定量方法（Quantitative Methods）— 描述性统计模块
一、背景：为什么需要切比雪夫不等式？
前面我们学了均值、标准差、偏度、峰度——这些工具告诉我们一堆数据的"形状"。但有一个问题还没有回答：
📌 核心问题：给定任意一个分布，有多大比例的数据落在均值 ± k 个标准差之内？
如果你是正态分布，这个比例是可以精确计算的（经验法则：68-95-99.7）。但现实中，收益率分布往往不是正态的——偏的、肥尾的、完全不对称的。
这时候就需要切比雪夫不等式。
二、切比雪夫不等式的核心陈述
定理（CFA 一级版本）：
对于 任何 具有有限均值和方差的分布，至少有 $(1 - \frac{1}{k^2})$ 比例的数据落在均值 ± k 个标准差范围内（其中 $k > 1$）。
即：
$$P(|X - \mu| \leq k\sigma) \geq 1 - \frac{1}{k^2}$$
🧠 记忆口诀：任意分布不用愁，k 倍标准差保底有，比例至少 1 - 1/k²。
三、关键特点（CFA 高频考点）
⭐ 三个核心特征：
- 不限制分布形态：偏态、肥尾、低峰态、任何形状——都能用！
- 下限估计（保守）：切比雪夫给出的是 "至少"，实际比例只会比它更大
- 只需 $k > 1$：$k \leq 1$ 时结果无意义（$1 - 1/k² \leq 0$）
💡 对比经验法则（Empirical Rule）：经验法则只适用于 正态分布，而且是精确值（不是下限）。切比雪夫适用于 任何分布，但只是保守的下限。两者适用条件不同！
四、关键 k 值的应用表
这是 CFA 考试必记的名词：
| k | 均值 ± kσ 范围 | 切比雪夫：至少比例 | 正态分布（经验法则） |
|---|---|---|---|
| 1 | μ ± 1σ | 无意义（0%） | ≈ 68% |
| 1.25 | μ ± 1.25σ | ≥ 36% | ≈ 79% |
| 1.5 | μ ± 1.5σ | ≥ 55.6% | ≈ 87% |
| 2 | μ ± 2σ | ≥ 75% | ≈ 95% |
| 2.5 | μ ± 2.5σ | ≥ 84% | ≈ 98.8% |
| 3 | μ ± 3σ | ≥ 88.9% | ≈ 99.7% |
| 4 | μ ± 4σ | ≥ 93.75% | ≈ 99.99% |
🔴 关键观察：k 越大，切比雪夫的下限越"慷慨"，但与正态分布的实际比例差距也越大。因为切比雪夫是"最坏情况"——它考虑了一切可能分布，包括那些极端难看的。
五、公式变形与应用
由「范围→比例」
已知：k = 2（均值 ± 2σ），问至少有多少数据在此范围内？
$$1 - \frac{1}{2^2} = 1 - \frac{1}{4} = 0.75 = 75\%$$
由「比例→范围」（反向用法）
已知：想保证至少 80% 的数据在范围内，需要 k 多大？
$$1 - \frac{1}{k^2} = 0.80 \quad \Rightarrow \quad \frac{1}{k^2} = 0.20 \quad \Rightarrow \quad k^2 = 5 \quad \Rightarrow \quad k = \sqrt{5} \approx 2.236$$
🧠 求 k 的通用公式：
$$k = \frac{1}{\sqrt{1 - p}}$$
其中 $p$ 是保底比例。
六、实战案例分析
案例 1：投资组合风险控制
某基金经理统计了旗下 500 只基金的年度收益率，发现平均收益率为 8%，标准差为 12%，但分布严重右偏（有少数基金年度涨幅超过 80%）。
问题：管理层想知道，无论这些基金收益率的分布长什么样，至少有多少比例基金的年度收益率落在 −16% 到 +32% 之间？
解题：
- 首先确定 k： $$k = \frac{32\% - 8\%}{12\%} = \frac{24\%}{12\%} = 2$$
验证下端：$\frac{8\% - (-16\%)}{12\%} = \frac{24\%}{12\%} = 2$ ✓ （两端距均值都是 2 个标准差，对称范围）
- 应用切比雪夫： $$1 - \frac{1}{2^2} = 1 - \frac{1}{4} = 75\%$$
答案：无论分布多奇怪，至少 75% 的基金收益率在 −16% ~ +32% 之间。
💡 如果用正态分布（经验法则）算，会是 ≈ 95%，但该分布严重右偏 → 正态分布假设不成立 → 必须用切比雪夫！
案例 2：反向求解——设定风险预算
某风控总监要求：无论如何（任何分布），至少 90% 的日收益率必须落在设定的风险区间内。日均收益率为 0.05%，标准差为 1.5%。
问题：风险区间的上下界应该设为多少？
解题：
-
先求 k： $$1 - \frac{1}{k^2} = 0.90 \quad \Rightarrow \quad \frac{1}{k^2} = 0.10 \quad \Rightarrow \quad k = \sqrt{10} \approx 3.16$$
-
计算区间：
- 上界：$0.05\% + 3.16 \times 1.5\% = 0.05\% + 4.74\% = 4.79\%$
- 下界：$0.05\% - 3.16 \times 1.5\% = 0.05\% - 4.74\% = -4.69\%$
