策与盈利能力。题干强调"银行已陷入困境"时，正确选项通常是可处置性。
数据差 → 模型风险：模型风险分四类——输入风险、估计风险、估值风险、对冲风险，其中输入风险直接来自数据获取。历史上银行数据按部门割裂收集、重复、存储介质代际不兼容；危机时无法快速识别跨条线的集中度（同一客户在车贷、房贷、信用卡三个系统里可能是三个"客户"）。
两条铁律：① 每类风险应力求单一权威数据来源（多套并行 = 违反准确性原则）；② 不可用一条原则去牺牲另一条（例如为赶时效牺牲完整性、为准确而放弃适应性，都是教材点名的反例）。

关键术语：risk data aggregation风险数据聚合、resolvability可处置性、input risk输入风险、manual workaround人工绕行、reference data参考数据。

🧭 知识网位置治理主题的「数据篇」：前面的章节讲「人怎么管风险」（流程、策略、董事会），本章讲「信息怎么支撑风险决策」——数据是风险管理的原料。前置 → 第 2 章 · 风险管理流程与策略 · 第 3 章 · 公司治理本章之后 → 第 8 章 · ERM（把治理与数据并入统一框架） · 第 9 章 · 金融灾难（数据割裂的反面教材）

7.1 What is risk data aggregation? 风险数据聚合

先背下委员会的原文定义：按银行的风险报告需求定义、收集与加工风险数据，使银行能够衡量自己相对于风险容忍度/风险偏好的表现；过程上就是对数据做拆解、分类与合并，目的是让报告可靠地反映风险。四类收益要能对应到情景：① 提升预见问题的能力（整体视角下才能看到地平线上的麻烦——孤立地看风险看不到）；② 压力时期更容易找到恢复路径（例如识别合适的并购方以恢复经营能力）；③ 提升可处置性（resolvability）——监管当局拥有聚合数据才能处置问题银行，这对全球系统重要性银行（G-SIB）尤其关键；④ 强化风险职能后，战略决策更好、效率更高、损失概率更低、最终盈利更好。

BCBS definition

"Defining, gathering and processing risk data according to the bank's risk reporting requirements to enable the bank to measure performance against its risk tolerance/appetite." The process means breaking down, sorting, and merging data and datasets so reports reflect risk reliably.

Benefits of effective aggregation ✅

Anticipate problems — see risks holistically ("see problems on the horizon").
Identify routes back to financial health in stress (e.g., a suitable merger partner).
Improved resolvability 可处置性 in stress/failure — key for G-SIBs (global systemically important banks).
Stronger risk function → better strategic decisions, efficiency, lower chance of loss, higher profitability.

Exam tip

If the scenario emphasises a bank that is already in trouble and asks for the most relevant benefit, choose improved resolvability.

7.2 Challenges & the impact of poor data 实施挑战与劣质数据的后果

这一节的逻辑链是「数据差 → 模型差 → 决策错」。银行用模型做从风险敞口分析到日常经营的一切，模型依赖数据，因此数据获取是模型风险的第一环——输入风险；模型开发者必须证明所用数据与模型背后的理论和方法一致，模型必须经过审查与验证（美国监管机构也发布了模型风险管理的指导意见）。模型风险四大类：输入、估计、估值、对冲。历史上的具体障碍：数据按部门/职能各自采集、重复采集、存储介质换代（打孔卡、磁带→软盘）互不兼容导致数据毁损；2007 年危机中许多银行无法快速、准确地识别跨业务条线与集团层面的集中度——这正是巴塞尔委员会设立专门小组调研、最终发布原则的背景。两个必记例子：① 同一客户在车贷、房贷、信用卡三个系统里用不同识别码，银行就看不出对"同一个客户"的真实敞口——所以标准必须跨部门一致；② 危机前虚假/欺诈按揭申请"一单接一单"进来，每单看似小问题，累积却酿成全局性失败——数据质量问题往往单点小、聚合大。背景补充：BCBS 239 发布后银行里出现了更多负责数据治理的首席数据官；AI 与数据分析正在自动化数据采集、分析与报告环节，委员会认为长期收益大于投入。

Model risk — the family

Poor data feeds model risk 模型风险, which includes: input risk, estimation risk, valuation risk, hedging risk 输入/估计/估值/对冲风险. Historically, bank data collection was disjointed (per department, duplicated, storage generations incompatible). During 2007–09, banks couldn't quickly identify concentrations across business lines or at the group level.

Illustration: the same customer with an auto loan + mortgage + credit card may appear as three "customers" if identification codes differ across departments — the bank cannot see its true single-name exposure. Standards must be consistent across departments.

The "one-at-a-time" failure

Before the 2007–09 crisis, erroneous or fraudulent mortgage applications arrived one at a time — each looked like an isolated flaw. The ultimate failure was global, driven by the aggregate of all those flawed applications. Data-quality problems are small point-by-point but large in aggregation.

7.3 Principles 1–2: governance & data architecture 治理与数据架构 ⭐

原则 1（治理）要求风险数据聚合与报告纳入全行风险管理框架，并有与其它原则一致的强治理安排：高管层在实施前批准框架、以财务与人力资源支持、纳入 IT 战略规划；实践要完全成文、由具备 IT 与数据专业能力的人独立审查与验证；在新产品、并购、剥离等新举措中必须被考虑——并购时要评估标的的聚合能力并设定整合时间表；聚合与报告决策不受银行物理位置、地理分布或法律架构影响；董事会持续关注实施与合规、并购后要复核。原则 2（数据架构与 IT 基础设施）要求「设计、建设并维护」能在正常与压力/危机两种状态下支撑聚合与报告（同时满足其它原则）的架构：纳入规划与业务影响分析、建立集团级统一的数据分类与架构、明确数据责任人；架构要包含元数据（数据特征信息）与命名规范（法律实体、对手方、客户、账户）。允许使用多个数据模型，但必须有稳健的自动化对账机制兜底。

#	Principle	Core requirement
1	Governance 治理	Part of the risk framework; senior mgmt approves; documented; independently reviewed/validated; considered in M&A; unaffected by bank structure
2	Data architecture & IT infrastructure	Supports aggregation/reporting in normal & stress; integrated classification; metadata & naming conventions; automated reconciliation
3	Accuracy & integrity	Accurate/reliable in normal & crisis; largely automated; controls as robust as accounting's; single authoritative source per risk type; documented manual workarounds
4	Completeness 完整性	All material risk data, on & off balance sheet; by business line/entity/asset class/industry/region; gaps identified & explained; not all risks need one metric
5	Timeliness 时效性	Timely & up-to-date in normal & crisis; critical risks (credit, counterparty, trading, market concentrations, liquidity) fast
6	Adaptability 适应性	On-demand / ad hoc reporting; stress-test friendly; absorbs new business & regulation; drill-down capability
7	Accuracy (reports)	Accurate & precise; reasonableness checks; error reports; precision standards like accounting materiality
8	Comprehensiveness	All material risks incl. stress results, capital, concentrations, funding plans; Pillar 1 + Pillar 2 risks
9	Clarity & usefulness 清晰有用	Tailored to recipients; data + analysis + interpretation + qualitative explanation
10	Frequency 频率	Set by board/senior mgmt; increase in stress; varies with risk-change speed (trading > lending)
11	Distribution	Delivered to relevant parties timely, confidentially

Governance principle (1) — the checklist

Part of the bank's overall risk management framework; senior management approves before implementation.
Supported with financial and human resources; included in strategic IT planning.
Fully documented; independently reviewed & validated by people with IT, data, and risk-reporting expertise.
Considered in new initiatives — new products, acquisitions/divestitures (assess the target's capabilities, set an integration time frame).
Unaffected by the bank's structure — independent of physical location, geographic presence, legal organisation.
The board stays aware; RDARR is reviewed after M&A.

Data architecture (2) — the details

Design, build and maintain architecture & IT infrastructure for normal AND stress/crisis conditions, while meeting the other principles.
Part of planning processes; subject to business impact analysis; integrated data classifications & architecture across the group; multiple data models allowed only with robust automated reconciliation.
Data architecture includes metadata 元数据 (data characteristics) and naming conventions for legal entities, counterparties, customers, accounts.
Accountability, roles, responsibilities and ownership defined; controls throughout the data life cycle.

The four data models (schemas)

Semantic — structures data in logical order + meaning and relationships ("the basic meaning of data").
Conceptual — the most abstract; maps concepts/relationships; confirms human understanding of systems & objectives.
Logical — describes data in as much detail as possible; not concerned with implementation.
Physical — the components to actually build the database (tables, columns, keys, relationships); translates concepts/logical models into implementable form.

Banks with compliant architecture understand risks better and can adjust to changes in business activity.

7.4 Principles 3–6: aggregation standards 聚合侧四标准 ⭐

原则 3（准确与完整）要求正常与危机下都生成准确可靠的数据，尽可能自动化以压低错误率；风险数据的控制强度要与会计数据同级；用电子表格、数据库等人工/桌面手段时必须有有效控制；数据要与其它数据（含会计数据）对账；每类风险力求单一权威来源；定义要全行一致（可维护数据词典）；允许在需要专业判断处保留人工介入，但要记录并解释人工绕行（manual workaround）——为什么它对准确性至关重要、如何把影响降到最低。原则 4（完整性）要求覆盖集团内全部重大风险数据（表内+表外），可按业务线、法律实体、资产类型、行业、地区等维度切分；并非所有风险都必须用同一个度量；若有缺口，必须识别并向监管说明。原则 5（时效性）要求正常与危机下都及时；快慢取决于风险的性质、波动性与重要性，也取决于银行自设的报告频率；系统必须能在危机中快速产出关键风险的聚合数据——考试常以此为清单记忆。原则 6（适应性）要求能应付
按需/临时请求（含压力情景与监管问询）：灵活支持压力测试与情景分析、可定制（异常提示、仪表盘、要点摘要）、能纳入新业务与外部因素、能吸收监管变化，并支持"下钻"（如按国家/地区提取某类信用敞口）。

Principle 3 highlights 准确与完整

Largely automated aggregation; controls as robust as accounting controls; reconcile with other bank data incl. accounting.
Single authoritative source for each specific risk type; consistent definitions; a risk-data dictionary of concepts/terms.
Human intervention appropriate when professional judgment is required — balance manual & automated; document manual workarounds (why critical, actions to minimise impact).
Monitor accuracy; corrective plans for poor data.

Principle 5 — critical risks to aggregate fast in crisis

Aggregated credit exposures to large corporate borrowers;
counterparty credit exposures (including derivatives);
trading exposures, positions and operating limits;
market concentrations by region & sector;
liquidity risk indicators;
time-critical operational risk indicators.

Timeliness varies by line: trading positions must be aggregated faster/more often than corporate lending.

⚠️ The principles interact — never trade one off against another

The accuracy & integrity, completeness, timeliness and adaptability standards work together. This chapter's own counter-examples: a bank takes shortcuts on completeness for speed; or accuracy suffers in a hurry to meet timeliness; or data is compiled for accuracy but becomes inflexible and not adaptable. The committee's instruction: meet each principle while still meeting the others.

7.5 Principles 7–11: reporting standards 报告侧五标准 ⭐

报告侧五原则的核心各一句话：7 准确——报告要准确、精确、可对账可验证；流程要定义、要做合理性检查、要描述并核验数据中的数学与逻辑关系、要有错误报告机制；对情景分析、敏感性分析、压力测试等"近似结果"同样要求可靠准确及时；准确要求对标会计的重要性（materiality）门槛——"若一项遗漏影响了风险决策，即视为重大"。8 全面——覆盖全部重大风险领域，深度与范围同银行规模、复杂度、接收者需求相称；须包含信用、市场、流动性、操作风险，压力测试结果、资本充足性、监管资本、流动性与资本预测、风险集中度、融资计划；报告应前瞻（含预测与压力测试），并结合新兴风险讨论风险偏好。Pillar 1（市场/信用/操作——最低资本要求挂钩）与 Pillar 2（业务、声誉、集中度、战略等）都要覆盖。9 清晰有用——针对接收者定制（董事会、高管、风险委员会、交易员、信贷员需求各不相同），内容 = 数据 + 分析 + 解释 + 定性说明；聚合程度越高，越需要定性解读；术语要分类、建清单。10 频率——由董事会/高管设定，反映接收者需求、风险性质与变化速度；压力期要提高频率（流动性、信用、市场报告可能需即时产出）；但有时也必须放慢——数据量过大（如随机现金流模拟）会拖垮质量核查，且情景迭代之间要保持一致。频率口径：交易条线 > 信贷条线；风险委员会 > 董事会。11 分发——及时送达相关方，同时保密；银行要能确认接收者及时收到。

💡 Materiality (Principle 7)

Supervisors expect accuracy requirements on risk data "commensurate with and analogous to accounting materiality": if an omission influences risk decision-making, it is deemed material. Precision/accuracy requirements for regular & stress reports are set by the board/senior management.

Comprehensiveness (8) — the report checklist

Credit risk · market risk · liquidity risk · operational risk · stress-test results · capital adequacy & regulatory capital · liquidity projections · capital projections · risk concentrations · funding plans. Pillar 1: market, credit, operational (Basel I core). Pillar 2: business, reputation, concentration, strategic risk.

