📖 参数 vs 统计量

📌 Topic: Distinguishing Population Truth from Sample Estimate — Crossing from Description to Inference

1. Why Does This Distinction Matter?

In L099 we covered three types of data. Now comes the next critical question: Does the data in your hand represent "everything" or "a part"?

💡 You always have only a sample. But what you want to know is the truth about the population. That is the entire purpose of statistical inference — and the parameter vs. statistic distinction is the language foundation of this framework.

2. Core Definitions

2.1 Parameter

Definition: A numerical measure that describes a characteristic of a population. It is a true but usually unknown constant.

Population vs. Sample: Intuitive Understanding

Population                        Sample
─────────────────────           ─────────────
   All individuals               A subset drawn from it
   Parameter (Greek letters)     Statistic (Latin letters)
   True value, fixed constant    Estimate, varies with sampling
   Usually unknown               What we can calculate

Examples of Populations:

🎯 Key insight: A population is not necessarily "everything that ever existed." In finance, populations are often theoretical — for example, a stock's "true expected return" is a parameter we can never directly observe.

2.2 Statistic

Definition: A numerical measure that describes a characteristic of a sample. It is calculated from sample data, known and computable, and used to estimate the corresponding population parameter.

💡 A statistic is what you can touch; a parameter is what you want to know. Statistical inference is the art of using the touchable to guess the untouchable.

3. Parameter vs. Statistic: Notation Reference Table

This is a symbol system you must memorize for CFA Level 1 — exam questions frequently test whether you understand the difference through notation alone.

Memory Aid

4. Key Distinctions In Depth

4.1 Parameters Are Constants; Statistics Are Random Variables

This is arguably the most important yet most overlooked insight in CFA Level 1 Quant:

Intuitive Understanding:

μ (the average height of all adult males in China) is a fixed number. We may not know exactly what it is, but it is what it is — it does not change.

In contrast, x̄ (the average height from a random sample of 100 people) might be 1.73m, then from another sample of 100 it might be 1.71m, then from yet another sample 1.74m...

μ stays still. x̄ jumps around. That is what "parameter = constant, statistic = random variable" means.

4.2 Why Is a Statistic Random?

Because sampling is random. Each time you draw a different sample from the population, you calculate a different x̄.

These three x̄ values differ, yet all three are estimates of the same μ. This is why a statistic is a random variable.

🎯 The upcoming topics L101–L103 on "sampling distributions" are essentially about studying the distributional behavior of x̄ as a random variable.

5. Real-World Cases

Case 1: Evaluating a Fund Manager's Performance

Background: You want to know a fund manager's "true stock-picking skill" (α).

⚠️ You can only ever see α̂ (the sample estimate), never the true α. This leads to the most fundamental question in finance: Does α̂ > 0 mean the manager truly has skill (α > 0), or is it just luck (sampling error)?

🎯 This is the core logic of the hypothesis testing module in CFA Level 1. L100 first gets you clear on parameters vs. statistics so you can later understand what hypothesis testing is actually testing.

Case 2: Equity Index Valuation

Case 3: Consumer Survey

6. Sampling Error

6.1 Definition

Sampling Error = Statistic − Parameter

6.2 Key Properties

🎯 Sampling error is not "doing something wrong" — it is the inevitable consequence of "not looking at everything." The goal of statistics is not to eliminate it but to quantify and control it.

6.3 Sampling Error vs. Non-Sampling Error

⚠️ CFA Level 1 likes to test this distinction. Sampling error is random and quantifiable; non-sampling error is systematic and hard to quantify.

7. Why Does This Distinction Matter for Investment Analysis?

🎯 The most expensive mistake in finance: treating a statistic as a parameter. You think you've found a profitable strategy (in-sample Sharpe = 2.0), when the parameter (true Sharpe) may be 0.2.

8. Typical CFA Level 1 Exam Questions

Example 1: Notation Recognition

An analyst's report uses the following symbols: μ, σ, x̄, s. Which of these represent population parameters?

A. x̄ and s B. μ and σ C. μ and x̄

Answer: B → Greek letters (μ, σ) = population parameters; Latin letters (x̄, s) = sample statistics.

Example 2: Conceptual Distinction

Which of the following statements about parameters and statistics is correct?

A. A parameter can be directly computed from a sample B. A statistic measures a characteristic of a population C. A parameter is a constant, and a statistic is a random variable

Answer: C → A is wrong (parameters cannot be directly computed), B is wrong (statistics measure samples, parameters measure populations), C is correct.

