课题：为什么"平均回报"永远在骗你——几何均值与算术均值的终极对决
一、引言：一个让无数投资者上当的魔术
场景： 某基金经理向你展示："过去 4 年我们的平均年化回报是 25%！"
你查了一下实际数据： - 第 1 年：+100% - 第 2 年：-50% - 第 3 年：+100% - 第 4 年：-50%
算术平均：(100% - 50% + 100% - 50%) / 4 = 25% ✅ 他没说谎。
但你投了 100 万，4 年后还剩多少？
- 第 1 年末：100 × 2.0 = 200
- 第 2 年末：200 × 0.5 = 100
- 第 3 年末：100 × 2.0 = 200
- 第 4 年末：200 × 0.5 = 100
100 万。一分没赚。 算术平均告诉你是 25%，真相是 0%。
🔥 这就是本课的核心：算术均值描述"单期期望"，几何均值描述"多期真实增长"。混淆两者，代价可能是你的全部收益。
二、为什么几何均值 ≤ 算术均值：数学证明与直觉理解
2.1 不等式：AM ≥ GM
对于任意一组正数 x₁, x₂, ..., xₙ：
AM = (x₁ + x₂ + ... + xₙ) / n ≥ ⁿ√(x₁ · x₂ · ... · xₙ) = GM
等式成立的条件：所有 xᵢ 完全相等（即零波动）。
2.2 直觉理解——"波动税"
想象你有两个投资方案：
| 方案 | 第 1 年 | 第 2 年 | 算术平均 | 几何平均 | 实际 2 年总回报 |
|---|---|---|---|---|---|
| 稳定 | +10% | +10% | 10% | 10% | 21% |
| 波动 | +50% | -30% | 10% | 2.47% | 5% |
算术平均完全相同（10%），但实际回报相差 4 倍以上。
波动越大 → 几何均值偏离算术均值越大 → 这就是"波动税"（Volatility Drag / Variance Drain）
波动税公式（近似）：
GM ≈ AM − σ²/2
其中 σ² 是收益率方差。波动每增加 1 个单位，"吃掉"约 0.5 个单位的复利回报。
2.3 波动税的威力
| 波动程度 | AM | σ² | GM ≈ AM − σ²/2 |
|---|---|---|---|
| 零波动 | 10% | 0 | 10% |
| 低波动 | 10% | 0.01 | 9.5% |
| 中波动 | 10% | 0.04 | 8.0% |
| 高波动 | 10% | 0.09 | 5.5% |
| 极端波动 | 10% | 0.16 | 2.0% |
📌 CFA 一级常考点：波动越大，AM 和 GM 的差距越大。
三、两种均值的本质区别
3.1 对比表
| 维度 | 算术均值 (AM) | 几何均值 (GM) |
|---|---|---|
| 数学本质 | 加法思维 | 乘法思维 |
| 回答的问题 | "平均每期是多少？" | "实际的复合增速是多少？" |
| 适用数据 | 横截面（同一时点的多个个体） | 时间序列（同一个体在多个时点） |
| 与复利的关系 | 不反映复利 | 精确反映复利 |
| 对异常值 | 敏感 | 相对稳健 |
| 对零/负值 | 可以处理 | 有零→GM=0；有负→需要(1+r)转换 |
| 代表场景 | 这个月各股票的平均涨幅 | 这只股票过去5年的年化回报 |
3.2 一句话法则
🎯 回顾看历史 → 几何平均；预测做决策 → 算术平均（期望值）
为什么？
- 历史回报（ex-post）： 钱是连本带利滚的，必须用几何平均衡量真实增长
- 未来预期（ex-ante）： 你不知道每年的顺序，用算术平均做期望值估计
⚠️ CFA 中计算要求回报率（required return）或做资产配置长期预期时，通常用几何平均（考虑复利效应）。
四、实战应用：什么时候用哪个？
4.1 必须用几何平均的场景
场景 1：计算 CAGR（Compound Annual Growth Rate）
CAGR = (V_end / V_begin)^(1/n) − 1
CAGR 就是几何平均！基金宣传材料中的"年化回报"必须是 CAGR。
场景 2：多期投资组合表现 "我投了 5 年，每年实际赚多少？"→ 几何平均。
场景 3：GDP 增长率、通货膨胀率的长期趋势 这些都是复利滚动的指标，必须用几何平均。