Compliance reality check

Banks broadly struggle with BCBS 239. A PwC study found better performance on reporting principles 7–11, but poorer on aggregation principles 3–6 — and governance/architecture (1–2) also lagged.

Ineffective practices (treated as non-compliant): ineffective quality control; improperly established quality rules (no minimum standards); lack of oversight or escalation; weak quality control; overuse of undocumented manual processes; no reconciliation between risk reports; no variance analysis; can't obtain foreign-subsidiary data in time; non-standardised reference data.

Effective practices: data-element certification; quality documentation; quality assurance mechanisms; per-risk-type quality assessment; documented & effective controls for manual processes.

🔑 英文要点速查 · English Essentials

risk data aggregation 风险数据聚合 — Defining, gathering, processing risk data per reporting requirements; breaking down, sorting, merging; measure performance vs risk tolerance.
resolvability 可处置性 — Authorities can access aggregated data to resolve a troubled/failing bank; critical for G-SIBs.
model risk family 模型风险四类 — Input risk, estimation risk, valuation risk, hedging risk; data acquisition is input risk.
BCBS 239 — Basel Committee's 14 principles (this chapter details 1–11) for overhauling data aggregation & reporting after the GFC.
Principles 1–2 治理与架构 — Governance (documented, independently validated, unaffected by structure, M&A-aware) + data architecture/IT (normal & stress, metadata, naming conventions).
Principles 3–6 聚合四标准 — Accuracy & integrity · completeness · timeliness · adaptability; interact — do NOT prioritise one at the expense of another.
single authoritative source 单一权威来源 — Strive for one authoritative source per risk type; controls as robust as accounting's.
manual workaround 人工绕行 — Documented, justified bypass of automation when judgment is needed; explain why and how impact is minimised.
Principles 7–11 报告五标准 — Accuracy · comprehensiveness · clarity & usefulness · frequency · distribution; frequent in stress, confidential delivery.
materiality (reporting) 重要性 — Accuracy analogous to accounting materiality: an omission that influences risk decisions is material.
frequency varies 频率差异 — Depends on recipient & risk speed: trading > lending; risk committee > board; increased in stress.
Pillar 1 / Pillar 2 risks 支柱一/二风险 — Pillar 1: market, credit, operational; Pillar 2: business, reputation, concentration, strategic (and others).

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. A global systemically important bank (G-SIB) has suffered heavy losses and now depends on emergency support. Its supervisor finds that cross-border and cross-business exposures cannot be aggregated quickly or accurately. According to the Basel Committee, which benefit of effective risk data aggregation is MOST directly relevant to this situation?

A. Higher profitability through lower funding costs.

B. Improved resolvability of the bank's problems by the authorities. ✅

C. Complete elimination of model risk.

D. Standardisation of external financial reporting.

B 正确。考点：四类收益与情景匹配
题干两个关键词——"已陷入困境"（stress/failure 语境）与"G-SIB"，对应教材中的可处置性（resolvability）：监管当局需要聚合数据来处置问题银行，这对 G-SIB 尤其重要。A 是一般性收益，不是本情形的"最直接"收益；C 夸大（数据质量无法消灭模型风险，只能降低输入风险）；D 偷换概念——要求的是与会计数据对账，不是把风险报告与财报合并标准化。

Q2. Which statement about BCBS 239's governance principle (Principle 1) is FALSE?

A. Risk data aggregation and reporting should be part of the bank's overall risk management framework, approved by senior management before implementation.

B. Aggregation and reporting practices should be fully documented and independently reviewed and validated.

C. Decisions about data aggregation and reporting may be tailored to the bank's physical location and legal organisation. ✅

D. Aggregation and reporting capabilities should be considered when the firm makes acquisitions or divestitures.

C 正确（它是错的）。考点：治理原则的"不受结构影响"铁律
原则 1 明确规定：聚合与报告决策应独立于银行的物理位置、地理分布与法律架构（unaffected by the bank's structure）——C 恰好说反，故为假。A（纳入框架+高管批准）、B（成文+独立验证）、D（并购/剥离中评估标的聚合能力并设定整合时间表）都是原则原文要点。

Q3. Which practice is MOST consistent with the accuracy-and-integrity principle (Principle 3)?

A. Maintaining multiple independent sources for each risk type so they can cross-check one another.

B. Aggregation predominantly automated, with documented manual workarounds where professional judgment is required. ✅

C. Eliminating all manual processes so that errors cannot be introduced by staff.

D. Applying lighter controls than accounting, since risk data consists of estimates.

B 正确。考点：P3 三件套（单一来源 / 同会计级控制 / 人工绕行报备）
教材要求聚合"尽可能自动化"、控制与会计同级、人工介入在需要专业判断时是恰当的（但要记录绕行原因与影响最小化措施）——B 完整贴合。A 违反"每类风险力求单一权威来源"；C 走极端（原则允许并管理人工介入，而非消灭）；D 与"控制强度对标会计数据"相悖。

Q4. To meet a reporting deadline, a bank shortens its data-completeness checks for the week. Under BCBS 239, this practice is:

A. Acceptable, provided the missing checks would not affect reported accuracy.

B. Acceptable during stress periods, when speed matters most.

C. Not acceptable — banks should meet each principle while still meeting the others, not sacrifice one for another. ✅

D. Acceptable if senior management approves the shortcut in writing.

C 正确。考点：原则互动铁律
教材原话：委员会要求银行在满足每条原则的同时满足其它原则——不得把一条原则置于另一条之上，并直接点名"为时效走捷径牺牲完整性"这一反例。A/B/D 都是教材里不存在的"自我豁免"路径。

企业风险管理（ERM）与未来趋势

8. Enterprise Risk Management (ERM) & Future Trends

学习单元 2 · 要点：ERM vs 筒仓式 · 三项整合动机 · 治理（Turnbull / SOX / CRO）· 五维度 · 风险文化（FSB 四指标 + 五大阻碍）· 情景分析（优劣势）· 压力测试与资本规划（SCAP / DFAST / CCAR / CoCo / 反向压力测试）

🇨🇳 中文导览

本章讲企业风险管理（ERM）：为什么要把信用、市场、流动性、操作等风险当成一个整体来管理，而不是各业务线各自为政（筒仓式）。ERM 的定义要背：一套综合、集成的框架，管理企业关键风险以实现业务目标、最小化非预期盈利波动、最大化企业价值——也是识别与排序风险的最佳方式。考试以概念辨析与定性案例为主。

ERM企业风险管理：集中、整体地看待风险，寻找分散化与抵补机会，避免重复对冲的浪费。
Silo-based筒仓式/各自为政：各单元孤立评估，指标彼此不可比，高管层收到碎片化信息。
三项整合（动机精炼版）：整合风险组织 → 组织效率提升；整合风险转移 → 风险报告更好；整合业务流程 → 经营绩效改善。此外还聚焦"威胁生存"的最大风险、管理交叉风险与相关性、把压力测试的资本成本嵌入定价决策。
治理：董事会与高管定义风险偏好与损失容忍度；Turnbull 报告与 SOX 是治理驱动的里程碑；五维度 = 目标 / 结构 / 识别与度量 / 策略 / 文化。新 CRO 不必然改变文化——若被视为"换招牌"反而适得其反；调查显示仅约一半 CRO 会审查薪酬计划，而薪酬恰恰可能鼓励冒险行为。
风险文化：FSB 四指标（最高层基调、有效沟通与异议空间、激励、问责）；五大阻碍：风险指标变"风险杠杆"、风险教育（须覆盖董事会）、风险跨组织跨时间、文化循环（危机记忆淡化后风险行为复活）、数据诅咒（数据爆炸需要机器学习）。
情景分析：多变量联动 + 叙事；2007–09 后与压力测试一起成为 ERM 最重要的识别工具（VaR 模型的"25 个标准差"事件教训）。
压力测试与资本规划：美国 2009 年 SCAP 起步 → 联邦储备自 2011 年起年度测试；DFAST（资产 ≥$10B，年中，更指令式、报告更少）与 CCAR（≥$50B，年末，9 个季度动态预测）；三情景 = 基线 / 不利 / 严重不利；最低资本口径（2018）：CET1 4.5%、Tier1 风险加权 6%、总资本 8%、杠杆率 4%；CoCo 可在危机中转股，是监管偏爱的资本工具；反向压力测试从最坏 KPI 反推情景。

关键术语：enterprise risk management企业风险管理、risk culture风险文化、scenario analysis情景分析、stress testing压力测试、reverse stress testing反向压力测试、contingent convertible bonds或有可转换债券。

🧭 知识网位置把前面各章的风险类型（信用/市场/流动性/操作）收进一个企业级框架——治理主题的"总装车间"。前置 → 第 3 章 · 风险治理 · 第 7 章 · 数据聚合与报告（ERM 的信息底座）本章之后 → 第 9 章 · 金融灾难（ERM 失效的教训库） · 第 10 章 · 大危机剖析

8.1 ERM vs silo-based risk management 企业风险管理 vs 各自为政（筒仓式）⭐

ERM 与筒仓式的本质区别在"整体视角"。筒仓式在每个业务单元内孤立评估和缓释风险：交易员管市场风险、精算师管保险风险、管理层分析业务风险——各自用不同的方法与格式，高管层因此收到碎片化信息；而且它忽略风险之间的动态关联（一种风险会影响另一种；风险或其对冲在公司整体层面可能相互抵消），结果造成低效且昂贵的过度对冲。ERM 则集中、整体地管理：跨业务线、职能与风险类型审查；寻找分散化机会，也识别集中度；配备综合风险经理与首席风险官（CRO）；用可比的指标（如情景分析与 VaR）跨线汇总；用多触发（multitrigger）对冲工具省成本；并把风险管理与资本、资产负债表与融资策略整合。定义必须背：ERM 是管理企业关键风险以实现业务目标、最小化非预期盈利波动、最大化企业价值的综合集成框架，是识别与排序风险的最佳方式。记住表格六组对比的"整体 vs 孤立"逻辑。

ERM definition & the silo contrast

ERM 企业风险管理 is "a comprehensive and integrated framework for managing a firm's key risks to meet business objectives, minimise unexpected earnings volatility, and maximise firm value." It integrates all risk types and views them as a whole, exploiting interdependencies and offsetting risks to avoid inefficient, costly overhedging. A silo-based 筒仓式 system evaluates each risk inside one unit, in isolation, with non-comparable metrics — leaving senior management and the board with fragmented information.

Silo-based	ERM
Risk managed within lines of business	Across all lines, functions & types; seek diversification & concentration, avoid overhedging costs
Isolated risk managers	Integrated risk managers + a CRO
Multiple non-comparable metrics	Integrated, comparable metrics (scenario analysis, VaR)
Hard to see enterprise/aggregate risk	Aggregate across lines & types
Specific transfer tools per risk	Cost savings from integrated multitrigger hedging
Risk separate from balance-sheet & financing	Integrated with capital, balance-sheet & financing strategy

8.2 Motivations to adopt ERM 采纳 ERM 的动机

动机可归纳为一组"整合"与一组"聚焦"：三项整合——整合风险组织带来组织有效性提升、整合风险转移带来更好的风险报告、整合业务流程带来经营绩效改善（这是官方对动机的三句总结，务必能默写）；聚焦——定义并约束企业风险偏好、聚焦威胁企业生存的重大风险（而非日常单元风险）、从单个业务线识别对整个企业的威胁、在企业管理新兴风险（网络、声誉、反洗钱）。另有四项实务收益：支持监管合规、安抚股东与其他利益相关方、管理交叉风险（一种风险催生另一种）与风险间相关性、优化风险转移总成本并把压力测试的资本成本纳入定价与决策、把风险融入商业模式与战略选择。

Integration trio: risk organisation → organisational effectiveness; risk transfer → better risk reporting; business processes → improved business performance.
Defines the enterprise risk appetite & enforces constraints; focuses on threats to survival, not day-to-day unit risks.
Identifies threats to the whole firm from individual lines; emerging risks (cyber, reputation, AML) are better managed at enterprise level.
Supports regulatory compliance; reassures shareholders/stakeholders.
Manages crossover risks & correlations; optimises total transfer costs; embeds stress-testing capital costs into pricing; ties risk to strategy.

8.3 Governance & the five ERM dimensions 治理与五维度

治理最佳实践：董事会与高管定义风险偏好与损失容忍度；管理层持续投入风险议题；配置所需技能与组织架构；把全部关键风险纳入 ERM；权责清晰（含 CRO 角色）；落实监督、审计与监控目标。治理实践由一系列监管举措推动，其中点名两个：Turnbull 报告与萨班斯-奥克斯利法案（SOX，2002）。五个维度是记忆要点：① 目标——风险偏好挂钩薪酬计划与全局限额，战略目标与偏好一致；② 结构——CRO、全球风险委员会等角色、报告线与治理架构，确保全企业风险被识别、直接与间接损失都被考虑；③ 识别与度量——按对企业的影响、严重度、（理想的）发生频率度量；工具含情景分析、压力测试、VaR、风险总成本、全企业风险映射、专项指标与风险标记工具；必须意识集中度（信用集中、行业集中、贷款类型集中、单一借款人依赖单一供应商、对数据/分析/技术供应商的过度依赖）；④ 策略——在企业层面决定规避/缓释/转移，并识别风险转移工具；⑤ 文化——ERM 的心脏（见 8.4）。两个警示细节：新任命 CRO 若被冷眼视为"给文化换招牌"，不会自动改善文化；调查显示仅约半数 CRO 审查薪酬计划——而薪酬计划往往催生甚至鼓励冒险行为，这是不充分的。

Best practice governance: board & senior management define risk appetite & loss tolerance; committed management; required skills & structure (incl. the CRO role); integrate key risks; clear roles & responsibilities; oversight, audit & monitoring. Governance drivers cited: the Turnbull Report & the Sarbanes-Oxley Act.

