Stress Testing

Stress Testing

                宏观情景（Baseline / Adverse / Severely Adverse）
                                   │
        ┌──────────────────────────┼──────────────────────────┐
        │                          │                          │
   Model Stress Testing        Accounting Loss           Regulatory Capital
   (PD / LGD / Models)         Forecasting               Stress Testing
        │                          │                          │
   “模型还能信吗？”          “我要计提多少钱？”        “资本还够不够？”
        │                          │                          │
     MRM / Model Team         CECL / IFRS 9               CCAR / DFAST

The Big Picture: Why Stress Testing Exists

Goal: Ensure the firm survives a severe but plausible shock without breaching regulatory capital/liquidity requirements or threatening solvency.
  • Regulatory Drivers:
    • CCAR/DFAST (U.S. Fed for large banks)
    • PRA Stress Test (UK)
    • ECB Stress Test (EU)
    • Basel III/IV (FRTB, CRR) – requires internal stress testing frameworks
  • Internal Drivers:
    • Capital planning
    • Risk appetite setting
    • Trading desk limits
    • Board & senior management oversight

The Significance of Stress Testing

1. Risk Assessment

Stress testing is a powerful tool to assess and quantify the risks associated with financial models and portfolios. By subjecting these entities to extreme scenarios, institutions can identify vulnerabilities and weaknesses that might not be apparent under normal conditions.

2. Capital Adequacy

Regulatory bodies often require financial institutions to conduct stress tests to ensure they maintain sufficient capital reserves to weather economic downturns. Stress testing helps determine if institutions can absorb losses without jeopardizing their stability.

3. Scenario Planning

Stress testing enables financial institutions to prepare for a wide range of scenarios, from economic recessions to market crashes. By analyzing model performance under these scenarios, institutions can develop contingency plans and risk mitigation strategies.

4. Risk Management

Stress testing is an integral part of risk management. It allows institutions to make informed decisions about their exposure to various risks, including credit risk, market risk, and liquidity risk. This, in turn, helps protect the institution’s financial health.

5. Investor Confidence

Demonstrating a commitment to stress testing can enhance investor confidence. When investors see that an institution is proactive in assessing and managing risks, they are more likely to trust and invest in that institution.

The Stress Testing Cycle (Annual + Ongoing)

PART 1: MARKET RISK STRESS TESTING

核心定义 (Core Concept)
Stress testing and scenario analysis are key tools for assessing market risk under extreme but plausible conditions.
They help banks estimate potential losses, test capital adequacy, and identify portfolio vulnerabilities.
中文助记：
压力测试和情景分析都是在“极端但合理”条件下检验风险敞口、资本充足性和脆弱点的工具。

🧩 1️⃣ What is Stress Testing?

Stress testing simulates extreme market conditions to assess how resilient a portfolio or firm is.
In market risk, it means shocking key market variables — rates, credit spreads, equities, volatility — and measuring portfolio losses.
Regulators (like the Fed under CCAR) require annual stress tests to ensure banks can survive severe downturns.
中文理解：
压力测试是通过模拟极端市场环境（利率暴涨、股市暴跌、信用利差扩大等）
来评估银行是否能承受冲击并保持资本充足。
面试短答模板：
“Stress testing measures portfolio resilience by simulating extreme but plausible shocks across key market factors, complementing VaR by focusing on tail risk and capital adequacy.”

Types of Stress Testing

(a) Historical Stress Testing

项目
内容
Definition
使用真实历史数据（如 2008、2020 疫情）重放历史冲击对当前组合的影响。
Data Source
实际历史市场数据：利率、汇率、股价、信用利差。
Strengths
✅ 真实可信、易沟通、可验证组合在过去危机下表现。
Limitations
❌ 仅限于过去事件、不能捕捉新风险、重复性低。
口述句：
“Historical stress tests replay real events like 2008 or COVID to see how today’s portfolio would have behaved.”