答案：风险区间设为 −4.69% ~ +4.79%，可以保证至少 90% 的日收益率落在此范围内（适用于任何分布）。
七、切比雪夫 vs 经验法则 vs 实际分布
CFA 考试中高频率出现的对比题：
| 特征 | 切比雪夫不等式 | 经验法则 |
|---|---|---|
| 适用范围 | 任何分布 | 仅正态分布 |
| 给出的结果 | 保底下限（"至少"） | 近似精确值（"约"） |
| 需要 k | k > 1 | 任意 k |
| 保守程度 | 非常保守 | 刚好准确（在正态下） |
| 典型用途 | 非正态/未知分布的风险估计 | 近似正态数据的快速估计 |
CFA 考试判断流程：
该数据是否近似正态？
├─ 是 → 用经验法则（更快更精确）
└─ 不确定 / 明显非正态 / 题目要求"最保守估计 / 任意分布 / 至少"
→ 用切比雪夫不等式
八、常见易错点总结
| 易错点 | 正确理解 |
|---|---|
| "切比雪夫给出的是精确比例" | ❌ 给出的是 下限（至少），不是精确值 |
| "k=1 时有 0% 的数据在 μ±σ 内" | ❌ 切比雪夫在 k=1 时无意义，不是说没数据 |
| "经验法则适用于所有分布" | ❌ 经验法则仅适用于正态/近似正态分布 |
| "切比雪夫比经验法则更准确" | ❌ 切比雪夫更 保守（下限值），经验法则在正态下更 精确 |
| "求 k 时忘记开根号" | 常见计算错误：$k^2 = 5$ → $k = 5$ ❌，应为 $k = \sqrt{5}$ |
| "分布是偏的就不能用切比雪夫" | ❌ 恰恰相反，偏的、肥尾的分布正是切比雪夫的用武之地 |
九、测试题
题目 1
根据切比雪夫不等式，对于任意分布，至少有 84% 的数据落在均值 ± k 个标准差范围内。k 最接近：
A. 2.0 B. 2.5 C. 3.0 D. 4.0
题目 2
某分析师正在分析一只小型成长股的日收益率。样本均值为 0.03%，标准差为 2.5%。该分析师不确定收益率是否服从正态分布（可能肥尾）。
要估计至少 75% 的日收益率落在什么范围内，应使用：
A. 经验法则，区间为 [−2.47%, +2.53%] B. 切比雪夫不等式，区间为 [−4.97%, +5.03%] C. 经验法则，区间为 [−4.97%, +5.03%] D. 切比雪夫不等式，区间为 [−7.47%, +7.53%]
题目 3
关于切比雪夫不等式，以下说法 正确 的是：
A. 当数据为正态分布时，切比雪夫不等式的估计比经验法则更精确 B. 切比雪夫不等式要求数据至少近似对称 C. 当 k = 1.5 时，至少有约 55.6% 的数据落在 μ ± 1.5σ 范围内，无论分布形态如何 D. 切比雪夫不等式只能用于峰度小于 3 的分布
题目 4（应用题）
某风控分析师被要求出具一份极端保守的风险报告：保证至少 95% 的月度收益率落在某一区间内，且该区间适用于任何分布形态。月均收益率为 0.6%，月标准差为 4.0%。
该区间的宽度（上界 − 下界）最接近：
A. 16.0% B. 25.3% C. 35.8% D. 72.0%
十、答案与解析
答案 1：B — 2.5
设 $1 - 1/k^2 = 0.84$，则 $1/k^2 = 0.16$，$k^2 = 1/0.16 = 6.25$，$k = \sqrt{6.25} = 2.5$。
答案 2：B — 切比雪夫不等式，区间为 [−4.97%, +5.03%]
分析师不确定分布是否正态（可能肥尾）→ 不能用经验法则，必须用保守的切比雪夫。
75% → $k = 2$。区间：$0.03\% \pm 2 \times 2.5\% = [−4.97\%, +5.03\%]$。
答案 3：C — 当 k = 1.5 时，至少有约 55.6% 的数据落在 μ ± 1.5σ 范围内，无论分布形态如何
A ❌：正态分布下经验法则更精确；B ❌：切比雪夫不要求对称；D ❌：切比雪夫对任何有穷方差的分布都适用。C 正确：$1 - 1/1.5^2 = 1 - 1/2.25 = 1 - 0.444 = 55.6\%$。
答案 4：C — 35.8%
设 $1 - 1/k^2 = 0.95$，$1/k^2 = 0.05$，$k^2 = 20$，$k = \sqrt{20} \approx 4.472$。
区间宽度 = 上界 − 下界 = $2 \times k \times \sigma = 2 \times 4.472 \times 4.0\% = 35.78\% \approx 35.8\%$。
📌 今日要点记住三句话： 1. 切比雪夫给出的是 任意分布 下数据落在 μ±kσ 内的 最低保底比例：$1 - 1/k^2$ 2. 当不确定分布是否正态时 → 用切比雪夫而非经验法则（更保守、更普适） 3. 反向公式记牢：$k = 1/\sqrt{1-p}$，求区间宽度 = $2k\sigma$
L111 切比雪夫不等式 | 2026-07-17 | CFA Level 1 定量方法
Quantitative Methods — Descriptive Statistics Module
1. Background: Why Do We Need Chebyshev's Inequality?