Five ERM dimensions

Targets — risk appetite linked to compensation plans & global limits; strategic goals aligned with appetite.
Structure — roles (CRO, global risk committee, other committees), reporting lines; ensure enterprise-wide risks identified, direct & indirect losses considered.
Identification & metrics — impact, severity, (ideally) frequency; tools: scenario analysis, stress testing, VaR, total-cost-of-risk, enterprise-wide risk mapping, risk-specific metrics, risk flagging; concentration awareness (loans by sector/type/borrower; reliance on single suppliers, data/analytics/technology providers).
ERM strategies — avoid / mitigate / transfer decided at the enterprise level; identify transfer instruments.
Culture — "the heart and soul of ERM" (next section).

8.4 Risk culture 风险文化 ⭐

风险文化的定义：影响员工行为的目标、习俗、价值观与信念（显性的与隐性的），这些规范指导个体如何理解和应对风险。银行监管者把风险文化列为 2007–09 危机后银行倒闭的首要贡献因素之一。弱文化的著名例子：2012 年曝光的 LIBOR 操纵丑闻、银行洗钱、以及危机前许多机构的次级贷款放贷文化。为什么难建？因为它是多层结构：个体带着自己的风险态度入职（人口特征、家庭背景、个性、职业准则都会影响），同侪与管理层也在塑造行为——文化同时发生在企业层、小组层（群体思维、只招"同类人"）与个体层。衡量方法：用调查等数据建立风险文化评分（能看出文化变化，但无法量化文化失效带来的损失）。FSB 给出四个核心指标；此外还有一组"建设要素"（风险偏好知晓度、风险素养、风险信息流、管理者的风险/回报一致性、风险管理地位、举报与升级通道、董事会议题优先级、对违规者的处理、文化事件识别）；两组"阻碍"（指标变杠杆、教育覆盖到董事会、跨组织跨时间、文化循环、数据诅咒）。命题方式：给一段文化描述让你判断强/弱，或直接考 FSB 四指标。

FSB's four risk-culture indicators FSB 四指标

Tone from the top — do management's actions match stated risk goals? Do compensation plans support firm values?
Effective communication & challenge — are opposing views valued? "Openness to dissent"? Stature of risk management vs performers?
Incentives — are compensation plans aligned with risk appetite/culture?
Accountability — are expectations clear? Are escalation processes used?

Why strong risk culture is hard 五大阻碍

Risk indicators become risk levers — it's easier to manage the indicator than to actually improve culture.
Risk education — needed throughout, incl. the board (directors should be able to list key enterprise risks and relate them to appetite).
Risk across the organisation & across time — risks form in business lines with their own sub-cultures (e.g., sales cultures); mechanisms needed to spot risks appearing in multiple lines.
Culture cycle — culture is most visible in crisis; memories fade and risky behaviour revives (e.g., pre-crisis lending standards).
Curse of data — exploding data volumes (even sick-day records can be an indicator) require machine learning to extract insight.

Culture also interacts with credit cycles, industry practices, professional/regulatory standards, and country risk & corruption indices. Hiring a new CRO doesn't "fix" culture (possible window dressing/rebranding); business-line employees' tolerances matter, not just managers'.

Scenario analysis vs sensitivity analysis 情景分析 vs 敏感性分析

Sensitivity: change one variable; assess the model's sensitivity. Scenario: change multiple variables at once with a narrative explaining why and describing effects — captures abnormal/tail events.

8.5 Scenario analysis 情景分析 ⭐

VaR 等概率化风险指标在 2007–09 全球市场熔断中暴露弱点：按模型几乎不可能的"连续多日 25 个标准差波动"真实发生了。因此情景分析与压力测试成为 ERM 项目最重要的风险识别工具，帮企业理解异常/尾部事件的影响。优势逐条：不要求风险频率——只要合理可信（plausible）即可；直观透明；迫使企业想象最坏情形及后果；聚焦关键风险与敞口；帮助发现预警信号、制定应急预案；可前瞻（假想事件）也可基于历史；可以简单也可以高度精细；可用于制定风险偏好、设定限额、做资本充足性规划。劣势逐条：概率难估、不产出风险量化（本质是定性过程，虽然会围绕情景建定量模型）；情景复杂；可能低估事件及其影响；"选对情景了吗"（能完整开发的情景数量有限）；"盯对预警、做对预案了吗"；情景常基于上一次大危机而非未来风险；可信度难评估；有用性取决于准确性、全面性、以及是否覆盖未来风险而非只有历史。

Advantages ✅	Disadvantages ❌
Frequency irrelevant — the risk must be plausible	Probabilities hard to estimate; no risk quantification
Intuitive & transparent	Scenarios are complex
Forces imagining the worst case & consequences	May underestimate events & effects; only limited scenarios fully developed
Focuses on key risks; develops contingency plans	Often based on the last crisis, not future risks
Forward-looking (hypothetical or historical); simple or sophisticated	Credibility hard to assess; usefulness depends on accuracy, comprehensiveness, capturing future risks
Used for risk appetite, limits & capital adequacy planning	—

8.6 Stress testing in capital planning 压力测试与资本规划 ⭐

危机前银行多用历史情景（1997 亚洲危机、1998 俄罗斯债务违约、2001 九一一、2007 次贷、2008 雷曼与对手方危机），但问题在于：该放几个情景、历史与假想如何搭配，并不清楚——例如俄罗斯违约与 LTCM 濒临倒闭本是关联事件，银行却未必联合测试；2007–09 后监管者明确认定银行没有看到风险如何相互交织、参与者行为如何在压力下改变，且测的情景比实际发生的更温和，因此监管要求银行证明能抵御更严酷、更现实的情景。美国路径：2009 年 SCAP（监管资本评估计划）起步；联邦储备自 2011 年起年度测试；DFAST（≥$100 亿资产，年中）与 CCAR（≥$500 亿资产，年末）两者情景相同（监管设定），但 DFAST 更指令式、报告更少、资本行动假设有限。联邦储备要求三个宏观情景：基线（大银行经济学家共识预测）、不利（经济温和下行）、严重不利（全球衰退/萧条）。CCAR 要求 9 个季度预测，动态预测资产负债表与利润表（收入、贷款损失准备、违约与降级信用损失、新增贷款规则、监管比率），并提交资本计划（资本来源与使用、资本充足性方法论、详细资本政策、业务计划变化、各情景下维持最低资本、必要时如何募资、分红与回购计划）。口径数字（2018）：最低 CET1 4.5%、Tier1 风险加权 6%、总资本 8%、Tier1 杠杆率 4%；若压力测试下不达标，银行必须降低风险偏好。CCAR 的复杂度惊人——2018 年情景含 28 个变量（VIX、利率、GDP、住宅与商业地产指数等），银行还要叠加自身数百个因素。但它的收益也实在：情景跨 9 个季度展开（而非时点冲击）、因素相互联动（而非筒仓）、风险因子动态化；同一套情景让监管者看到系统性风险、并做"苹果对苹果"的比较。银行也可做反向压力测试：先定关键 KPI 的最坏结果，再倒推哪些情景导致它，从而暴露最脆弱的业务线与关键风险因子。最后一句要记：研究表明，造成上市公司企业价值重大损失的主因是战略失误，而非风险管理疏漏——情景分析要用于战略决策。

	DFAST	CCAR
Coverage	Banks ≥ $10bn assets	Banks ≥ $50bn assets
Timing	Mid-year	Year-end
Scenarios	Same — devised by supervisors
Nature	More prescriptive; less reporting; limited capital action assumptions	Complex: dynamic balance-sheet & income projections over 9 quarters

Fed's three macroeconomic scenarios

Baseline — consensus forecasts of large-bank economists.
Adverse — moderately declining economy.
Severely adverse — global recession/depression; falling demand for fixed-income investments.

Minimum capital standards (2018): CET1 4.5% · Tier 1 risk-based 6% · total risk-based 8% · Tier 1 leverage 4%. Banks must meet them under all scenarios; failing → lower risk appetite.

💡 Contingent convertible bonds (CoCos) 或有可转换债券

Regulators favour CoCos: they convert to equity if the bank is in capital trouble (easing cash outflows in stress). Conversion triggers are accounting events (e.g., Tier 1 level) or market events (e.g., share-price drop). They act as insurance → a risk transfer in ERM terms. Because they lose value after a shock, they may discourage risk-taking (executive compensation exposed to downside). And unlike classical insurance, there's no need to define the risk type in advance.

💡 Reverse stress testing 反向压力测试

Identify worst-case outcomes on key performance indicators (KPIs), then work backwards to see which scenarios produced them — exposing the most vulnerable business lines and the key risk factors. Other uses beyond compliance: specifying risk appetite & limits, checking capital plans' reasonableness, developing warning signals, contingency planning for liquidity/credit/funding.

🔑 英文要点速查 · English Essentials

enterprise risk management (ERM) 企业风险管理 — Comprehensive, integrated framework: meet objectives, minimise unexpected earnings volatility, maximise firm value; best way to identify & prioritise risks.
silo-based risk management 筒仓式 — Per-unit, isolated risk management; non-comparable metrics; ignores interdependencies → costly overhedging.
integration trio 三项整合 — Risk organisation → organisational effectiveness; risk transfer → better reporting; business processes → business performance.
five ERM dimensions 五维度 — Targets · structure · identification & metrics · strategies · culture.
risk culture 风险文化 — Goals, customs, values, beliefs influencing behaviour; a primary contributor to bank failures (seen post-2007).
FSB four indicators FSB 四指标 — Tone from the top; effective communication & challenge; incentives; accountability.
culture cycle 文化循环 — Culture most visible in crisis, fades with memory; risky behaviour revives.
scenario analysis 情景分析 — Multiple variables + narrative; frequency irrelevant (плаusibility enough); became THE key ERM identification tool post-crisis.
stress testing (US) 压力测试 — SCAP 2009; annual since 2011; three scenarios: baseline, adverse, severely adverse.
DFAST vs CCAR — ≥$10bn mid-year, prescriptive, less reporting vs ≥$50bn year-end, 9-quarter dynamic projections; same supervisor-set scenarios.
contingent convertible bonds (CoCos) 或有可转债 — Convert to equity in capital trouble; risk transfer; discourage risk-taking; no pre-defined risk type needed.
reverse stress testing 反向压力测试 — Start from worst-case KPIs, work backwards to scenarios; spots vulnerable lines.
strategic blunders — Studies: strategic mistakes, not risk-management lapses, drive significant enterprise-value loss.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. A key advantage of ERM over silo-based risk management is that ERM:

A. eliminates the need for a Chief Risk Officer.

B. allows interdependent and offsetting risks to be viewed together, avoiding costly overhedging. ✅

C. replaces scenario analysis with sensitivity analysis.

D. guarantees that losses cannot occur.

B 正确。考点：ERM 的本质 = 整体视角
筒仓式的最大缺陷是忽略风险相互关联、导致公司层面过度对冲；ERM 跨线整合、寻找分散化与抵销（offsetting）机会——B 正是教材原文逻辑。A 反了（ERM 通常配 CRO，但非"消除需求"）；C 反了（ERM 用情景分析与压力测试，且不排斥敏感性分析）；D 夸大（没有任何框架能保证零损失）。

Q2. Regarding U.S. stress testing, which statement is MOST accurate?

A. CCAR applies to smaller banks than DFAST does.

B. DFAST and CCAR use different supervisory scenarios.

C. DFAST is more prescriptive, requires less reporting, and has limited capital action assumptions. ✅

D. CCAR projections cover a single point-in-time shock.

C 正确。考点：DFAST vs CCAR 三组对照
两者情景相同（监管设定）；DFAST 覆盖 ≥$100 亿、年中、更指令式、报告少、资本行动假设有限；CCAR 覆盖 ≥$500 亿、年末、9 个季度动态预测（不是时点冲击）。A 说反（CCAR 覆盖更大的银行）；B 与"情景相同"矛盾；D 与"9 个季度展开"矛盾。

Q3. Which of the following is an advantage of scenario analysis over probabilistic measures such as VaR?

A. It assigns precise probabilities to tail events.

B. The frequency of the risk is irrelevant — it only needs to be plausible. ✅

C. It changes only one variable at a time.

D. It ignores relationships among variables.

B 正确。考点：情景分析优势首条
教材优势清单第一条：频率无关——风险只需"合理可信"。A 恰是情景分析的劣势（概率难估、无法量化）；C 描述的是敏感性分析（情景是多变量联动）；D 反了（情景正因多变量联动才能刻画尾部）。

Q4. Which statement about risk culture is MOST accurate?