(b) Hypothetical / Anticipatory Stress Testing

项目
内容
Definition
构建基于假设或专家判断的未来冲击场景。
Data Source
假设数据：如“油价暴跌 + 利率急升”。
Strengths
✅ 捕捉未出现的风险；可定制当前市场脆弱点；用于资本充足评估。
Limitations
❌ 主观性强；需严格验证合理性与一致性。
口述句：
“Hypothetical stress testing is forward-looking — it captures emerging risks not yet seen in history.”

Stress Testing Framework (实务框架)

步骤
内容
关键词
1. Define Objective
明确目标（监管、内部容忍度、资本规划）
Objective
2. Identify Risk Factors
确定主要风险因子（利率、信用、股市、波动率、相关性）
Risk Drivers
3. Design Scenarios
历史 or 假设型；需“极端但合理”
Scenario Design
4. Apply Shocks & Revalue
使用全重估或敏感度近似（ΔΓVega / DV01 / CS01）计算损失
Revaluation
5. Aggregate Results
汇总损失 → 资产类别 / 交易台 / 因子层级
Aggregation
6. Interpret & Report
评估合理性 + 严重性；报告风险委员会 / 资本充足
Governance
中文口诀：
目标 → 因子 → 场景 → 重估 → 汇总 → 报告。
面试一句话总结：
“We follow a structured process — define objective, identify key risk factors, design scenarios, revalue the portfolio, aggregate results, and interpret them for capital and governance.”

4️⃣ Why Conduct Stress Testing?

类型
目的
中文解释
Historical
理解组合在已知危机下的反应，识别关键损失驱动因素。
再现已知危机，验证模型反应。
Hypothetical
预判未来黑天鹅风险、评估潜在尾部损失。
前瞻规划、发现潜在脆弱点。
面试加分点：
“Historical stress tells us how we would have performed; hypothetical stress helps us prepare for what hasn’t happened yet.”

🧠 5️⃣ Governance & Use in Practice

  • 场景由 风险管理部 + 前台 (FO) 共同讨论确认。
  • 结果提交给 风险委员会 / 高管层 审议。
  • 结果会影响 限额设定、资本规划、风险偏好声明 (Risk Appetite Statement)。
  • 符合 SR 11-7、CCAR、Basel ICAAP 等监管要求。
口述补充：
“Results are reviewed by senior management and used to calibrate risk appetite and capital buffers.”

PART 2: CREDIT RISK STRESS TESTING

Step 1: Scenario Design (Same macro shocks, different transmission)
Same macro scenarios, but now we model how they impact borrowers:
  • Higher unemployment → ↑ consumer defaults
  • GDP contraction → ↑ corporate defaults
  • Rate hikes → ↑ debt servicing costs → ↓ credit quality
✅ Key Credit Risk Drivers:
  • PD (Probability of Default)
  • LGD (Loss Given Default)
  • EAD (Exposure at Default)
  • Collateral values (for secured loans)
Step 2: Portfolio Segmentation
Credit portfolios are split into:
  1. Wholesale Credit (corporate loans, bonds, counterparty exposure)
  1. Retail Credit (mortgages, credit cards, auto loans)
  1. Counterparty Credit Risk (CCR) from derivatives (this is where Market + Credit intersect)
For each segment, different models apply:
  • Wholesale: Portfolio models (e.g., CreditMetrics, KMV)
  • Retail: Vintage models + macro regressions
  • CCR: Simulate MtM of derivatives under stress → calculate EAD/PD
Step 3: Modeling Credit Losses

A. Expected Credit Loss (ECL) under IFRS 9 / CECL

  • Stressed PDs are fed into ECL models:ECL = PD × LGD × EAD
  • Example:
    • Baseline PD for BBB corp: 0.5%
    • Stressed PD (recession): 3.0%
      → ECL increases 6x

B. Counterparty Credit Risk (CCR)

  • For each counterparty (e.g., hedge fund, bank):
      1. Simulate portfolio MtM under stress scenarios
      1. Calculate Peak Exposure (e.g., 95th percentile of exposure over 1 year)
      1. Apply stressed PD → CVA = LGD × ∫ PD(t) × EE(t) dt
🔥 Critical insight:
In 2008, market moves caused counterparty defaults (Lehman, AIG). So CCR stress testing depends on market risk scenarios.