4.2 必须用算术平均的场景
场景 1：横截面分析 "2024 年标普 500 中 500 只股票的日均回报均值"→ 算术平均（同一年、不同股票）
场景 2：概率加权期望值 E(R) = Σ pᵢ × Rᵢ —— 这就是算术平均的加权版。
场景 3：作为方差 / 标准差的基础 σ² = Σ(xᵢ − x̄)²/(n−1)，这里的 x̄ 必须是算术平均。不能用几何平均算标准差。
场景 4：单期预测 "明年这个策略大概赚多少？"→ 用算术平均做点估计。
4.3 容易混淆的边界案例
| 问题 | 正确答案 | 理由 |
|---|---|---|
| 过去10年标普500平均年回报？ | 几何平均 | 跨期表现，复利在起作用 |
| 2025年100只基金的平均回报？ | 算术平均 | 同一时点、不同基金，无复利关系 |
| 某股票过去30天平均日收益率？ | 算术平均 | 日收益率用算术平均做统计推断 |
| 某股票过去10年CAGR？ | 几何平均 | CAGR = 几何平均 - 1，定义如此 |
五、回报率数据的几何平均计算
5.1 标准公式
对于回报率序列 r₁, r₂, ..., rₙ：
GM = [(1+r₁)(1+r₂)...(1+rₙ)]^(1/n) − 1
⚠️ 关键细节：回报率有负值时，先加 1 转化为增长因子，再求几何平均，最后减 1。
5.2 计算示例
某投资 5 年回报：+15%, +8%, −4%, +12%, +6%
步骤： 1. 转为增长因子：1.15, 1.08, 0.96, 1.12, 1.06 2. 连乘：1.15 × 1.08 × 0.96 × 1.12 × 1.06 = 1.4189 3. 开 5 次方：1.4189^(1/5) = 1.0725 4. 减 1：GM = 7.25%
验证： 100 × 1.0725⁵ = 141.89 ≈ 实际终值 141.89 ✅
算术平均对比： (15% + 8% − 4% + 12% + 6%) / 5 = 7.4%
只差 0.15%，因为这里波动不算太大。波动越大 → 差距越大。
六、几何均值的数据要求
6.1 只能用于比率尺度（Ratio Scale）
几何平均要求数据具有绝对零点，因为乘法运算依赖"相对比例"。
| 尺度 | 能否用几何平均 | 例子 |
|---|---|---|
| 名义 | ❌ | 行业分类 |
| 序数 | ❌ | 晨星评级 1-5 星 |
| 间隔 | ❌ | 温度（0°C 不表示"没有温度"） |
| 比率 | ✅ | 回报率、增长率、价格比 |
6.2 零值和负值处理
- 有零： GM = 0（乘积为 0，全部本金损失）
- 有负数： 在 (1+r) 空间处理；若 (1+r) 为负 → 几何平均无实数解
七、跨期绩效归因：AM 和 GM 联用
7.1 分解收益来源
某基金经理在 3 年牛熊市中：
| 年份 | 市场回报 | 基金回报 | 超额收益 |
|---|---|---|---|
| 牛市 | +30% | +35% | +5% |
| 熊市 | −20% | −15% | +5% |
| 震荡 | +8% | +12% | +4% |
问题：基金相对于市场的"平均超额"是多少？
超额的平均（算术）： (5% + 5% + 4%) / 3 = 4.67% ← 用算术，超额是加法概念
基金的 CAGR： [(1.35)(0.85)(1.12)]^(1/3) − 1 = 8.75% 市场的 CAGR： [(1.30)(0.80)(1.08)]^(1/3) − 1 = 4.00% 实际复利超额： 8.75% − 4.00% = 4.75%
🔑 超额收益用算术平均衡量跟踪误差和 alpha 稳定性；真实回报用几何平均衡量复利效果。
八、常见陷阱与误区
陷阱 1：用算术平均代替 CAGR
"这只基金过去 10 年平均年回报 15%。"→ 绝大多数情况下，CAGR < 15%。 检测方法： 把每年回报列出来，用 (1+r) 连乘，开 n 次方，看是不是 15%。
陷阱 2：忽略波动对复利的影响
两个组合同样有 AM = 8%： - A：每年固定 8%（GM = 8%） - B：−10%, +30%, −5%, +25%, −8%（GM = 5.2%）
算术平均可以相同，几何平均可以差很远！
陷阱 3：横截面和时间序列混用
- 同一年 50 只基金 → 算术平均