A. Appointing a new Chief Risk Officer is generally sufficient to fix a weak risk culture.

B. Risk culture matters only for senior managers; lower-level employees' risk attitudes are irrelevant.

C. The FSB's four indicators include tone from the top, effective communication & challenge, incentives, and accountability. ✅

D. Risk culture can be precisely quantified in monetary terms.

C 正确。考点：FSB 四指标
A 错——教材明确：新 CRO 可能被视为"换招牌"（rebranding），不必然改变文化；B 错——文化必须浸润整个组织，业务线员工的容忍度同样重要；D 错——问卷调查可形成文化评分，但无法量化文化失效对应的损失。

从金融灾难中学习

9. Learning from Financial Disasters

学习单元 3 · 要点：九大案例与风险类型一一对应 · 静态 vs 动态对冲 · contango/backwardation · 相关性骤升与杠杆（LTCM）· 前台后台分离（巴林）· 治理失败（安然）· 网络风险（SWIFT）

🇨🇳 中文导览

本章用真实金融灾难案例，把抽象风险类型落到具体事件，是高频定性考点。每起案例都对应一种风险与一条教训：利率风险（储贷危机）、融资流动性风险（雷曼/大陆伊利诺伊/北岩）、对冲策略失误（Metallgesellschaft）、模型风险（Niederhoffer / LTCM / 伦敦鲸）、违规交易与误导性报告（巴林）、金融工程（Bankers Trust / 奥兰治县 / 萨克森）、声誉风险（大众）、公司治理（安然）、网络风险（SWIFT）。考试常给案例名让你选对应风险类型与教训。

Funding-liquidity risk融资流动性风险：用短债融长资产 → 期限错配，融资一断即崩。
Model risk模型风险：错误假设/错误模型/错误估计量致损（Niederhoffer 的卖出虚值看跌期权、LTCM 的相关性假设、伦敦鲸的估值法操纵）。
Rogue trading违规交易：交易员越权并控制后台隐藏亏损（巴林李森）。
Static vs dynamic hedge静态/动态对冲：静态一次匹配不调整；动态（滚动）需频繁再平衡，带来融资流动性风险（MGRM 案）。
Contango / backwardation期货升水/贴水：升水（远期价格高于现货）下滚动对冲亏钱；贴水下赚钱。

关键术语：maturity mismatch期限错配、flight to quality避险逃跑、agency risk代理风险、synthetic derivatives合成衍生品、cyber risk网络风险。

🧭 知识网位置前面所有风险类型的"事故数据库"：每个案例都是某章理论的失效现场（利率→S&L；流动性→雷曼；模型→LTCM；治理→安然）。背案例名 → 风险类型 → 一条教训是本章的考试形态。前置 → 第 8 章 · ERM（失效的背景板）本章之后 → 第 10 章 · 2007–09 大危机剖析（把这些教训放大到系统层面） · 第 3 章 · 治理（安然类失败的机制）

9.1 Interest-rate risk → the 1980s U.S. S&L crisis

利率风险 = 利率波动导致损失的可能，其敏感度用久期衡量。储贷危机的机制要能复述：1933 年设立、1966 年扩展到储贷机构（S&L）的 Regulation Q 限制存款利率上限，使 S&L 能以低成本吸存，再用资金发放当时常见的 10 年期固定利率房贷；在 60-70 年代向上的收益率曲线下，短端（负债成本）远低于长端（资产收益），S&L "riding the curve" 稳稳吃利差。1970 年代末通胀飙升，美联储加息推高短端利率——付 6% 吸存、只收 5% 房贷，利差转负、亏损累积；S&L 用降低放贷标准、做高风险贷款来弥补，最终约 35% 倒闭、纳税人 1,600 亿美元救助。教训：管理利率风险可用两招——资产与负债的久期匹配（近似同久期时，利率变动的影响相互抵消），或用衍生品（caps / floors / swaps）对冲。考点提示：给一家银行的资产负债结构判断风险→先看久期错配方向与利率变动方向。

Savings & Loan (S&L) crisis 储贷机构危机

S&Ls funded long-term fixed-rate mortgages with short-term deposits. Regulation Q (extended to S&Ls in 1966) capped deposit rates → cheap funding while the yield curve was upward-sloping ("ride the curve"). Inflation surged in the late 1970s → Fed raised rates → S&Ls paid ~6% on deposits but earned only ~5% on mortgages → negative spread → looser lending standards into risky loans → ~35% failed → $160 billion taxpayer bailout.

Lesson: manage interest-rate risk — match the duration of assets & liabilities, or hedge with derivatives (caps, floors, swaps).

9.2 Funding-liquidity risk → Lehman, Continental Illinois, Northern Rock ⭐

流动性风险 = 无法满足短期现金需求的风险，可来自外部市场、内部运营、资产负债结构或三者混合。三起案例的共同线程：用短期批发融资（回购、同业、外币市场）支撑长期非流动资产，一旦市场停止滚动融资即崩溃。雷曼 1850 年成立，重仓证券化房地产、把大量证券化资产留在表内；2007 年杠杆（资产/权益）达 31:1，核心融资靠每日回购；房价 2006 年起下跌它仍在加码；2007 下半年起投资者拒绝短期放款，2008 年 9 月 15 日破产——150 年老店死于流动性风险管理失败。大陆伊利诺伊：1981 年能源市场下行、Penn Square 银行倒闭，它持有约 10 亿美元关联贷款致巨亏；融资靠美联储短借、CD 与海外（如日本）货币市场；1984 年 5 月海外市场得知其困境后无法再融资，储户 10 天抽走约 20% 活期存款，随后倒闭、获当时美联储史上最大救助——"太大不能倒"一词据说由此而来。北岩：英国以 20%/年速度扩张的按揭银行，采用 OTD（发起-分销）模式、依赖美欧亚的短期批发市场；2007 年初违约率上升传导到全球证券化市场、融资冻结；恰在事发前英国监管还给了 Basel II 豁免允许其增派股息；英格兰银行的救助消息泄露引发零售挤兑（当时存款保险仅 £2,000 全额 + 至 £33,000 的 90%），最终被国有化。教训与延伸（要记）：危机后美国监管（尤其美联储）开始强制大型银行定期做流动性压力测试；管理融资流动性 = 优化融资来源与构成，通过资产负债的契约期限管理（直接或经利率互换等衍生品）实现。ALM 的两个内生权衡：① 利率风险 vs 流动性风险——短债融长资产降利率风险但升流动性风险；反之久期匹配降流动性风险但重定价机会少、利率风险可能更高；② 成本 vs 风险缓释——拉长融资期限有效但更贵。另外：理论上可缩短资产久期，但受借款人需求与竞争约束，通常不现实；因此银行必须维持应急流动性缓冲（高流动性资产或已承诺的信贷额度，均有机会成本），把 ALM 纳入整体资产负债管理。

Lehman Brothers (2008)

Founded 1850; sourced, repackaged & sold securitised real-estate assets, retaining large portions on its balance sheet — even after housing prices fell in 2006. Leverage (assets/equity) 31:1 by 2007; funded long, illiquid assets with daily repo borrowing. When short-term lenders stopped rolling, bankruptcy on Sept 15, 2008. Root cause: unmanaged funding-liquidity risk.

Continental Illinois (1984) — "too big to fail"

Held ~$1bn of Penn Square-linked energy loans. Funded via Fed borrowing, CDs & foreign money markets. When news hit foreign markets (May 1984), depositors withdrew ~20% of demand deposits in 10 days; largest Fed rescue to date. Some say this coined "too big to fail".

Northern Rock (2007, UK) — a classic run on wholesale-funded lending

GreW 20%/yr on an OTD model funded by short-term wholesale money (US/Europe/Asia). When securitisation markets froze, funding couldn't be rolled. A leaked Bank of England rescue triggered a deposit run (UK insurance: £2,000 + 90% to £33,000). Ended in public ownership. Post-crisis response: mandatory liquidity stress testing for large banks (US Fed).

ALM trade-offs 资产负债管理的两个权衡

Interest-rate risk vs liquidity risk: funding long assets with shorter liabilities ↓ IR risk but ↑ funding-liquidity risk; duration-matching ↓ liquidity risk but may ↑ IR risk (fewer repricing opportunities).
Cost vs risk mitigation: lengthening funding maturities reduces liquidity risk but costs more.

Shortening asset maturities is usually impractical (borrower needs, competition) → banks maintain emergency liquidity buffers (liquid assets or committed credit lines — both carry opportunity costs).

9.3 Hedging strategy → Metallgesellschaft (MGRM) ⭐

先立概念：静态对冲＝买入与敞口高度匹配的工具、部署时测量一次关系、之后不调整，关注到期时的结果——监督需求低、交易成本低，但不会跟踪敞口变化；动态对冲＝持续再平衡（日/月/季），更贴合变化的市场，但需要持续监督、交易成本高、还多一层模型风险。滚动对冲是动态对冲的典型：用 1 个月期货对冲长期敞口，每月到期前平仓再买入下一批，循环至到期。MGRM 案例：1993 年它向客户提供 5 年或 10 年固定价格的取暖油/汽油合约（固定价 = 未来 12 个月合约均价之上 $3-5/桶；客户若现货高于合约价可取差价的一半作为选择权）；这给公司一个长期远期空头敞口，它用短期限货的滚动（stack-and-roll）多头对冲——因为长期限合约流动性极差。1993 年秋其无铅汽油合约未平仓量达 5,500 万桶，而日均成交量仅 1,500–3,000 万桶（体量本身即风险）。 关键机制：远期合约损益到期才实现，期货每日盯市——客户合约是 5–10 年的长周期，期货却天天产生现金流。1993 年末油价急跌，出现 13 亿美元保证金追缴（未实现损失），公司没有现金应对（因为客户合约全是长期性质），母公司下令平掉全部对冲头寸——把未实现亏损变成已实现亏损。经济学上许多学者认为该对冲策略是基本稳健的（若扛得住期间现金流，远期与期货现金流长期会平衡），MGRM 的死因是现金流/流动性：没能在保证金压力下坚持。滚动对冲还有三道坎：① 会计规则——按德国/IFRS 口径，期货损失必须当期报告、客户合约的对应收益不许确认，收入表巨亏会伤信用评级；② 交易流动性风险——平掉大仓位需要时间（MGRM 用了 10 天）以控制市场冲击；③ 税务处理跨辖区不一，税后结果可能与税前大不同。

Static vs dynamic hedge 静态 vs 动态对冲 ⭐

	Static	Dynamic (rolling)
Mechanism	Match the hedge to the exposure once, at deployment	Rebalance frequently (daily/monthly/quarterly); e.g., roll 1-month futures
Supervision/costs	Low supervision, low transaction costs	Ongoing supervision; higher transaction costs; extra model risk
Weakness	Doesn't adjust to changing exposures	Creates funding & trading liquidity risk when long liabilities are hedged with short derivatives

MGRM (1993)

Sold fixed-price 5–10yr supply contracts (~150m barrels of gasoline/heating oil; price = avg. futures + $3–5), creating a long-term forward short position; hedged with rolling short-dated futures (stack-and-roll) — open interest 55m barrels vs 15–30m daily volume. Oil fell sharply in late 1993 → $1.3bn margin call; no cash to cover (customer contracts were long-term) → parent ordered hedges closed → unrealised loss became realised.

Curve regimes ⭐ Backwardation 现货升水/期货贴水: spot > futures → rolling hedge earns (sell existing at profit vs next month's cost).
Contango 期货升水: futures > spot → rolling hedge loses when rolling.

Lesson: many economists consider the strategy fundamentally sound — MGRM's undoing was a cash-flow problem (margin calls, credit, funding liquidity). Three extra considerations: accounting rules (futures losses reported, customer-hedge gains not), trading liquidity (10 days to unwind), and tax treatment differences across jurisdictions.