We have learned about the mean, standard deviation, skewness, and kurtosis—these tools tell us the "shape" of a dataset. But one question remains unanswered:
📌 The Core Question: Given any distribution, what proportion of data falls within ±k standard deviations of the mean?
If the distribution is normal, we can calculate this precisely (Empirical Rule: 68-95-99.7). But in reality, return distributions are often not normal—skewed, fat-tailed, or completely asymmetric.
This is where Chebyshev's Inequality comes in.
2. Core Statement of Chebyshev's Inequality
Theorem (CFA Level 1 Version):
For any distribution with finite mean and variance, at least $(1 - \frac{1}{k^2})$ proportion of the data falls within ±k standard deviations of the mean (where $k > 1$).
That is:
$$P(|X - \mu| \leq k\sigma) \geq 1 - \frac{1}{k^2}$$
🧠 Memory Aid: For any distribution, the proportion within k standard deviations is at least 1 − 1/k².
3. Key Characteristics (High-Frequency CFA Exam Points)
⭐ Three Core Features:
- No Restriction on Distribution Shape: Skewed, fat-tailed, platykurtic, any shape—Chebyshev still works!
- Lower-Bound Estimate (Conservative): Chebyshev gives an "at least" figure; the actual proportion can only be larger.
- Requires $k > 1$ Only: When $k \leq 1$, the result is meaningless ($1 - 1/k^2 \leq 0$).
💡 Contrast with the Empirical Rule: The Empirical Rule applies only to normal distributions and gives approximate exact values (not lower bounds). Chebyshev applies to any distribution but provides only a conservative lower bound. Different applicability conditions!
4. Key k-Value Application Table
These are must-memorize values for the CFA exam:
| k | Range μ ± kσ | Chebyshev: At Least | Normal (Empirical Rule) |
|---|---|---|---|
| 1 | μ ± 1σ | Meaningless (0%) | ≈ 68% |
| 1.25 | μ ± 1.25σ | ≥ 36% | ≈ 79% |
| 1.5 | μ ± 1.5σ | ≥ 55.6% | ≈ 87% |
| 2 | μ ± 2σ | ≥ 75% | ≈ 95% |
| 2.5 | μ ± 2.5σ | ≥ 84% | ≈ 98.8% |
| 3 | μ ± 3σ | ≥ 88.9% | ≈ 99.7% |
| 4 | μ ± 4σ | ≥ 93.75% | ≈ 99.99% |
🔴 Key Observation: As k increases, Chebyshev's lower bound becomes more "generous," but the gap from the normal distribution's actual proportion also widens. This is because Chebyshev considers the "worst-case scenario"—it accounts for every possible distribution, including extremely ugly ones.