- 一只基金 50 年 → 几何平均
陷阱 4：认为几何平均"更准确"所以永远用它
几何平均不适用于：横截面比较、统计推断（需算术平均计算方差）、单期期望预测
九、测试题
选择题
Q1：某投资 3 年回报率分别为 +50%、−50%、+50%，以下哪个说法正确？ - A. 算术平均回报为 16.67%，实际终值高于初始值 - B. 几何平均回报低于算术平均，且算术平均为 16.67% - C. 算术平均和几何平均均为 50% - D. 几何平均回报为 16.67%，算术平均为 50%
Q2：当收益率波动（方差）增加时，算术平均与几何平均之间的差距会？ - A. 缩小 - B. 扩大 - C. 保持不变 - D. 取决于回报率的正负
Q3：以下哪个场景最适合用算术平均而非几何平均？ - A. 计算某基金过去 10 年的年化复合回报 - B. 计算 50 只不同科技股在 2025 年的平均回报率 - C. 计算你的投资组合自成立以来的 CAGR - D. 计算某国过去 20 年的平均 GDP 增长率
Q4：已知某资产的年化算术平均回报为 12%，年化波动率（σ）为 30%。根据波动税近似公式 GM ≈ AM − σ²/2，该资产的年化几何平均约为？ - A. 12.0% - B. 9.0% - C. 7.5% - D. 16.5%
Q5：以下哪种数据类型可以使用几何平均？ - A. 晨星基金评级（1-5 星） - B. 各城市的温度（摄氏度） - C. 股票的年化回报率 - D. 投资风格分类（成长/价值/平衡）
Q6：一个投资组合 5 年的财富增长因子（即每年 (1+r)）分别为：1.10, 0.90, 1.20, 0.95, 1.15。几何平均增长率最接近？ - A. 6.0% - B. 5.3% - C. 10.6% - D. 4.8%
Q7：当一组数据的所有值完全相同时，以下哪项成立？ - A. AM > GM - B. GM > AM - C. AM = GM - D. AM 和 GM 的关系不确定
Q8：某分析师说："这只基金算术平均年回报 20%，CAGR 是 15%，相差 5 个百分点的原因是？" - A. 基金有管理费 - B. 基金回报率波动大，存在波动税 - C. CAGR 计算有误 - D. 基金使用了杠杆
答案
Q1：B — 算术平均 = (50% − 50% + 50%) / 3 = 16.67%。增长因子连乘：1.5 × 0.5 × 1.5 = 1.125，GM = 1.125^(1/3) − 1 ≈ 4.0%。AM > GM。
Q2：B — 波动越大，波动税（Variance Drain）越大，AM 和 GM 差距越大。参考公式 GM ≈ AM − σ²/2。
Q3：B — 50 只不同股票在同一年的回报率是横截面数据，用算术平均。A、C、D 都是时间序列的跨期复合增长，必须用几何平均。
Q4：C — GM ≈ 12% − (0.30² / 2) = 12% − (0.09/2) = 12% − 4.5% = 7.5%。仅 30% 的波动就吃掉了 4.5 个百分点的复利！
Q5：C — 股票年化回报率是比率尺度数据，有绝对零点，可以用几何平均。评级是序数，温度是间隔，分类是名义。
Q6：B — 连乘：1.10 × 0.90 × 1.20 × 0.95 × 1.15 = 1.29843。GM = 1.29843^(1/5) − 1 = 1.0536 − 1 ≈ 5.36%，最接近 5.3%。
Q7：C — 当所有数据完全相同时，AM = GM。这是 AM ≥ GM 中等号成立的唯一条件。
Q8：B — AM 和 CAGR（即 GM）的差距来自波动税。回报率波动越大，AM 和 GM 差距越大。管理费同时影响两者，杠杆放大回报但不直接造成 AM-GM 差距。
十、备考要点
| 优先级 | 考点 | 关键记忆 |
|---|---|---|
| ⭐⭐⭐ | AM vs GM 适用场景判断 | 横截面→AM / 时间序列→GM |
| ⭐⭐⭐ | 波动税公式 GM ≈ AM − σ²/2 | 波动每↑1单位，复利↓0.5单位 |
| ⭐⭐ | AM ≥ GM，等号仅在所有值相等时 | 考判断题 |
| ⭐⭐ | GM 只能用于比率尺度数据 | 名义/序数/间隔尺度不能用 |
| ⭐⭐ | CAGR = GM − 1 | 基金回报率的正确衡量 |