9.4 Model risk → Niederhoffer, LTCM, London Whale 模型风险 ⭐

模型风险的五种形态先记名：用错模型、模型设定错误、数据不完整、用错估计量、假设错误。三个案例各展示一个侧面。Niederhoffer：他卖出巨量深度虚值的标普 500 看跌期权收权利金——只要日跌幅小于 5% 就稳定小赚，按正态分布假设 5% 单日跌幅"近乎不可能"；1997 年 10 月亚洲危机外溢，标普单日暴跌 7%，5,000 万美元保证金追缴无法满足，全部仓位被强平、基金权益清零——教训是假设可能错，竞争市场没有免费的午餐。LTCM：1994 年成立，合伙人有前美联储副主席、两位诺奖得主与所罗门债券套利老将；破产前 48 亿美元权益对 1,250 亿美元资产（25:1 杠杆），名义敞口超过 1 万亿——贷款机构因合伙人声望常常豁免初始保证金，放大了杠杆。核心策略是相对价值（价差收敛）：买一资产、空另一资产（如买英国公司债空英国国债；买意/西主权债空德国 bund），理论上降低波动与系统性风险。1998 年 8 月俄罗斯意外违约并贬值，引发避险逃跑（flight to quality）——投资者涌入的恰是 LTCM 做空的国债与 bund，一个月内权益损失超 40%（48 亿中的 20 亿），几乎引发系统性崩溃，纽约联储撮合银行团注资 35 亿美元换 90% 股份与完全管理权。死因是模型错误：没有预见全球危机中相关性的飙升（他们调高过相关性，但远不及实际），也没有预警真实的波动率——模型错导致流动性危机，最终摧毁公司。五条教训：监测相关性（地理分散化在危机中失效；模型假设低频高损事件跨期不相关——错）；盯流动性（极高杠杆+模仿者同向平仓放大市场冲击）；反思损失假设（10 天期 VaR 计算值 3.2 亿美元，实际损失超 10 亿——短窗口不足以扛住冲击，应做压力测试）；加强披露（对冲基金披露要求低）；初始保证金无例外（若贷方坚持保证金要求，不会有如此杠杆，也有缓冲垫）。伦敦鲸：2012 年摩根大通 CIO 管理 3,500 亿美元超额存款，用合成信用衍生品大规模下注；2011 年该行为满足资本要求本想降险，领导层却加倍做多去"抵消"空头——反而扩大风险追求利润；2012 年亏损出现后，领导层改估值方法：最佳实践是用日内区间中点定价，他们却从区间内挑价（cherry-picking）让仓位显得更安全；3 月被自家投行发现两个部门估值口径不一；5 月副 CRO 令其回归中点法并第三方验证，账面与经济损失开始显性化。1-4 月间 CIO 共违反自身 VaR 限额 330 次——系统性忽略或上调限额；还修订 VaR 假设把计算值压掉 50%；最终损失 62 亿美元。教训：当限额被突破、交易难看时，绝不允许调整假设或估值模型让坏决策显得好看。

Niederhoffer (1997)

Wrote large volumes of deep OTM S&P 500 puts, capturing small premiums as long as the daily drop was <5%. Oct 1997: Asia crisis spillover → S&P fell 7% in one session → $50m margin call unmet → positions liquidated → equity wiped out. Lesson: assumptions can be flawed; there is no free lunch.

Long-Term Capital Management (LTCM, 1998)

$4.8bn equity / $125bn assets (25:1); notional > $1tn. Relative-value convergence trades; lenders often waived initial margin on reputation. Aug 1998 Russia defaulted & devalued → flight to quality → the very assets LTCM shorted (Treasuries, bunds) rallied → –40%+ of equity in a month → NY Fed brokered a $3.5bn bank rescue (90% of shares, full management control).

Lessons: monitor correlations (they spike in crises; low-frequency/high-severity events are NOT uncorrelated); watch liquidity & leverage (imitators liquidating together amplified market impact); question loss assumptions (10-day VaR said $320m; actual loss > $1bn → stress test instead); enhance disclosure; require initial margin without exception.

London Whale (JPMorgan CIO, 2012)

$350bn of excess deposits; massive synthetic-credit bets. Instead of reducing risk, leadership doubled down long to offset shorts — expanding risk while chasing profit. Valuation switched to cherry-picked prices (vs midpoint) to look safer; CIO breached VaR limits ~330 times (Jan–Apr 2012) and revised VaR assumptions to cut calculated VaR by 50%. Final loss: $6.2 billion.

Lesson: when limits are breached or trades look unprofitable, never adjust assumptions or valuation models to make bad decisions look better.

9.5 Rogue trading & misleading reporting → Barings

巴林银行 1762 年成立于伦敦、世界第二古老的商人银行。1992 年李森（Nick Leeson）赴新加坡任运营主管，职责从执行客户交易扩展到自营套利——理论上应做"买低估、卖高估"的对冲套利，他却做成单边方向性下注（只买不空）。会计上看他在"大赚"：因为他还控制自己交易的后台记账，用秘密对账账户上报——1994 年表面盈利 1.02 亿英镑，实际亏损约 2 亿英镑。1994 年末风险经理对其反常收益起疑，但无权挑战被超额奖金安抚的李森主管层；1995 年 1 月李森单月"盈利"1,000 万英镑再次被压下；等高层明白真相为时已晚，亏损大到银行被迫清算，ING 以1 英镑收购——200 年老店毁于一个不受制衡的交易员。教训要背四条：前台与后台必须分离监督；好得反常或异常稳定的盈利要有健康的怀疑并核查账簿；要有独立验证机制、后台绝不能被前台影响者控制；按整体策略的最终结果评价交易，而不是挑时段看。

Barings Bank (1995) — Nick Leeson

Leeson ran Nikkei futures "arbitrage" in Singapore but made unidirectional bets and controlled the back office, hiding losses via a secret reconciliation account ("£102m profit" = £200m loss). Warnings came (suspicious gains, £10m in one month) but were dismissed; risk managers couldn't override supervisors fattened by bonuses. Barings — the world's 2nd-oldest merchant bank — was sold for £1 (to ING).

Lessons: segregate front/back-office oversight; treat too-good/strangely-consistent profits with scepticism; independent verification; judge trades by ultimate outcome, not cherry-picked periods.

9.6 Financial engineering → Bankers Trust, Orange County, Sachsen 金融工程 ⭐

金融工程的积木：远期、期货、互换、期权与证券化产品——可对冲单一敞口或一篮子敞口（例如投资日本固收的美国基金同时有日本利率风险与汇率风险，可用远期/互换只对冲其一，甚至用 quanto swap 这类多币种利率互换）。关键警示：要对冲就要认清目标——是缓释风险，还是"借对冲之名增厚收益"；后者往往是在叠加风险而非缓释。Bankers Trust（BT）：90 年代初以风控见长的 BT 帮宝洁（P&G）管理美元与马克利率风险，结果演变成20:1 杠杆的复杂互换押注利率下行（BT 付固定、P&G 付浮动）；1994 年美联储加息 250 个基点，P&G 巨亏——因为它选择了当投机者而不是对冲者，杠杆又放大了损失；P&G 起诉 BT 并获 7,800 万美元判决；其他公司类似遭遇使 BT 声誉损失难以挽回，最终被德意志银行收购。奥兰治县：1994 年该县资产 77 亿美元，财务主管 Robert Citron 用短期回购（repo）再借 129 亿美元，全部投入反向浮息票据（利率上升时票息下降），并且完全依赖 repo 到期滚动；起初比同行多赚 2%；美联储加息 250 个基点后票息下跌、投资人拒绝展期，最终破产（约 16 亿美元损失）；Citron 后来承认他其实并不理解反向浮息的敞口——教训：不懂的东西不要投，损失可能是终局性的。萨克森州立银行：德国州立银行体系中的一员，2000 年代中期看上次贷利润，在都柏林（爱尔兰）设立表外载体买入大量次贷资产；2007–09 危机袭来、亏损惨重，被迫卖给一家未涉足次贷证券化的德国银行——教训：表外载体能隐藏真实的信用与流动性风险。

Bankers Trust & P&G (1994)

P&G asked BT (a skilled risk manager) to manage USD/DM interest-rate risk; it ended up in 20:1 leveraged swaps betting on falling rates. The Fed raised rates by 250bp (1994) → substantial losses — speculating, not hedging. P&G sued; won $78m. Similar cases with other firms inflicted lasting reputational damage; BT was eventually acquired by Deutsche Bank.

Orange County (1994)

$7.7bn portfolio + $12.9bn borrowed via short-term repos (rolled at maturity — full reliance on lenders), invested in inverse floaters (coupons fall when rates rise). Initially +2% vs peers; Fed hikes (+250bp) → coupons fell & repos weren't rolled → bankruptcy (~$1.6bn loss). Citron admitted he didn't understand the exposures. Lesson: don't invest in what you don't understand — losses can be terminal.

Sachsen Landesbank (2007)

A German Landesbank that set up an off-balance-sheet entity in Dublin to hold securitised US subprime assets. Heavy losses in the crisis → had to sell itself to a German bank that had avoided subprime. Lesson: off-balance-sheet vehicles can hide real credit/liquidity risk.

9.7 Reputation risk → Volkswagen

声誉 = 公众对公司公平性、道德行为与对待利益相关方的看法；声誉风险 = 因负面事实或传闻导致负面经营结果的可能。三个关键利益相关方：客户、监管者、股东。ESG 监测是声誉影响日益上升的领域；互联网时代事实与传闻扩散极快——坏传闻数小时就能暂时摧毁声誉，传闻属实的情况伤害最大。大众案例：2015 年 9 月美国环保署（EPA）公布大众在环保责任上不道德——给车辆编程让排放控制只在监管测试时生效，测试全优、日常使用却超标；2009–2015 年间影响全球超 1,000 万辆汽车；丑闻发酵中股价被砍掉三分之一，面临数十亿美元罚款叠加销量下滑（消费者转投其他品牌）；德国政府一度担忧"德国制造"这块招牌受损。教训：声誉风险常由其他风险（欺诈/违规）触发，且直接打击企业价值。

Volkswagen "Dieselgate" (2015)

Software controlled emissions only during regulatory tests (2009–2015, >10m vehicles). Share price –1/3; billions in fines + lost sales; even "Made in Germany" reputation was questioned. Lesson: reputation risk (loss of trust) can be triggered by another risk (fraud) and severely harms firm value — watch customers, regulators, shareholders.

9.8 Corporate governance → Enron

公司治理 = 指导公司如何运转的政策与流程体系；检查点包括：足够的透明度与问责、对高管与风险管理政策的监督、合理的多元化、董事会独立性、以及匹配公司所需技能组合的董事。安然是治理失败的"教科书"：1985 年 InterNorth 与 Houston Natural Gas 的高杠杆合并诞生安然，放松管制后它成为天然气经纪商，并创建了能源衍生品市场来覆盖气价风险——1995–2000 年被《财富》评为"美国最具创新力公司"，2000 年底有 2 万名员工、记录收入近 1,010 亿美元。2001 年 12 月它以当时美国史上最大破产收场，直接原因是治理失败与代理风险（agency risk）：① 代理风险——高管以自我利益优先，追逐短期利润最大化个人财富、牺牲整个公司；② 董事会监督缺失——董事长与 CEO 由同一人担任（自我监督），董事会还放任 CFO 经营自己的私募股权公司（实为向其个人输送资金的渠道）；③ 会计欺诈——用 SPV 造假销售、把亏损藏进表外载体；④ 收入确认——资产一完工就确认收入（无论是否真卖出），真出售时的亏损埋进 SPV；⑤ 审计失败——五大会计师事务所之一的安达信是安然唯一审计师，国会调查认定其卷入欺诈、吊销执业资格——安然与安达信一同倒下。结局催生 2002 年SOX 法案：要求主要高管对披露财报的可靠性负责，并创建 PCAOB（公众公司会计监督委员会）整体提升治理标准。教训：CEO 与董事长应分设；董事会必须监督管理层；警惕激进会计；独立审计师须制衡管理层而非只收费用。

Enron (2001)

Gas broker turned energy-derivatives pioneer ("America's Most Innovative Company" 1995–2000; ~$101bn recorded revenue, 20,000 employees). Dec 2001: largest U.S. bankruptcy at the time. Failures: agency risk; chairman = CEO (self-supervision); CFO's private equity side business; SPV accounting fraud hiding losses; revenue recognised when assets were built; auditor Arthur Andersen complicit, practice revoked. Result: Sarbanes-Oxley (2002) + PCAOB.

Lessons: separate CEO & chairman; board must supervise; scrutinise aggressive accounting; independent auditors must hold management accountable.

9.9 Cyber risk → SWIFT

网络风险 = 内部技术基础设施被攻破导致财务或声誉损失的风险——黑客入侵系统盗取资金、信息或身份数据；做数字交易的任何机构都相关，缓释途径：内部加强、聘外部 IT 顾问、或购买网络保险外包损失风险。SWIFT（环球银行金融电信协会）是全球金融机构间电子资金转移的领导者，由比利时、美国、英国、欧洲央行、日本等央行组成的财团监管，每天处理约 2,500–3,500 万笔交易。2016 年 2 月黑客用员工凭证入侵 SWIFT 系统（凭证如何获得始终未完全查明），从孟加拉央行在纽约联储的账户发起一系列转款请求、盗走 8,100 万美元——目标本是 10 亿美元，因一笔请求里的拼写错误（"fandations" 而非 "foundations"）引起纽约联储警觉、拦下后续转账。资金转入菲律宾一家银行后，黑客用恶意软件删除转账记录并关闭确认通知；孟加拉央行因长期收不到通知而重启系统，瞬间涌入大量转账通知才东窗事发。资金经菲律宾银行转至多间赌场并被提走，始终未追回。教训：网络攻击的筹码极高——面具与武器不再是抢劫银行的必需品。

SWIFT heists (2016)

Stolen employee credentials → fraudulent transfer requests; $81m taken from Bangladesh Bank's NY Fed account (goal $1bn; a typo "fandations" stopped the rest). Malware deleted transfer records & disabled confirmations; funds to the Philippines → casinos → never recovered. SWIFT handles 25–35m transactions daily. Lesson: cyber risk — financial/reputational loss from breaching internal technology infrastructure; mitigate internally, via IT consultants, or cyber insurance.