PART 3: HOW MARKET & CREDIT RISK STRESS TESTING INTEGRATE

Key Integration Points:

Linkage
Mechanism
CVA Risk
Market moves → derivative MtM → exposure → credit loss
Collateral Haircuts
Stressed asset prices → lower collateral value → higher EAD
Liquidity Spirals
Market sell-off → margin calls → forced liquidation → further price drops
Rating Downgrades
Credit stress → rating downgrade → higher capital charges (FRTB DRC)
Example:
  • Equity crash → hedge fund’s equity derivatives go deep OTM →
  • Bank calls margin → fund can’t pay → default →
  • Bank books market loss (MtM) + credit loss (CVA)

PART 4: GOVERNANCE & OUTPUTS

Who’s Involved?

  • Risk Teams: Market Risk, Credit Risk, Model Risk
  • Finance: Capital planning, accounting (ECL)
  • Treasury: Liquidity stress testing (LCR/NSFR impact)
  • CRO Office: Signs off on scenarios & results
  • Board & Regulators: Review final submission (e.g., CCAR)

Key Deliverables:

  1. Pre-Provision Net Revenue (PPNR) Impact
  1. CET1 Capital Ratio under stress (must stay above minimum + buffer)
  1. Liquidity Coverage Ratio (LCR) under outflow stress
  1. Risk Concentration Reports (sector, region, counterparty)
  1. Action Plan: Capital actions, risk limits, hedging strategies
 
 
 

AI Stress Testing

一、什么是 AI Stress Testing

在金融行业，压力测试（Stress Testing）指的是：
在极端但合理的情景下，测试系统是否仍然稳定。
例如银行会模拟：
  • 经济衰退
  • 房价暴跌
  • 利率大幅上升
然后测试：
  • 贷款违约率
  • 银行资本充足率
  • 流动性风险
目的是确保：
系统在极端情况下不会崩溃。

二、为什么 AI 也需要压力测试

AI 系统正在进入越来越关键的领域：
  • 金融决策
  • 医疗诊断
  • 自动驾驶
  • 政策分析
如果 AI 在极端情况下表现异常，可能造成严重后果。
例如：
  • 错误的贷款审批
  • 自动交易系统失控
  • 医疗诊断错误
因此需要测试：
AI在极端输入或复杂环境下的表现。

三、AI压力测试可能测试什么

未来 AI Stress Testing 可能包括几个方面。

1 数据分布变化（Distribution Shift）

测试模型在数据变化时的表现。
例如：
  • 新市场环境
  • 新用户行为
  • 宏观经济变化
在金融模型中，这类似于：
宏观压力情景。

2 对抗性输入（Adversarial Inputs）

测试 AI 是否容易被操纵。
例如：
  • 特殊输入让模型产生错误判断
  • 提示攻击（Prompt Injection）
这在 LLM 系统中非常常见。

3 极端场景

测试 AI 在非常罕见情况下的表现。
例如：
  • 市场崩盘
  • 极端天气
  • 数据缺失

4 系统级风险

AI系统往往不是单一模型，而是：
复杂系统
例如：
  • 数据管道
  • 模型
  • API
  • 决策系统
压力测试需要评估整个系统。

四、AI压力测试可能成为监管要求

一些监管机构已经开始讨论类似框架。
例如：
AI风险管理框架通常包括：
  • Risk identification
  • Risk measurement
  • Risk monitoring
  • Risk mitigation
未来可能增加：
AI Stress Testing
特别是在：
  • 金融
  • 医疗
  • 自动驾驶
这些高风险领域。