| ⭐ | GM 计算：先 (1+r) 连乘，再开 n 次方，再减 1 | 计算题经常出现 |
📊 Sindy姐的投资笔记： 下次有人跟你讲"年均回报 XX%"，第一反应不是"赚好多"，而是"把每年回报列出来我看看"。几何平均才是你账户里真正的钱。算术平均？那是销售的话术。🌹
Topic: Why "Average Returns" Always Lie — The Ultimate Showdown Between Geometric Mean and Arithmetic Mean
1. Introduction: A Magic Trick That Fools Countless Investors
Scenario: A fund manager proudly presents: "Our average annualized return over the past 4 years is 25%!"
You check the actual data: - Year 1: +100% - Year 2: −50% - Year 3: +100% - Year 4: −50%
Arithmetic mean: (100% − 50% + 100% − 50%) / 4 = 25% ✅ He's not lying.
But you invest 1 million. How much is left after 4 years?
- End of Year 1: 100 × 2.0 = 200
- End of Year 2: 200 × 0.5 = 100
- End of Year 3: 100 × 2.0 = 200
- End of Year 4: 200 × 0.5 = 100
1 million. Zero gain. The arithmetic mean says 25%, the truth is 0%.
🔥 Core takeaway: The arithmetic mean describes "single-period expectation." The geometric mean describes "multi-period actual growth." Confusing the two can cost you all your returns.
2. Why GM ≤ AM: Mathematical Proof and Intuition
2.1 The Inequality: AM ≥ GM
For any set of positive numbers x₁, x₂, ..., xₙ:
AM = (x₁ + x₂ + ... + xₙ) / n ≥ ⁿ√(x₁ · x₂ · ... · xₙ) = GM
Equality holds only when all xᵢ are exactly equal (zero volatility).
2.2 Intuition — The "Volatility Tax"
Imagine two investment strategies:
| Strategy | Year 1 | Year 2 | AM | GM | Actual 2-Year Return |
|---|---|---|---|---|---|
| Stable | +10% | +10% | 10% | 10% | 21% |
| Volatile | +50% | −30% | 10% | 2.47% | 5% |
The arithmetic means are identical (10%), yet actual returns differ by over 4×.
Greater volatility → Larger deviation between GM and AM → This is the "Volatility Drag" (Variance Drain)
Volatility Drag Formula (approximation):
GM ≈ AM − σ²/2
Where σ² is the variance of returns. Each unit increase in volatility "eats" about 0.5 units of compound return.