9.10 Exam cheat-sheet: name → risk → lesson 案例速查表 ⭐

这是高频"名称→风险→教训"速查表。记忆法：利率→S&L；融资流动性→雷曼/大陆/北岩；对冲→MGRM；模型→Niederhoffer/LTCM/伦敦鲸；违规交易→巴林；金融工程→BT/奥兰治县/萨克森；声誉→大众；治理→安然；网络→SWIFT。考试直接给案例考归类与对应教训，务必把每个名字和"风险类型 + 一条核心教训"绑死。

Case	Risk / factor	Key lesson
S&L crisis (1980s)	Interest-rate risk	Match duration / hedge rates
Lehman / Continental Illinois / Northern Rock	Funding-liquidity risk	Short-term funding of long assets = danger; deposit runs; ALM trade-offs
Metallgesellschaft	Hedging strategy (dynamic/rolling)	Match hedge horizon; manage margin cash flows; contango losses; accounting & tax frictions
Niederhoffer / LTCM / London Whale	Model risk	Assumptions fail; correlation spikes; VaR horizon limits; never game limits/valuation
Barings	Rogue trading / misleading reporting	Segregate front/back office; scepticism on too-good profits
Bankers Trust / Orange County / Sachsen	Financial engineering	Hedge ≠ speculate; don't invest in what you don't understand; off-balance-sheet vehicles hide risk
Volkswagen	Reputation risk	Loss of trust damages value; customers/regulators/shareholders
Enron	Corporate governance	Separate CEO/chair; independent audit; agency risk; SOX aftermath
SWIFT	Cyber risk	Protect payment/IT infrastructure; insurance option

🔑 英文要点速查 · English Essentials

interest-rate risk & duration 利率风险与久期 — S&L: short deposits funding 10-yr fixed mortgages; rate hikes → negative spread → 35% failed, $160bn bailout; fix by duration matching or caps/floors/swaps.
funding-liquidity risk 融资流动性风险 — Lehman (31:1 leverage, daily repo), Continental Illinois (~20% deposits withdrawn in 10 days; "too big to fail"), Northern Rock (wholesale-funded OTD; leaked rescue → retail run).
ALM trade-offs ALM 权衡 — IR risk vs liquidity risk; cost vs mitigation; maintain emergency liquidity buffers (liquid assets / committed lines).
static vs dynamic hedge 静态 vs 动态对冲 — Static: set once, low cost, doesn't track exposure. Dynamic: rebalance, flexible, costlier, model risk; rolling futures create margin/liquidity needs.
contango / backwardation 期货升水/贴水 — Contango (futures > spot) loses money when rolling; backwardation (spot > futures) earns.
MGRM lesson — Fundamentally sound hedge can die on cash-flow needs: $1.3bn margin call, accounting (losses shown, gains deferred), 10-day unwind, taxes.
model risk 模型风险 — Wrong model / specification / data / estimators / assumptions; Niederhoffer's 5% assumption; LTCM's correlation & VaR assumptions.
LTCM — Relative-value trades + extreme leverage; flight to quality after Russia 1998; –40% equity in a month; $3.5bn rescue; monitor correlations & liquidity; initial margin without exception; disclose more.
London Whale — Doubling down instead of de-risking; cherry-picked valuations; ~330 limit breaches; VaR assumptions cut 50%; $6.2bn loss.
rogue trading controls 违规交易控制 — Barings: separate back-office oversight; scepticism of too-good results; judge by ultimate outcome.
financial engineering 金融工程 — Forwards/futures/swaps/options/securitised products; Bankers Trust (speculating not hedging; $78m to P&G), Orange County (inverse floaters, repos, bankruptcy), Sachsen (off-balance-sheet subprime, Dublin).
reputation risk 声誉风险 — Public perception; customers, regulators, shareholders; VW: test-cheating software, –1/3 share price, >10m cars.
corporate governance 公司治理 — Enron: agency risk, chairman=CEO, SPVs, Arthur Andersen; result: SOX + PCAOB.
cyber risk 网络风险 — SWIFT: $81m from Bangladesh Bank; malware deleted records; mitigate/insurance.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. Northern Rock most directly illustrates which risk faced by financial institutions?

A. Market risk from marked-to-market securities.

B. Funding-liquidity risk from heavy reliance on short-term wholesale markets. ✅

C. Model risk from mis-specified valuation models.

D. Cyber risk from breached IT infrastructure.

B 正确。考点：北岩 = 融资流动性风险的标本
北岩用美欧亚短期批发市场为 OTD 模式（放贷→证券化→卖出）融资；证券化市场冻结后融资无法滚动，救助消息泄露又引发零售挤兑——教科书式的期限错配。A/C/D 均无依据。

Q2. Which statement about static and dynamic hedging strategies is MOST accurate?

A. Dynamic hedges require no ongoing supervision.

B. Static hedges continuously adjust to changes in the underlying exposure.

C. A rolling (dynamic) hedge can create funding-liquidity problems even when the hedge is economically sound over its life. ✅

D. Static hedges incur higher transaction costs than dynamic hedges.

C 正确。考点：静态 vs 动态对冲 + MGRM 的核心教训
MGRM 的对冲策略长期看基本稳健（远期与期货现金流最终会平衡），但它扛不住短期保证金追缴——死在现金流/流动性。A 反了（动态需持续监督）；B 反了（静态一次性匹配、不调整）；D 反了（静态交易成本更低）。

Q3. Which of the following is NOT a lesson from the LTCM collapse?

A. Short-term VaR parameters are always the best risk management tool. ✅（这句话本身就是错的）

B. Lenders should impose initial margin requirements without exceptions.

C. Position liquidity should be carefully monitored, especially under extreme leverage.

D. Correlations must be monitored for convergence during stress periods.

A 正确（它不是教训，且违背教材）。考点：LTCM 五条教训的逆向识别
LTCM 用 10 天期 VaR 算出 3.2 亿美元"安全垫"，实际损失超 10 亿——教材明确 10 天窗口太短，压力测试才是更好的方法。B（保证金无例外）、C（流动性监测）、D（相关性在压力期飙升）都是标准教训。

Q4. A board wants to reduce the chance of an Enron-style governance failure. Which action is LEAST likely to help?

A. Separating the roles of chairman and CEO.

B. Requiring independent auditors to hold management accountable.

C. Relying on continuous communication between the CEO and CFO to prevent fraud. ✅

D. Scrutinising aggressive accounting techniques.

C 正确（它最不可能有帮助）。考点：安然的治理教训
安然的教训恰恰是：不能把监督寄托在管理层自身（董事长兼 CEO = 自我监督；CFO 兼职私营业务无人制约）。治理要靠结构——职位分设、独立审计与董事会监督、对激进会计的警惕。A/B/D 都是标准教训，"CEO-CFO 多沟通"不是教材认可的控制手段。

2007–09 大金融危机剖析

10. Anatomy of the Great Financial Crisis of 2007–2009

学习单元 3 · 要点：危机酝酿链条 · 次贷特征（2–28 ARM / 100% LTV / NINJA / liar）· CDO 瀑布与 CDO² · 机构激励错位 · 短期批发融资（ABCP / repo）与 systemic risk · haircut 与 LIBOR-OIS 两个压力指标 · 央行与政府应对（贴现窗口扩至投行 / TAF / PDCF / TARP）

🇨🇳 中文导览

本章把危机拆成"酝酿 → 次贷与 CDO → 机构角色 → 短期融资与系统风险 → 政策应对"五段。核心链条：低利率 → 房价不可持续上涨 → 证券化（OTD）让放贷者不再承担信用风险 → 放贷标准放松 → 次贷繁荣 → 银行/SIV 用短期批发融资（ABCP、回购）支持长期资产 → 期限错配 → 房价下跌后融资断流、市场冻结。两个必记的压力指标与四个必记的政策工具是本章最"硬"的考点。

Subprime次级贷款：借款人信用差（逾期史/高 LTV/高收入负债比）；2–28 ARM（前 2 年低诱惑利率）、100% LTV（无首付）、只付息、NINJA（无收入无工作无资产）、liar loans（收入不核）。
OTD & SIV发放-分销 / 结构投资载体：贷款不再留表，转入破产隔离的 SIV，经证券化切成 CDO 分档出售 → 银行资本占用下降、放贷标准放松。
CDO waterfall瀑布式分层：senior（AAA，先受偿、后承担损失）→ junior → equity（先吸收损失）；junior 再打包 = CDO²，不透明且难估值。
机构激励错位：银行放松标准；经纪商按数量而非质量收费；评级机构给次贷支持的高级档 AAA、依赖发行人数据、发行人付费（利益冲突）；投资者追逐收益率却无法估值 CDO。
短期融资：ABCP（有资产抵押的商业票据）与 repo（回购，以证券为抵押）。房价下跌 → 质疑 SIV 资产质量 → 拒绝展期 → 2007 年 8 月 ABCP/回购市场完全停摆。
两个压力指标：haircut 从危机前的约 0% 升至 2008 年 9 月雷曼违约后的近 45%；LIBOR-OIS 利差从近 0% 升至峰值 3.6% 以上——利差越高，感知信用风险越高、同业越不愿放款。
政策工具：美联储降息 + 流动性支持；把贴现窗口（历史上只供商业银行日常准备金）开放给投行；TAF（定期拍卖工具，向存款机构供资）、PDCF（一级交易商信贷工具，经回购向一级交易商放款）、2008 年 9 月房利美/房地美 bailout、2008 年 10 月 TARP（问题资产救助计划，购买金融机构不良资产）；央行资产负债表大幅扩张（QE）。

关键术语：haircut折扣率/保证金折扣、LIBOR-OIS spreadLIBOR 与隔夜指数互换利差、deleveraging去杠杆、systemic risk系统性风险。

🧭 知识网位置第 9 章是个体案例的"点"，本章是系统层面的"面"——把 OTD、CDO、期限错配、激励错位全部串进一个事件链。回收一条主线索：第 4 章的信用风险转移机制如何在系统层面失控。前置 → 第 4 章 · 信用风险转移机制 · 第 9 章 · 雷曼等案例本章之后 → 第 11 章 · 职业行为准则（危机后的行为反思）

10.1 The build-up 危机的酝酿

酝酿四步链条：① 美国利率长期处于历史低位 → 借钱便宜 → 房地产需求推动房价不可持续地上涨；② 证券化让房贷可以被发起、打包、卖给追逐收益的投资者，发起机构不再承担信用风险 → 不再关心借款人信用质量 → 放贷标准放松 → 次贷因"低息环境下的高收益"日益流行；③ 与以往美国危机不同，始于次贷的问题迅速蔓延到其他资产类别与地区，最终波及全球市场；④ 银行越来越依赖短期负债支持长期资产 → 资产与负债久期错配 → 重大流动性风险；房价停涨时，短期负债无法滚动。高潮（2008 年 9 月）：雷曼破产 → 信心崩塌、同业拆借市场冻结；各大机构间失去信任、囤积超额准备金不愿外借；贝尔斯登与美林被收购、摩根士丹利与高盛转为银行控股公司（接受美联储监管）、房利美与房地美被国有化、AIG 被救助以防系统性扩散。

Low interest rates → cheap borrowing → rapid, unsustainable house-price increase.
Securitization: mortgages originated → repackaged → sold to yield-seeking investors → originators bore less credit risk → relaxed lending standards → subprime 次级贷款 boomed (higher yields in a low-rate world).
Unlike previous U.S. crises, the subprime problem escalated to other asset classes and geographies — global impact.
Banks financed long-term assets with short-term liabilities → maturity mismatch → funding-liquidity risk; when house prices stalled, short-term liabilities could not be rolled over.

Climax (Sept 2008): Lehman Brothers bankruptcy → loss of confidence → interbank market froze; institutions hoarded reserves. Bear Stearns & Merrill sold; Morgan Stanley & Goldman became bank holding companies; Fannie Mae & Freddie Mac nationalised; AIG bailed out.

10.2 Subprime mortgages & CDOs 次级房贷与 CDO

次贷的正式定义：以住宅为抵押、借给信用差借款人的贷款——可能有逾期记录、高 LTV（低首付）或高贷款收入比。典型结构是 30 年期 2–28 ARM：前 2 年相对低的固定诱惑利率，之后 28 年转成可能高到无法承受的浮动利率；还叠加 100% LTV（无首付、无权益缓冲）、只付息（本金不降）、NINJA、liar loans（就业与收入未核实）。借款人与炒房者指望房价续涨后转按揭或卖出获利；房价下跌后他们陷入负权益（贷款额超过房屋价值）选择违约离场 → 法拍增加 → 供给过剩 → 房价进一步下跌（恶性循环）。CDO 结构：贷款池分档（senior / junior / equity），按瀑布分配现金流与损失——senior 先拿钱、最后承担损失；即便底层是 NINJA 与 liar 贷款，senior 档仍被打造成 AAA；多个 CDO 的 junior 档再打包成 CDO²——极度不透明、正常时期都难估值，连老练投资者也看不懂自己买的是什么。评级机构问题：给出不现实的高评级（基于优质房贷的历史数据、忽视市场日益投机化）、过度依赖发行人数据且不自查、发行人付费产生利益冲突。

Subprime mortgage features 次贷特征

2–28 ARM — 2-yr low "teaser" rate, then a much higher variable rate for 28 years.
100% loan-to-value — no down payment → no equity cushion for the lender.
Interest-only — no principal reduction during the life of the mortgage.
NINJA loans — no income, no job, no assets.
"Liar loans" — little evidence collected to confirm employment/income.