5. Formula Transformations and Applications
From "Range → Proportion"
Given: k = 2 (mean ± 2σ). What is the minimum proportion of data within this range?
$$1 - \frac{1}{2^2} = 1 - \frac{1}{4} = 0.75 = 75\%$$
From "Proportion → Range" (Reverse Application)
Given: we want at least 80% of data within the range. What k is needed?
$$1 - \frac{1}{k^2} = 0.80 \quad \Rightarrow \quad \frac{1}{k^2} = 0.20 \quad \Rightarrow \quad k^2 = 5 \quad \Rightarrow \quad k = \sqrt{5} \approx 2.236$$
🧠 General Formula for k:
$$k = \frac{1}{\sqrt{1 - p}}$$
where $p$ is the guaranteed minimum proportion.
6. Practical Case Analysis
Case 1: Portfolio Risk Control
A fund manager compiled the annual returns of 500 funds under management, finding an average return of 8% and a standard deviation of 12%. However, the distribution is severely right-skewed (a few funds had annual gains exceeding 80%).
Question: Management wants to know, regardless of the shape of the return distribution, what is the minimum proportion of funds whose annual returns fall between −16% and +32%?
Solution:
- First, determine k: $$k = \frac{32\% - 8\%}{12\%} = \frac{24\%}{12\%} = 2$$
Verify the lower bound: $\frac{8\% - (-16\%)}{12\%} = \frac{24\%}{12\%} = 2$ ✓ (Both ends are 2 standard deviations from the mean; symmetric range.)
- Apply Chebyshev: $$1 - \frac{1}{2^2} = 1 - \frac{1}{4} = 75\%$$
Answer: No matter how strange the distribution, at least 75% of fund returns fall between −16% and +32%.
💡 If we used the normal distribution (Empirical Rule), it would be ≈ 95%, but the distribution is severely right-skewed → the normal distribution assumption does not hold → must use Chebyshev!
Case 2: Reverse Calculation—Setting a Risk Budget
A risk director mandates: no matter what (any distribution), at least 90% of daily returns must fall within a specified risk interval. The average daily return is 0.05%, with a standard deviation of 1.5%.
Question: What should the upper and lower bounds of the risk interval be?
Solution:
-
First, find k: $$1 - \frac{1}{k^2} = 0.90 \quad \Rightarrow \quad \frac{1}{k^2} = 0.10 \quad \Rightarrow \quad k = \sqrt{10} \approx 3.16$$
-
Calculate the interval:
- Upper bound: $0.05\% + 3.16 \times 1.5\% = 0.05\% + 4.74\% = 4.79\%$
- Lower bound: $0.05\% - 3.16 \times 1.5\% = 0.05\% - 4.74\% = -4.69\%$
Answer: The risk interval should be set at −4.69% to +4.79%, ensuring at least 90% of daily returns fall within this range (applicable to any distribution).
7. Chebyshev vs. Empirical Rule vs. Actual Distribution
A high-frequency comparison in the CFA exam:
| Feature | Chebyshev's Inequality | Empirical Rule |
|---|---|---|
| Scope of Application | Any distribution | Normal distributions only |
| Result Type | Conservative lower bound ("at least") | Approximate exact value ("about") |
| Requires k | k > 1 | Any k |
| Conservatism | Very conservative | Exactly accurate (under normality) |
| Typical Use | Risk estimation for non-normal/unknown distributions | Quick estimation for approximately normal data |
CFA Exam Decision Flow:
Is the data approximately normal?