Example 3: Scenario Application

An analyst wants to estimate the average P/E ratio of all companies listed on the NYSE. She randomly selects 100 companies and calculates an average P/E of 18.5. In this scenario:

(i) What is the population? (ii) Is 18.5 a parameter or a statistic? (iii) What does μ represent?

Answer: - (i) Population: All companies listed on the NYSE - (ii) 18.5 is a statistic (x̄), because it was calculated from a sample of 100 companies - (iii) μ represents the true average P/E of all NYSE-listed companies (a parameter, unknown)

Example 4: Sampling Error

Suppose a stock's true annualized volatility σ = 25%. An analyst uses 60 monthly returns to compute a sample standard deviation s = 22%. The sampling error is:

A. 3% B. −3% C. Cannot be determined without knowing the sample mean

Answer: B → Sampling Error = Statistic − Parameter = 22% − 25% = −3%.

⚠️ Sampling error has nothing to do with the "mean" — you do not need x̄. Error depends only on comparing the statistic to the parameter.

9. Practice Questions (5 Questions)

Question 1

Which of the following best describes a statistic?

A. The true mean of a population B. A numerical value computed from sample data, used to estimate a population parameter C. A constant that can never be observed

Question 2

An analyst collects dividend yield data for 100 companies in the S&P 500 and calculates an average dividend yield of 2.3%. In this scenario, 2.3% is a:

A. Parameter B. Statistic C. Neither a parameter nor a statistic

Question 3

Which of the following notation pairs is correct?

A. Population mean = x̄, Sample mean = μ B. Population standard deviation = s, Sample standard deviation = σ C. Population mean = μ, Sample mean = x̄

Question 4

Regarding sampling error, which of the following statements is correct?

A. Sampling error can be completely eliminated by increasing the sample size B. Sampling error equals the statistic minus the parameter C. Sampling error is a type of non-sampling error

Question 5

Suppose a population has a true mean μ = 50. A researcher draws four different samples and computes x̄₁ = 52, x̄₂ = 48, x̄₃ = 51, x̄₄ = 49. Which of the following is correct?

A. μ varies across different samples B. The variation in x̄ demonstrates that a statistic is a random variable C. The variation in sample means indicates that the parameter is also random

10. Answers and Explanations

Question 1 — Answer: B

Explanation:

• ❌ A: This describes a parameter (population true mean = μ)
• ✅ B: Correct. A statistic is computed from sample data to estimate a population parameter
• ❌ C: This is closer to the description of a parameter (usually unknown, but it is a constant, not "never observable" — a census could observe it)

Question 2 — Answer: B (Statistic)

Explanation:

2.3% was calculated from 100 companies (a sample, not all 500).

• ❌ A: The parameter would be the true average dividend yield of all 500 companies (unknown). 2.3% is only the sample mean of 100 companies — a statistic (x̄).
• ✅ B: Correct.
• ❌ C: It is a statistic.

Question 3 — Answer: C

Explanation:

• ❌ A: Reversed — population mean = μ, sample mean = x̄
• ❌ B: Reversed — population standard deviation = σ, sample standard deviation = s
• ✅ C: Correct.

Question 4 — Answer: B

Explanation:

• ❌ A: Sampling error cannot be completely eliminated — as long as you sample instead of census, it exists. Increasing sample size only reduces it.
• ✅ B: Correct. Sampling error is defined as Statistic − Parameter.
• ❌ C: Sampling error is not a type of non-sampling error. They are parallel concepts, not hierarchical.

Question 5 — Answer: B

Explanation:

• ❌ A: μ is a constant (50) and does not vary across samples
• ✅ B: Correct. Four different samples produced four different x̄ values, vividly demonstrating that a statistic is a random variable — this is the foundation of "sampling distribution."
• ❌ C: The parameter (μ) is not random; it is fixed at 50.

11. Key Takeaways

Summary Table

Three-Step Identification Method

1. Check the symbol → Greek letter = Parameter, Latin letter = Statistic
2. Check the source → From the whole = Parameter, From a sample = Statistic
3. Check the nature → Fixed = Parameter, Varies with sampling = Statistic

Three Takeaways

12. Next Lesson Preview

L101 Sampling and Sampling Distributions — If x̄ is a random variable, what does its distribution look like? The Central Limit Theorem tells us: no matter what shape the population takes, the distribution of the sample mean eventually approaches normality. This is one of the most powerful tools in statistical inference.

📊 Core Creed: Greek letters are truth; Latin letters are estimates. Parameters stay fixed; statistics fluctuate with sampling. Crossing from description to inference — this step opens the real door to CFA Level 1 Quant.