When prices fell: negative equity → walk away & default → foreclosures → oversupply → prices fell further (vicious spiral).

The CDO waterfall CDO 瀑布式现金流结构

Loans pooled and sliced into tranches — senior, junior, equity. Cash flows & defaults follow the waterfall: senior (AAA-designated; receives cash flows first, absorbs losses last), junior, equity (paid last, absorbs losses first). Senior tranches held AAA ratings despite NINJA/liar collateral. Junior tranches of multiple CDOs were rebundled as CDO-squared — a CDO backed by other CDO tranches; extremely opaque and hard to value.

10.3 Roles of different institutions 各机构角色与激励错位

四类角色，一条主线——激励错位：银行在 OTD 下放松标准、把房贷转进 SIV 以少占资本；按揭经纪与放贷机构按数量而非质量收佣金，向买不起或本可买更便宜产品的人推销次贷（不合适销售）；评级机构给次贷支持的高级档 AAA、依赖发行人数据、被发行人付费；投资者与中介在低利率下追逐收益、买入复杂难估的产品、无法给 CDO 定价。考试常给一段行为让你判断属于哪类机构、错在何处。

Institution	Role in the crisis
Banks	Relaxed lending standards under OTD; moved mortgages into SIVs → less regulatory capital held
Mortgage brokers & lenders	Compensated by quantity, not quality → pushed unsuitable subprime products (higher commissions); suitability ignored
Rating agencies	Unrealistically high ratings (AAA on senior tranches); relied on issuer data without own checks; conflict of interest (paid by issuer)
Investors / intermediaries	Chased yield in opaque, complex products; could not value CDOs even in normal times

10.4 Short-term wholesale funding & systemic risk 短期批发融资与系统性风险 ⭐

两种短期融资工具要精确区分：商业票据（CP）是短期无担保融资，主要供高质量发行人使用，本质假设是到期能滚动续发；ABCP（资产支持商业票据）是特殊情形——以信用卡贷款、房贷等资产作抵押。回购（repo）：银行卖出一项资产并同时约定未来以略高价买回，价差就是这段融资的利息（repo rate）；repo 是有担保的短期借款——卖的资产即抵押品，到期不还，贷款方可不经破产法院直接保留或出售抵押品；抵押品可以是国债或高质量公司债，依质量设定折扣率（haircut）——例如 100 美元的资产愿付 90 美元，haircut 就是 10%。机制：SIV 以 ABCP 与 repo 滚动融资长期资产，因此高度依赖展期能力 → 重大融资流动性风险。房价与 MBS 价格下跌后，贷款方质疑 SIV 内资产质量、不愿续贷，到 2007 年 8 月 ABCP 与 repo 市场完全停摆；为 SIV 提供备用信贷额度（backstop lines）的银行也被拖下水；货币市场基金持有大量 ABCP，价格下跌引发机构投资者挤兑货基，进一步恶化流动性；对冲基金无法滚动债务，被迫抛售 CDO 等资产应对保证金追缴。系统层面的数字证据（必记）：haircut 从危机前约 0% 升至 2008 年 9 月雷曼违约后的近 45%；LIBOR-OIS 利差从危机前近 0% 升至峰值 3.6% 以上。更高的 haircut 与借不到钱迫使机构去杠杆——抛售压低价格、侵蚀权益，机构被迫求助政府/竞争者，或以破产收场。这就是系统性风险：由于（如资产负债久期错配等）脆弱性，导致整个金融市场停摆的系统失败风险。最终教训一句话：就算银行自认资本充足，过度依赖短期融资也极其危险——这种资金会在危机中一夜蒸发。

Funding instruments 融资工具对照

Commercial paper (CP) — short-term, unsecured; high-quality issuers; relies on roll-over at maturity.
Asset-backed commercial paper (ABCP) — CP backed by collateral (credit-card loans, mortgages).
Repo — sell an asset & agree to repurchase at a slightly higher price (the difference = repo rate); secured: the asset is collateral the lender may keep/sell without bankruptcy courts; haircut set by collateral quality (pay $90 for a $100 asset → 10% haircut).

Stress metrics & the deleveraging spiral 两个压力指标

Haircuts: ~0% pre-crisis → nearly 45% by Sept 2008 (post-Lehman).
LIBOR-OIS spread: ~0% pre-crisis → over 3.6% at the peak — higher = higher perceived credit risk & reluctance to lend interbank.

Higher haircuts + inability to borrow → forced deleveraging → lower prices → eroded equity → government/competitor help or bankruptcy. This is systemic risk — system failure from vulnerabilities like asset-liability maturity mismatch.

Lesson: even a bank with sufficient capital cannot rely heavily on short-term funding — it can disappear overnight in a crisis.

10.5 Central-bank & government responses 央行与政府应对 ⭐

美联储的路数（五条动作清单）：提供以高质量抵押品担保的长期贷款；允许投行与证券公司经贴现窗口直接从美联储借款（危机前投行没有这个通道——这是最具标志性的变化）；以高质量非流动资产为抵押提供流动性；提供资金购买 ABCP；收购房利美与房地美发行的资产。这些动作导致全球央行资产负债表大幅扩张（量化宽松）。具体工具（必记四个）：TAF 定期拍卖工具——向存款机构提供资金；PDCF 一级交易商信贷工具——美联储经回购向一级交易商放款；2008 年 9 月房利美与房地美救助；2008 年 10 月起的 TARP 问题资产救助计划——从金融机构购买不良资产。考试角度：辨析"哪些是美联储做过的、哪些不是"，以及"贴现窗口向投行开放"为何特殊。

Fed actions: long-term loans secured by high-quality collateral; investment banks & securities firms borrow via the discount window (previously unavailable to them); liquidity against high-quality illiquid assets; funding to purchase ABCP; acquiring Fannie/Freddie-issued assets → massive central-bank balance-sheet expansion (QE).

U.S. interventions to know 美国干预工具

TAF (Term Auction Facility) — funds to depository institutions.
PDCF (Primary Dealer Credit Facility) — Fed lending to primary dealers via repos.
Fannie Mae & Freddie Mac bailout — September 2008.
TARP (Troubled Asset Relief Program) — purchased toxic assets from financial institutions starting October 2008.

🔑 英文要点速查 · English Essentials

build-up factors 酝酿因素 — Low rates → cheap credit → unsustainable house prices; securitization → relaxed standards → subprime boom; problem then spread across asset classes & geographies.
subprime features 次贷特征 — 2–28 ARM teaser; 100% LTV; interest-only; NINJA; liar loans; negative equity → defaults → oversupply → deeper price falls.
originate-to-distribute (OTD) 发放-分销 — Mortgages moved to bankruptcy-remote SIVs via securitization; less capital, relaxed underwriting, focus on volume.
CDO & waterfall CDO 瀑布 — Pool sliced into senior/junior/equity; senior paid first, absorbs losses last; AAA ratings on subprime-backed tranches; CDO-squared = CDO of CDO tranches.
rating-agency failures 评级机构问题 — Unrealistically high ratings, issuer-provided data, issuer-pays conflict of interest.
ABCP vs CP vs repo 三种短融工具 — CP: unsecured; ABCP: collateral-backed; repo: secured borrowing via asset sale/repurchase (repo rate; haircut).
rollover reliance 依赖展期 — SIVs funded long assets via ABCP/repos → funding-liquidity risk; markets shut down by August 2007; backstop lines dragged banks; money-market runs; hedge-fund fire sales.
stress metrics 压力指标 — Haircuts ~0% → ~45% (Sept 2008); LIBOR-OIS ~0% → >3.6% peak.
deleveraging spiral 去杠杆螺旋 — Higher haircuts/borrowing refusal → fire sales → lower prices → eroded equity → help or bankruptcy.
systemic risk 系统性风险 — System failure that shuts down the entire financial market, driven by vulnerabilities like asset-liability maturity mismatch.
central-bank toolkit 央行工具箱 — Rate cuts, liquidity support, discount window extended to investment banks, balance-sheet expansion (QE).
U.S. facilities 美国工具 — TAF (depositories), PDCF (primary dealers via repos), Fannie/Freddie bailout (Sept 2008), TARP (toxic assets, Oct 2008).
core lesson — Overreliance on short-term funding is dangerous even with adequate capital: funding can evaporate overnight.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. Under the originate-to-distribute (OTD) model, the originating lender most likely:

A. retains all credit risk until the mortgages mature.

B. holds less regulatory capital and has a stronger incentive to relax underwriting standards. ✅

C. must hold more regulatory capital because SIVs sit on its balance sheet.

D. focuses more on loan quality than loan volume.

B 正确。考点：OTD 的激励结构
OTD 下贷款转入破产隔离的 SIV 并证券化出售 → 信用风险转移出去 → 资本占用下降，同时收费与数量挂钩 → 标准放松。A 说反（风险已转移）；C 说反（SIV 是表外载体）；D 说反（OTD 重数量）。

Q2. Which pair of developments best evidences the seizing-up of short-term funding markets during the crisis?

A. Haircuts rising from about 0% to nearly 45%, and the LIBOR-OIS spread rising from about 0% to above 3.6%. ✅

B. Haircuts rising while the LIBOR-OIS spread narrowed sharply.

C. Haircuts falling and LIBOR-OIS holding steady near zero.

D. Repo rates falling to zero as collateral quality improved.

A 正确。考点：两个压力指标的方向与含义
haircut 上升 = 抵押品折价加大、放款更谨慎；LIBOR-OIS 走高 = 感知信用风险上升、同业不愿拆借。两者同时恶化正是融资市场冻结的直接证据。B/C/D 都与史实相反或方向矛盾。

Q3. At the peak of the crisis, the Federal Reserve did all of the following EXCEPT:

A. lowering interest rates and providing liquidity support.

B. allowing investment banks to borrow via the discount window.

C. closing the discount window to commercial banks. ✅（它没有这样做）

D. providing funding to purchase asset-backed commercial paper.

C 正确（"对商业银行关闭贴现窗口"不是史实）。考点：贴现窗口的"扩围"而非"关闭"
贴现窗口历史上供商业银行日常准备金使用；危机顶峰的特殊之处是把投行也纳入（B 对）。降息、流动性支持、出资购买 ABCP 均为美联储动作，C 无中生有。

Q4. Which statement best captures the funding lesson of the 2007–2009 crisis?

A. Banks with sufficient regulatory capital can always withstand market stress.

B. Overreliance on short-term funding is dangerous because such funding can evaporate overnight in a crisis. ✅

C. Repos are unsecured borrowings, which is why the repo market froze.

D. ABCP was safe because it was collateralised, so its failure was purely a ratings problem.

B 正确。考点：危机的一条核心教训
教材原话：即便银行自认资本充足，过度依赖短期融资也极其危险——危机中这类资金会一夜消失（期限错配 + 展期依赖是系统性风险的传导核心）。A 被雷曼等案例直接否证；C 说反（repo 是有担保融资）；D 以偏概全（ABCP 虽有抵押，但底层资产质量遭质疑即被抛弃）。

职业行为准则

11. 职业行为准则 Code of Conduct

学习单元 3 · 要点：准则两大板块（原则 / 职业标准）· 五组标准（诚信 1.1–1.7 · 利益冲突 2.1–2.2 · 保密 3.1–3.2 · 基本责任 4.1–4.5 · 最佳实践 5.1–5.4）· 违规后果（暂停/除名/撤销称号）· 情境判断题的三大陷阱（礼品、保密、夸大）

🇨🇳 中文导览

本章是职业行为准则，考试形式为"该会员是否违反准则"的情境判断，题干往往故意把"已披露""想为公司好""对方确实违法"等动机包装成免责理由——判断依据只看行为本身是否触碰标准。准则分两大部分：原则（① 职业诚信与道德行为、② 利益冲突、③ 保密）与职业标准（④ 基本责任、⑤ 最佳实践）。

礼品与独立性（1.2）：不得提供、索取或接受任何"可被合理预期损害自己或他人独立性与客观性"的礼品、利益、报酬或对价——是否披露、动机是否高尚都不豁免；昂贵或稀缺的馈赠风险最高。
保密（3.1/3.2）：未经事先同意须为客户与雇主的保密信息守密；不得将其用于不当目的或个人利益。注意：客户涉嫌违法也不能私下外泄——只可依规向有权机构披露。
反误导（1.4 + 4.4 + 5.3/5.4）：不得明知故犯地虚假陈述分析、建议或专业活动；不得夸大结果的准确性与确定性；沟通须含真实数据；区分事实与观点。
责任不可外包（4.2）：道德责任不能委托他人。法律与准则冲突时，法律优先。
后果：经正式调查，违规可致临时暂停或永久除名，并可撤销使用本职业资格称号的权利。

关键术语：independence of thought独立思维、conflict of interest利益冲突、confidentiality保密义务、misrepresentation虚假陈述、highest standard取最高标准。

