1. Sampling Methods (LOS 7.

a)

The core idea is how we select a sample from a population. The

method we choose determines how much we can trust our results.
The primary distinction is between probability sampling (good) and
nonprobability sampling (less reliable).

Probability Sampling (Random Selection)

1. Simple Random Sampling: This is the baseline. Every

member of the population has an equal chance of being
selected. Think of drawing names from a hat. It's unbiased but
may not be the most efficient.

2. Systematic Sampling: A practical shortcut to random

sampling. You select every kth member from a population (e.g.,
every 50th name on a list). It's simple but can be biased if
there's an underlying pattern in the data that aligns with your
selection interval.

3. Stratified Random Sampling: This is a highly important and

testable concept. You first divide the population into subgroups
(strata) based on specific characteristics (e.g., bond ratings,
industry sectors, market cap). Then, you draw random
samples from each subgroup in proportion to that group's size
in the overall population.

o Purpose: To ensure the sample is a representative "mini-

version" of the population. If 30% of your bond universe is
AAA-rated, a stratified sample will ensure 30% of your
sample bonds are also AAA-rated.

o Advantage: Reduces sampling error and provides a more

precise estimate of the population mean compared to
simple random sampling.

4. Cluster Sampling: The population is divided into subgroups

(clusters), often based on geography. Unlike stratified sampling,
you then randomly select some of the clusters and sample
every observation within those selected clusters (one-stage) or
randomly sample from within the selected clusters (two-stage).

o Purpose: To save time and money when the population is

geographically dispersed.

o Disadvantage: Increases sampling error because the

clusters may not be representative of the overall
population.
Exam Trap (Distractor): Stratified vs. Cluster Sampling
This is a classic point of confusion that the CFAI loves to test.

 The Distractor: A question will describe a process of dividing a

population into subgroups and then randomly selecting some of
the subgroups to analyze. It will then ask you to identify this
method as "stratified sampling." This is incorrect.

 The Distinction:

o Stratified = Sample from ALL groups.

o Cluster = Randomly select SOME groups.

Nonprobability Sampling (Non-Random Selection)

1. Convenience Sampling: Selecting a sample based on ease of

access (e.g., the first 20 people you meet). It is easy but often
highly biased.

2. Judgmental Sampling: The researcher uses their expertise to

select a sample they believe is representative. It can be
efficient but is also subject to the researcher's bias.

2. The Central Limit Theorem and Standard Error (LOS 7.b)

This is one of the most important theorems in statistics.

Central Limit Theorem (CLT): The CLT states that for any
population with a finite variance, the sampling distribution of the
sample mean will be approximately normal if the sample size (n) is
sufficiently large (typically n ≥ 30).

Why this is so important: The CLT allows us to use the predictable

properties of the normal distribution to make inferences about the
population mean, even if the underlying population itself is not
normally distributed.

Key Distinction: Standard Deviation vs. Standard Error

This is a critical distinction that is frequently tested.

 Standard Deviation (s or σ): Measures the dispersion

of individual observations within a sample or population. It
answers, "How spread out is the data?"

 Standard Error of the Sample Mean (sₓ or σₓ): Measures

the dispersion of sample means around the true population
mean. It is the standard deviation of the sampling distribution. It
answers, "How much can I expect my sample mean to vary from
 the true population mean?" It quantifies the precision of your
estimate.

Formula for Standard Error:

s
s x́ =
√n

Where s is the sample standard deviation and n is the sample size.

Interpretation: As the sample size (n) increases, the standard error

decreases. A larger sample gives you a more precise estimate of the
true population mean.

Exam Trap (Distractor): A question will provide the sample

standard deviation (s) and ask about the precision or confidence in
the sample mean. The distractor answer will be based on using the
standard deviation directly. The correct approach requires you to first
calculate the standard error by dividing the sample standard
deviation by the square root of the sample size.

3. Resampling Methods (LOS 7.c)

Resampling methods use the original sample to generate new

samples, allowing us to estimate the sampling distribution of a
statistic without relying on theoretical formulas.

1. Bootstrap: The primary resampling method in the curriculum.

It involves drawing repeated samples of size n with
replacement from the original sample. "With replacement" is
the key phrase—it means an observation can be selected
multiple times in a single resampled dataset. This is
computationally intensive but very powerful as it makes no
assumptions about the underlying population's distribution.

2. Jackknife: A simpler method where you create n samples by

taking the original sample and deleting one observation at a
time. It is less computationally demanding but also less
versatile than the bootstrap method.

Exam Trap (Distractor): The distractor will confuse the resampling

methods. It might describe the "with replacement" process of
bootstrapping but call it the jackknife. Remember: Bootstrap = With
Replacement.

Module 8.1: The Hypothesis Testing Framework (LOS 8.a)

This is the 7-step process for conducting any hypothesis test.

1. State the Hypotheses:

o Null Hypothesis (H₀): The hypothesis you want

to reject. It is the "dull" hypothesis, the status quo.
It always includes the "equals" sign (e.g., =, ≤, or ≥).
It is the statement that is actually being tested.

o Alternative Hypothesis (Hₐ): The hypothesis you want

to support. It is what you conclude if you reject the null.
It never includes the "equals" sign (e.g., ≠, <, or >).