2.3 The Power of Volatility Drag
| Volatility Level | AM | σ² | GM ≈ AM − σ²/2 |
|---|---|---|---|
| Zero | 10% | 0 | 10% |
| Low | 10% | 0.01 | 9.5% |
| Moderate | 10% | 0.04 | 8.0% |
| High | 10% | 0.09 | 5.5% |
| Extreme | 10% | 0.16 | 2.0% |
📌 CFA Level 1 key point: The greater the volatility, the wider the gap between AM and GM.
3. The Essential Difference Between the Two Means
3.1 Comparison Table
| Dimension | Arithmetic Mean (AM) | Geometric Mean (GM) |
|---|---|---|
| Mathematical nature | Additive thinking | Multiplicative thinking |
| Question answered | "What is the average per period?" | "What is the actual compound growth rate?" |
| Data type | Cross-sectional (multiple entities at one point in time) | Time series (one entity across multiple points in time) |
| Relation to compounding | Does not reflect compounding | Precisely reflects compounding |
| Sensitivity to outliers | Sensitive | Relatively robust |
| Handling zero/negative values | Can handle | Zero → GM = 0; Negative → requires (1+r) transformation |
| Typical use case | Average return of all stocks this month | Annualized return of this stock over the past 5 years |
3.2 One-Line Rule
🎯 Looking backward at history → Geometric Mean; Forecasting for decisions → Arithmetic Mean (expected value)
Why?
- Historical returns (ex-post): Money compounds period over period; must use geometric mean to measure actual growth
- Future expectations (ex-ante): You don't know the sequence of returns; use arithmetic mean for expected value estimation
⚠️ Note: In CFA, when calculating required return or making long-term asset allocation forecasts, the geometric mean is typically used (to account for compounding effects).
4. Practical Application: When to Use Which?
4.1 Must Use Geometric Mean
Case 1: Calculating CAGR (Compound Annual Growth Rate)
CAGR = (V_end / V_begin)^(1/n) − 1
CAGR is the geometric mean! The "annualized return" in fund marketing materials must be CAGR.
Case 2: Multi-period portfolio performance "I invested for 5 years — how much did I actually earn per year?" → Geometric mean.
Case 3: Long-term trends in GDP growth, inflation rates These are compounding indicators; must use geometric mean.
4.2 Must Use Arithmetic Mean
Case 1: Cross-sectional analysis "Average daily return of all 500 S&P 500 stocks in 2024" → Arithmetic mean (same year, different stocks)
Case 2: Probability-weighted expected value E(R) = Σ pᵢ × Rᵢ — this is the weighted version of arithmetic mean.
Case 3: As the basis for variance / standard deviation σ² = Σ(xᵢ − x̄)²/(n−1), where x̄ must be the arithmetic mean. You cannot use the geometric mean to calculate standard deviation.
Case 4: Single-period forecasting "How much will this strategy likely earn next year?" → Use arithmetic mean for point estimate.
4.3 Boundary Cases — Easy to Confuse
| Question | Correct Answer | Rationale |
|---|---|---|
| Average annual return of S&P 500 over past 10 years? | Geometric Mean | Cross-period performance; compounding at work |
| Average return of 100 funds in 2025? | Arithmetic Mean | Same point in time, different funds; no compounding relationship |
| Average daily return of a stock over past 30 days? | Arithmetic Mean | Daily returns typically use arithmetic mean for statistical inference |
| CAGR of a stock over past 10 years? | Geometric Mean | CAGR = GM − 1, by definition |
5. Calculating Geometric Mean for Return Data
5.1 Standard Formula
For a sequence of returns r₁, r₂, ..., rₙ:
GM = [(1+r₁)(1+r₂)...(1+rₙ)]^(1/n) − 1
⚠️ Key detail: When returns include negative values, first add 1 to convert to growth factors, then compute the geometric mean, then subtract 1.