├─ Yes → Use the Empirical Rule (faster, more precise)
└─ Uncertain / Clearly non-normal / Question asks for "most conservative estimate / any distribution / at least"
→ Use Chebyshev's Inequality
8. Common Pitfalls Summary
| Pitfall | Correct Understanding |
|---|---|
| "Chebyshev gives exact proportions" | ❌ It gives a lower bound ("at least"), not an exact value |
| "At k=1, 0% of data is within μ±σ" | ❌ Chebyshev is meaningless at k=1; it does not mean there is no data |
| "The Empirical Rule applies to all distributions" | ❌ The Empirical Rule applies only to normal / approximately normal distributions |
| "Chebyshev is more accurate than the Empirical Rule" | ❌ Chebyshev is more conservative (lower bound); the Empirical Rule is more precise under normality |
| "Forgetting to take the square root when solving for k" | Common calculation error: $k^2 = 5$ → $k = 5$ ❌, should be $k = \sqrt{5}$ |
| "Chebyshev cannot be used on skewed distributions" | ❌ On the contrary, skewed and fat-tailed distributions are exactly where Chebyshev shines |
9. Practice Questions
Question 1
According to Chebyshev's Inequality, for any distribution, at least 84% of the data falls within ±k standard deviations of the mean. k is closest to:
A. 2.0 B. 2.5 C. 3.0 D. 4.0
Question 2
An analyst is examining the daily returns of a small-cap growth stock. The sample mean is 0.03% and the standard deviation is 2.5%. The analyst is uncertain whether the returns follow a normal distribution (they may be fat-tailed).
To estimate the range within which at least 75% of daily returns fall, which approach should be used?
A. Empirical Rule, range [−2.47%, +2.53%] B. Chebyshev's Inequality, range [−4.97%, +5.03%] C. Empirical Rule, range [−4.97%, +5.03%] D. Chebyshev's Inequality, range [−7.47%, +7.53%]
Question 3
Regarding Chebyshev's Inequality, which of the following statements is correct?
A. When data is normally distributed, Chebyshev's Inequality provides a more precise estimate than the Empirical Rule B. Chebyshev's Inequality requires the data to be at least approximately symmetric C. When k = 1.5, at least approximately 55.6% of the data falls within μ ± 1.5σ, regardless of the distribution shape D. Chebyshev's Inequality can only be used for distributions with kurtosis less than 3
Question 4 (Application)
A risk analyst is asked to produce an extremely conservative risk report: ensuring at least 95% of monthly returns fall within a certain interval, and that interval must hold for any distribution shape. The average monthly return is 0.6%, and the monthly standard deviation is 4.0%.
The width of the interval (upper bound − lower bound) is closest to:
A. 16.0% B. 25.3% C. 35.8% D. 72.0%
10. Answers and Explanations
Answer 1: B — 2.5
Set $1 - 1/k^2 = 0.84$, then $1/k^2 = 0.16$, $k^2 = 1/0.16 = 6.25$, $k = \sqrt{6.25} = 2.5$.
Answer 2: B — Chebyshev's Inequality, range [−4.97%, +5.03%]
The analyst is uncertain whether the distribution is normal (may be fat-tailed) → Cannot use the Empirical Rule; must use conservative Chebyshev.
75% → $k = 2$. Interval: $0.03\% \pm 2 \times 2.5\% = [−4.97\%, +5.03\%]$.
Answer 3: C — When k = 1.5, at least approximately 55.6% of the data falls within μ ± 1.5σ, regardless of the distribution shape
A ❌: Under normality, the Empirical Rule is more precise; B ❌: Chebyshev does not require symmetry; D ❌: Chebyshev applies to any distribution with finite variance. C is correct: $1 - 1/1.5^2 = 1 - 1/2.25 = 1 - 0.444 = 55.6\%$.
Answer 4: C — 35.8%
Set $1 - 1/k^2 = 0.95$, $1/k^2 = 0.05$, $k^2 = 20$, $k = \sqrt{20} \approx 4.472$.
Interval width = Upper − Lower = $2 \times k \times \sigma = 2 \times 4.472 \times 4.0\% = 35.78\% \approx 35.8\%$.
📌 Today's Three Key Takeaways: 1. Chebyshev provides the minimum guaranteed proportion of data within μ±kσ for any distribution: $1 - 1/k^2$ 2. When uncertain whether a distribution is normal → Use Chebyshev, not the Empirical Rule (more conservative, more universal) 3. Memorize the reverse formula: $k = 1/\sqrt{1-p}$, and interval width = $2k\sigma$
L111 Chebyshev's Inequality | 2026-07-17 | CFA Level 1 Quantitative Methods