🧭 知识网位置全套资料的"行为层"收束：第 8–10 章讲制度与市场怎么会失控，本章讲人为什么会失控——巴林（欺诈+隐瞒）、安然（代理风险）都是准则的负面教材。前置 → 第 9 章 · 金融灾难（行为失范案例） · 第 3 章 · 治理（结构与文化的约束） · 第 8 章 · 风险文化

11.1 Professional integrity & ethical conduct 职业诚信与道德行为（1.1–1.7）

这一组七条标准围绕"诚信"展开，逐条记忆：1.1 与雇主、现有或潜在客户、公众及金融业其他从业者的所有往来中，须专业、道德、诚信；1.2 在提供风险服务时保持合理判断与独立思维——不得提供、索取或接受可能被合理预期损害独立性或客观性的礼品、利益、补偿或对价（注意限定语是"可合理预期损害"，不以实际损害为前提，也不因已披露或动机良好而豁免）；1.3 采取合理预防措施，确保自己的服务不被用于不当、欺诈或非法目的；1.4 不得明知故犯地就分析、建议、行动或其他专业活动作出虚假陈述；1.5 不得从事不诚实或欺骗行为，不得有任何损害其诚信、品格、可信度、专业能力或损害职业声誉的行为；1.6 不得从事损害本职业资格体系（资格称号与考试）完整性或有效性的行为；1.7 注意道德行为与文化习俗的差异，避免在本地习俗下显得不道德的行为；当标准出现冲突或重叠时，始终取最高标准。做题法则：先定位题干触碰的是哪条标准，再看选项里的"免责理由"（披露了/为雇主好/对方违法/金额小）是否在准则中被认可——都不认可。

Act professionally, ethically, with integrity in all dealings (employers, clients, public, practitioners).
Exercise reasonable judgment with independence of thought; never offer/solicit/accept gifts, benefits, compensation or consideration that could reasonably be expected to compromise independence & objectivity (1.2).
Take precautions that services are not used for improper/fraudulent/illegal purposes (1.3); no knowing misrepresentation (1.4).
No dishonesty/deception (1.5); nothing that compromises the profession's designation/examination integrity (1.6).
Mind cultural differences; on conflict/overlap of standards apply the highest standard (1.7).

11.2 Conflicts of interest & confidentiality 利益冲突与保密（2.x / 3.x）

利益冲突两条：2.1 在所有情形下公平行事，向所有受影响方充分披露任何实际或潜在的利益冲突；2.2 充分且公平地披露一切"可被合理预期损害独立性与客观性、或妨碍对雇主、客户及潜在客户履行各自职责"的事项。保密两条：3.1 不得将保密信息用于不当目的；未经事先同意，须对自己的工作、雇主及客户的信息保密；3.2 不得利用保密信息谋取个人利益。考试陷阱：客户涉嫌违法（如税务调查）时，保密义务依然存在——信息只可向有权机构依规披露，向朋友/其他机构泄密即违规；"我为公司挽回了损失"不是免责理由。

2.1 Act fairly; fully disclose actual or potential conflicts to all affected parties.
2.2 Full & fair disclosure of matters that could impair independence/objectivity or interfere with duties to employer/clients.
3.1 No use of confidential information for inappropriate purposes; maintain confidentiality of work/employer/client absent prior consent.
3.2 Never use confidential information for personal benefit.

11.3 Fundamental responsibilities & best practices 基本责任与最佳实践（4.x / 5.x）

基本责任五条：4.1 遵守管辖其职业活动的一切适用法律、规则与规章（包括本准则），不得明知而参与或协助违规；4.2 道德责任不可外包或委托他人；4.3 理解雇主或客户的需求与复杂度，提供适当且合适的风险管理服务与建议；4.4 勤勉地避免夸大结果或结论的准确性、确定性；4.5 清晰披露自己在风险评估、行业实践与适用法律法规方面的具体知识或专业能力的边界。最佳实践四条：5.1 以勤勉执行所有服务，工作方式独立于利益相关方，以最高专业客观性收集、分析、分发风险信息；5.2 熟悉当前公认的风险管理实务，偏离时须明确说明；5.3 确保沟通包含真实数据、不含虚假信息；5.4 在呈报分析与建议时区分事实与观点。考试陷阱：把"历史上的增长"当事实陈述、"预测未来增长"当观点陈述且明示风险——合规；反之把观点包装成事实、或对不可实现的表现作隐性保证（如宣称某组合既能跟踪大盘又保证稳定超额）——违规。

4.1 Comply with applicable laws/rules (incl. this Code); no knowing participation in violations. 4.2 Ethical responsibilities cannot be outsourced.
4.3 Understand needs/complexity; give suitable services & advice. 4.4 Don't overstate accuracy/certainty. 4.5 Disclose limits of knowledge & expertise.
5.1 Diligent execution, independent of interested parties; highest professional objectivity. 5.2 Know generally accepted risk practices; flag departures.
5.3 Communications contain factual data, no false information. 5.4 Distinguish fact vs opinion.

11.4 Violations & consequences 违规与后果 ⭐

会员须同时遵守本准则与管辖风险管理职业的当地法律法规；准则与法律冲突时，法律优先。违规的可能后果（按严重度）：临时暂停资格、永久除名，以及撤销使用本职业资格称号的权利——制裁须经正式调查程序后作出。另有一条容易被忽略的软要求：高标准的道德行为超出准则条文——遇到准则未具体列举的情形，会员仍须以道德方式行事；在与职业相关的所有情境中保持审慎，以维护风险管理领域与从业者的公信力。考试拿分口诀：披露不等于豁免、职务便利不是理由、"对方违法"不破保密、预测≠事实。

Consequences 后果

If the Code and applicable laws conflict, laws take priority. Violations may lead to temporary suspension or permanent removal from membership, and the revocation of the right to use the professional designation — issued after a formal investigation. High ethical behaviour goes beyond the Code: when a situation isn't specifically covered, act ethically and with prudence.

🔑 英文要点速查 · English Essentials

Code structure 准则结构 — Principles: integrity & ethical conduct; conflicts of interest; confidentiality. Professional standards: fundamental responsibilities; best practices.
independence of thought (1.2) 独立思维 — No offering/soliciting/accepting gifts that could reasonably be expected to compromise independence — disclosure doesn't cure; motives don't matter.
misrepresentation (1.4) 虚假陈述 — No knowingly misrepresenting analysis, recommendations, or actions (e.g., implicit performance guarantees).
conflicts (2.1–2.2) 利益冲突 — Act fairly; full disclosure of actual/potential conflicts to all affected parties.
confidentiality (3.1–3.2) 保密 — No inappropriate use; maintain confidentiality absent prior consent; never for personal benefit — even client wrongdoing doesn't allow outside disclosure.
laws & compliance (4.1) 守法 — Comply with applicable laws; no knowing participation in violations; laws take priority over the Code.
no outsourcing of ethics (4.2) 责任不可外包 — Ethical responsibilities cannot be delegated.
overstatement (4.4) 夸大 — Never overstate the accuracy or certainty of results.
fact vs opinion (5.4) 事实 vs 观点 — Separate historical facts from forecasts/opinions in every communication (5.3: factual data, no false information).
highest standard (1.7) 取最高标准 — Cultural conflicts: apply the strictest applicable standard.
sanctions 制裁 — Suspension, permanent removal, revocation of designation rights — after a formal investigation.
beyond the Code — Unlisted situations still require ethical, prudent conduct upholding the profession's integrity.

📝 本章自测 · Self-Check

以下为自编练习题（非真题）：题干保留英文考试形态，解析为中文——先自己判断，再点开核对。共 4 题。

Q1. An analyst has been invited by a major corporate client to visit its facilities. She declines the company's offer of a private jet and luxury hotel suite, but accepts two hard-to-obtain courtside seats to a championship final, which she promptly discloses to her employer. Has she violated the Code?

A. No — she disclosed the gift to her employer and declined the other offers.

B. No — accepting gifts that create goodwill is a normal business courtesy.

C. Yes — accepting gifts that could reasonably be expected to compromise independence and objectivity. ✅

D. Yes — because employees must decline all gifts of any kind.

C 正确。考点：独立性标准的"可合理预期"口径（1.2）
昂贵、难求的馈赠可能影响未来的研究与判断——披露与拒绝部分馈赠都不能"治愈"这一风险（A 的免责不成立）。B 太宽松（准则以"可合理预期损害独立性"为准绳，而非一概允许）；D 太严苛（准则并未禁止一切礼品，"名义价值"的礼节性馈赠是被允许的，被禁的是可能损害独立性者）。

Q2. A portfolio manager learns that a client's company is under investigation for customs fraud. Without telling anyone else, he passes the information to a friend at a bank so the bank can withdraw its lending proposal. Which statement is most likely correct?

A. No violation — the information concerned the client's illegal activity.

B. Violation — he failed to maintain the confidentiality of client information. ✅

C. Violation — he should first have reported the fraud to the authorities.

D. No violation — preventing potential losses for the bank justifies the disclosure.

B 正确。考点：保密义务的边界（3.1/3.2）
保密义务不因信息涉及客户违法而自动解除——此类信息只能依规向有权机构披露，不能私下告知朋友或第三方（A、D 都是常见错误辩护）。C 混淆义务：准则并未要求会员自行调查并举报客户犯罪，其违反点在于泄密本身。

Q3. In promotional materials, a manager states: "Our quantitative process selects roughly 30 small-cap stocks but is engineered to track the broad market index while consistently earning a 3–5% annual premium." This statement is:

A. permissible — it describes the fund's risk/return objective.

B. permissible — the claim is based on the firm's model estimates.

C. not permissible — it misrepresents the performance the manager can reasonably expect to achieve. ✅

D. not permissible — only because it mentions a specific percentage range.

C 正确。考点：虚假陈述标准（1.4）
30 只小盘股组合不可能可靠跟踪宽基指数，且"持续赚取溢价"是无法保证的隐性业绩承诺——属于对可合理预期业绩的虚假陈述。A、B 的辩护不成立（目标不等于保证，模型估计不豁免误导）；D 错在理由（百分比本身不是问题，问题是对不可实现表现的暗示保证）。

Q4. An analyst's report states: "Sector revenues in this emerging market grew 12% last year, and we expect that growth in orders will translate into higher industry profits over the next three to five years." The report then details major political and currency risks. Regarding the Code, the report:

A. has not violated the Code — historical facts and clearly framed expectations are properly distinguished, and key risks are disclosed. ✅

B. violated the Code by failing to distinguish facts from opinions.

C. violated the Code by recommending an unsuitable investment.

D. violated the Code by providing insufficient operational detail.

A 正确。考点：事实与观点的区分（5.3/5.4）
已发生的增长率是事实；"我们预计……"是明确标注的观点，且报告披露了主要风险——符合"沟通含真实数据、区分事实与观点"的要求。B 的指控与文本不符；C/D 无中生有（报告描述的是行业分析，且已履行风险披露义务）。

Interactive Tools · 交互计算器

🧮 Interactive Calculators & Formula Sheet

Enter numbers and press the button — instant answers for the /6 arithmetic you'll face on the exam. Always input percentages as decimals of the percentage point used in each field.

1 · CAPM required return E(Rᵢ) = R_F MRP × β

Risk-free rate R_F (%) Market risk premium E(R_M) − R_F (%) Beta (β)

2 · Beta from correlation & volatilities β = ρ·(σᵢ/σ_M)

Correlation ρ_i,M Asset volatility σᵢ (%) Market volatility σ_M (%)

3 · Performance ratios Sharpe / Treynor / Jensen / IR / Sortino

Portfolio return R_P (%) Risk-free rate R_F (%) Market return R_M (%) Beta β Portfolio std dev σ_P (%) Benchmark return R_B (%) Tracking error (%) Min acceptable return R_MIN (%) Downside deviation (%)

4 · Fama-French three-factor return 三因子模型预期收益

R_F (%) β_M Equity risk premium RP_M (%) β_SMB SMB factor (%) β_HML HML factor (%)

⭐ Formula Sheet — memorise these (英文为主, 中文注释)

Formula	Usage 用途
EL = PD × EAD × LGD	Expected credit loss 预期信用损失
β_i = Cov(i,M)/σ_M² = ρ_i,Mσ_i/σ_M	Systematic risk 系统性风险
E(R_i) = R_F [E(R_M)−R_F]·β_i	CAPM required return (SML) CAPM 要求回报
CML: E(R_P) = R_F MRP/σ_Mσ_P	Efficient portfolios 资本市场线
Sharpe = (R_P−R_F)/σ_P	per unit total risk 总风险
Treynor = (R_P−R_F)/β_P	per unit market risk 系统性风险
Jensen α = R_P − [R_Fβ_P(E(R_M)−R_F)]	alpha vs CAPM 超额收益
Tracking error = σ(R_P−R_B)	active risk 跟踪误差
IR = E(R_P−R_B)/TE	active return/risk 信息比率
Sortino = (R_P−R_MIN)/downside dev.	downside-only 下行风险
APT: R_i = E(R_i) + Σβ_kF_k e_i	multifactor return 多因子收益
Fama-French: E(R_i) = R_F β_MRP_M β_SMBF_SMB β_HMLF_HML	size & value factors 规模与价值因子