5.2 Calculation Example
A 5-year investment with returns: +15%, +8%, −4%, +12%, +6%
Steps: 1. Convert to growth factors: 1.15, 1.08, 0.96, 1.12, 1.06 2. Multiply: 1.15 × 1.08 × 0.96 × 1.12 × 1.06 = 1.4189 3. Take the 5th root: 1.4189^(1/5) = 1.0725 4. Subtract 1: GM = 7.25%
Verification: 100 × 1.0725⁵ = 141.89 ≈ Actual terminal value 141.89 ✅
Arithmetic mean comparison: (15% + 8% − 4% + 12% + 6%) / 5 = 7.4%
Only a 0.15% difference — because the volatility here is moderate. Greater volatility → larger gap.
6. Data Requirements for Geometric Mean
6.1 Only for Ratio Scale Data
The geometric mean requires data with an absolute zero point, since multiplication depends on "relative proportions."
| Scale | Can Use GM? | Example |
|---|---|---|
| Nominal | ❌ | Industry classification |
| Ordinal | ❌ | Morningstar ratings (1-5 stars) |
| Interval | ❌ | Temperature (0°C does not mean "no temperature") |
| Ratio | ✅ | Returns, growth rates, price ratios |
6.2 Handling Zero and Negative Values
- If any value is zero: GM = 0 (product is zero — all principal lost)
- If any value is negative: Handle in (1+r) space. If (1+r) is negative → GM has no real solution
7. Cross-Period Performance Attribution: Using AM and GM Together
7.1 Decomposing Return Sources
A fund manager over 3 years of bull and bear markets:
| Year | Market Return | Fund Return | Excess Return |
|---|---|---|---|
| Bull | +30% | +35% | +5% |
| Bear | −20% | −15% | +5% |
| Sideways | +8% | +12% | +4% |
Question: What is the fund's "average excess return" relative to the market?
Arithmetic mean of excess: (5% + 5% + 4%) / 3 = 4.67% ← Use arithmetic; excess is an additive concept
Fund CAGR: [(1.35)(0.85)(1.12)]^(1/3) − 1 = 8.75% Market CAGR: [(1.30)(0.80)(1.08)]^(1/3) − 1 = 4.00% Actual compounded excess: 8.75% − 4.00% = 4.75%
🔑 Use arithmetic mean for excess returns to measure tracking error and alpha stability; use geometric mean to measure actual compounded returns.
8. Common Pitfalls and Misconceptions
Pitfall 1: Substituting AM for CAGR
"This fund averaged 15% annual returns over 10 years." → In the vast majority of cases, CAGR < 15%. Detection method: List each year's returns, compound (1+r) factors, take the nth root, check if it equals 15%.
Pitfall 2: Ignoring the impact of volatility on compounding
Two portfolios, both with AM = 8%: - A: Constant 8% each year (GM = 8%) - B: −10%, +30%, −5%, +25%, −8% (GM = 5.2%)
Same arithmetic mean, vastly different geometric means!