2. Select the Appropriate Test Statistic: Choose the right tool

for the job (z, t, χ², F) based on what you are testing (mean,
variance) and the characteristics of the sample.

3. Specify the Significance Level (α): This is the probability of

making a Type I error. It is your threshold for "reasonable
doubt." Common levels are 1%, 5%, and 10%.

4. State the Decision Rule: Based on the test statistic and

significance level, determine the critical value(s) from the
appropriate statistical table. The decision rule is your pre-
defined condition for rejecting H₀ (e.g., "Reject H₀ if the test
statistic is > 1.96 or < -1.96").

5. Collect Data and Calculate the Test Statistic: Compute the

value from your sample data. The general formula is:

Sample Statistic−Hypothesized Value

Test Statistic=
Standard Error of the Sample Statistic
6. Make the Statistical Decision: Compare the calculated test
statistic to the critical value(s). If it falls in the rejection region,
reject H₀. Otherwise, you "fail to reject" H₀.

7. Make the Economic Decision: Translate the statistical

decision back into the context of the investment problem.

Key Concepts: Errors, Significance, and Power

This is a highly testable area focused on definitions.

 Type I Error: Rejecting the null hypothesis when it is actually

true. The probability of a Type I error is the significance level
(α).
  Type II Error: Failing to reject the null hypothesis when it is
actually false. The probability of a Type II error is denoted by β.

 Power of a Test: The probability of correctly rejecting a false

null hypothesis. This is what you want your test to do. Power =
1 - β.

The Trade-off: For a given sample size, there is a trade-off between

Type I and Type II errors.

 Decreasing your significance level (e.g., from 5% to 1%) makes

it harder to reject the null. This decreases the probability of a
Type I error but increases the probability of a Type II error,
thus reducing the power of the test.

Exam Trap (Distractor): A question will state that the power of a

test is 40% and the significance level is 5%. It will then ask for the
probability of correctly rejecting a true null.

 The Distractor: The candidate sees "correctly rejecting" and

thinks of the power (40%).

 The Reality: You can never "correctly reject" a true null.

Rejecting a true null is a Type I error. The question is a trick. The
probability of rejecting a true null is simply the significance
level, 5%.

Module 8.2: Types of Hypothesis Tests (LOS 8.b & 8.c)

This section details which statistical tool to use for which job. This is
the most important part of the reading for exam purposes.

Summary of Use Cases

Test Key Degrees of

What are you testing? to Assumption Freedom
Use s (df)

Mean of a single t-test Population is n-1

population normal (or
approx.
normal).
Use z-test if
population
variance is
known OR
sample size
is large (n ≥
 30).

Populations
Difference between
are normal;
means of
t-test variances n₁ + n₂ - 2
two independent populati
are assumed
ons
equal.

Mean of
Differences n - 1 (where
differences between
t-test are normally n is # of
two dependent (paired)
distributed. pairs)
populations

Chi-
Population is
Variance of a single squar
normally n-1
population e (χ²)
distributed.
test

Equality of Numerator:
Populations
variances between n₁ - 1,
F-test are normally
two independent populati Denominato
distributed.
ons r: n₂ - 1

Key Distinctions and Exam Focus

1. Tests for Means: Independent vs. Dependent Samples

o Use Case (Independent): Comparing the mean returns

of two different groups, e.g., US equity funds vs. European
equity funds. The performance of one group does not
systematically affect the other.

o Use Case (Dependent/Paired): Comparing the mean

returns of the same group of companies before and
after an event (e.g., a change in CEO). The samples are
dependent because they are the same companies. This is
a paired comparisons test.

o Exam Trap (Distractor): A question will describe a

"before and after" scenario and provide a distractor
answer based on the t-test for independent samples. You
must recognize that this requires a paired comparisons
test.

2. Tests for Variance: One Population vs. Two Populations

 o Use Case (Chi-square): Testing if the volatility
(variance) of a single mutual fund is equal to a specific
value, e.g., σ² = 0.05.

o Use Case (F-test): Testing if the volatilities

of two mutual funds are equal to each other, σ₁² = σ₂².
This is often a preliminary step to decide if you can use
the "pooled variance" version of the t-test for difference in
means.

o Exam Trap (F-test): The F-test statistic is a ratio of

variances: F = s₁² / s₂². The convention is to always put
the larger sample variance in the numerator. This
simplifies the test to a one-tailed (right-tail) test and is a
critical detail for finding the correct critical F-value.

3. Parametric vs. Nonparametric Tests (LOS 8.c)

o Parametric Tests: The tests listed above (t, z, χ², F) are

all parametric. They require assumptions about the
distribution of the population (usually that it's normal).

o Nonparametric Tests: Used when the assumptions for

parametric tests are not met.

 When to use:

1. The data is ranked (ordinal scale), not values.

2. The population distribution is clearly not

normal, and the sample size is small.

3. The hypothesis doesn't concern a parameter

(e.g., testing if data is random).

o Exam Focus: This is primarily a conceptual topic. You

need to know when a nonparametric test is appropriate,
not necessarily how to calculate one (with the exception
of the Spearman rank correlation from Reading 9).