Pitfall 3: Mixing up cross-sectional and time-series data
- 50 funds in the same year → Arithmetic Mean
- One fund over 50 years → Geometric Mean
Pitfall 4: Thinking GM is "more accurate" so always use it
GM is NOT suitable for: cross-sectional comparisons, statistical inference (needs AM for variance), single-period expected forecasts
9. Practice Questions
Multiple Choice
Q1: An investment has 3-year returns of +50%, −50%, +50%. Which statement is correct? - A. AM = 16.67%, terminal value exceeds initial value - B. GM is lower than AM, and AM = 16.67% - C. Both AM and GM equal 50% - D. GM = 16.67%, AM = 50%
Q2: When return volatility (variance) increases, the gap between AM and GM: - A. Shrinks - B. Widens - C. Remains unchanged - D. Depends on whether returns are positive or negative
Q3: Which scenario is best suited for arithmetic mean rather than geometric mean? - A. Calculating a fund's annualized compound return over 10 years - B. Calculating the average return of 50 different tech stocks in 2025 - C. Calculating your portfolio's CAGR since inception - D. Calculating a country's average GDP growth rate over 20 years
Q4: An asset has an annualized AM of 12% and annualized volatility (σ) of 30%. Using the volatility drag approximation GM ≈ AM − σ²/2, the annualized GM is approximately: - A. 12.0% - B. 9.0% - C. 7.5% - D. 16.5%
Q5: Which data type CAN use the geometric mean? - A. Morningstar fund ratings (1-5 stars) - B. Temperatures across cities (Celsius) - C. Annualized stock returns - D. Investment style classification (Growth/Value/Balanced)
Q6: A portfolio's 5-year wealth growth factors (i.e., (1+r) each year) are: 1.10, 0.90, 1.20, 0.95, 1.15. The geometric mean growth rate is closest to: - A. 6.0% - B. 5.3% - C. 10.6% - D. 4.8%
Q7: When all values in a dataset are identical, which holds true? - A. AM > GM - B. GM > AM - C. AM = GM - D. The relationship between AM and GM is indeterminate
Q8: An analyst says: "This fund has an AM of 20% and a CAGR of 15%. The 5 percentage point gap is due to:" - A. Management fees - B. High return volatility creating volatility drag - C. CAGR calculation error - D. The fund using leverage
Answers
Q1: B — AM = (50% − 50% + 50%) / 3 = 16.67%. Growth factor product: 1.5 × 0.5 × 1.5 = 1.125, GM = 1.125^(1/3) − 1 ≈ 4.0%. AM > GM.
Q2: B — Greater volatility means greater Variance Drain. The gap between AM and GM widens. See formula GM ≈ AM − σ²/2.
Q3: B — Returns of 50 different stocks in the same year are cross-sectional data; use arithmetic mean. A, C, and D are all time-series compound growth scenarios requiring geometric mean.
Q4: C — GM ≈ 12% − (0.30² / 2) = 12% − (0.09/2) = 12% − 4.5% = 7.5%. Just 30% volatility eats 4.5 percentage points of compound return!
Q5: C — Annualized stock returns are ratio-scale data with an absolute zero; geometric mean applies. Ratings are ordinal, temperature is interval, classification is nominal.
Q6: B — Product: 1.10 × 0.90 × 1.20 × 0.95 × 1.15 = 1.29843. GM = 1.29843^(1/5) − 1 = 1.0536 − 1 ≈ 5.36%, closest to 5.3%.
Q7: C — When all data points are identical, AM = GM. This is the only condition where the equality in AM ≥ GM holds.
Q8: B — The gap between AM and CAGR (GM) comes from volatility drag. The more volatile the returns, the wider the gap. Management fees affect both simultaneously; leverage amplifies returns but does not directly cause the AM-GM gap.
10. Key Exam Takeaways
| Priority | Topic | Key Memory Aid |
|---|---|---|
| ⭐⭐⭐ | AM vs GM scenario judgment | Cross-sectional → AM / Time series → GM |
| ⭐⭐⭐ | Volatility drag: GM ≈ AM − σ²/2 | Each unit ↑ in volatility → 0.5 units ↓ in compound return |
| ⭐⭐ | AM ≥ GM, equality only when all values identical | Tested as True/False |
| ⭐⭐ | GM only for ratio-scale data | Cannot use on nominal/ordinal/interval scales |
| ⭐⭐ | CAGR = GM − 1 | The correct measure of fund returns |
| ⭐ | GM calculation: compound (1+r), nth root, subtract 1 | Frequently appears in calculation problems |
📊 Sindy's Investment Note: Next time someone tells you "average annual returns of XX%," your first reaction shouldn't be "great returns" — it should be "show me each year's returns." The geometric mean is the money in your account. The arithmetic mean? That's marketing copy. 🌹