Excellent choice ✅

We’ll go module by module — written in Roman numerals (I, II, III, etc.) — just like a proper
CFA-style study summary.

Here’s what I’ll do for you:

 Cover every formula, concept, and example from the PDF.

 Use clear bullet points, definitions, and short notes for memorization.
 Add Professor’s Notes, Examples, and Exam Tips where needed.
 No detail will be skipped.

Let’s begin with Module I (10.1): Linear Regression Basics.

Ⅰ. MODULE 10.1 — LINEAR REGRESSION BASICS

🎯 Learning Outcome (LOS 10.a)

Describe a simple linear regression model, how the least squares criterion is used to estimate
regression coefficients, and interpret these coefficients.

1. Purpose of Simple Linear Regression

 Explains variation in a dependent variable (Y) using one independent variable (X).
 "Variation" means how much values deviate from the mean — not the same as variance.
 Regression answers:
➤ “What explains the fluctuations in Y?”

2. Key Terms

Term Also Known As Meaning

Explained / Endogenous / Variable whose variation is being
Dependent Variable (Y)
Predicted explained
Independent Variable Explanatory / Exogenous /
Variable used to explain changes in Y
(X) Predictor

3. Example
  Predicting stock returns (Y) using GDP growth (X).
→ Stock returns are dependent; GDP is independent.
 Another example:
→ S&P 500 excess returns (X) → explain ABC stock’s excess returns (Y).

4. Scatter Plot & Correlation

 Plot ABC returns (Y) vs. S&P 500 returns (X).

 Example correlation ≈ 0.40 → positive relation.

5. Simple Linear Regression Equation

[
Y_i = b_0 + b_1 X_i + \varepsilon_i
]
where:

 ( Y_i ): actual dependent variable

 ( X_i ): independent variable
 ( b_0 ): intercept (Y-value when X = 0)
 ( b_1 ): slope coefficient (change in Y per unit change in X)
 ( \varepsilon_i ): residual/error term

6. Estimated Regression Line (Line of Best Fit)

[
\hat{Y_i} = b_0 + b_1 X_i
]

 The hat (ˆ) indicates a predicted value.

 Line minimizes Sum of Squared Errors (SSE):
[
SSE = \sum (Y_i - \hat{Y_i})^2
]
 The method that minimizes SSE is called Ordinary Least Squares (OLS).

7. Formulas for Coefficients

[
b_1 = \frac{Cov(X, Y)}{Var(X)}
]
[
b_0 = \bar{Y} - b_1 \bar{X}
]
→ The regression line always passes through the mean point ((\bar{X}, \bar{Y})).

8. Example Calculation

Given:

 Cov(S&P, ABC) = 0.000336

 Var(S&P) = 0.000522
 Mean S&P = –2.70%, Mean ABC = –4.05%

Then:
[
b_1 = \frac{0.000336}{0.000522} = 0.64
]
[
b_0 = -4.05% - (0.64)(-2.70%) = -2.3%
]

Equation:
[
\hat{Y} = -2.3% + 0.64 X
]

If S&P 500 return = –7.8%,

[
\hat{Y} = -2.3 + 0.64(-7.8) = -7.3%
]
Residual = Actual – Predicted = ( 1.1 - (-7.3) = 8.4% )

9. Interpretation

 Intercept (b₀ = –2.3%): Expected return on ABC when S&P = 0.

 Slope (b₁ = 0.64): When S&P excess return ↑1%, ABC’s excess return ↑0.64%.
10. Professor’s Note

 In finance, the slope (b_1) = Beta (β) of the stock.

o β < 1 → less volatile than market
o β = 1 → average risk
o β > 1 → more volatile
 Statistical significance (tested by t-test or F-test) determines if X truly explains Y.

11. Assumptions of Simple Linear Regression (LOS 10.b)

1. Relationship between X and Y is linear.

2. Variance of residuals is constant (homoskedasticity).
3. Residuals are independent (no autocorrelation).
4. Residuals are normally distributed.

12. Common Issues

 Nonlinearity → use residual plots to detect.

 Heteroskedasticity → unequal variance of errors.
 Serial Correlation (Dependence) → residuals show patterns over time.
 Non-normality → affects hypothesis testing validity.
 Outliers → extreme values distort regression fit.

🧠 Module Quiz 10.1

1. Slope = 10 → For each 1-unit change in X, Y changes by 10 units. ✅

2. Least likely assumption: Dependent variable uncorrelated with residuals ❌ (should be
independent variable).

That’s Module I (10.1) fully covered — every key line, example, and formula from your PDF ✅

Shall I move to Module II (10.2 — ANOVA and Goodness of Fit) next?

Perfect 👌
Let’s continue with Module Ⅱ (10.2): Analysis of Variance (ANOVA) & Goodness of Fit —
this part connects regression output with how well the model explains Y.
We’ll cover formulas, interpretations, F-test, t-test, and examples — exactly as in your PDF.
Ⅱ. MODULE 10.2 — ANALYSIS OF VARIANCE
(ANOVA) & GOODNESS OF FIT
🎯 Learning Outcomes

LOS 10.c:
Calculate and interpret measures of fit, and evaluate tests of fit and regression coefficients.

LOS 10.d:
Describe how ANOVA is used in regression, interpret ANOVA results, and compute the
Standard Error of Estimate (SEE).

1. Purpose of ANOVA in Regression

 ANOVA (Analysis of Variance) breaks down the total variation in the dependent
variable ( Y ) into:
o Explained variation (due to regression)
o Unexplained variation (due to residuals)

This helps evaluate how well X explains Y.

2. Key Sums of Squares

Term Symbol Meaning / Formula

Total variation of Y around its
Total Sum of Squares ( SST = \sum (Y_i - \bar{Y})^2 )
mean
Regression Sum of ( SSR = \sum (\hat{Y_i} - \ Variation explained by the
Squares bar{Y})^2 ) regression line
( SSE = \sum (Y_i - \
Error Sum of Squares Unexplained variation (residuals)
hat{Y_i})^2 )

Relationship:
[
SST = SSR + SSE
]

Total variation = Explained + Unexplained variation

3. Mean Squares

Measure Formula Interpretation

Mean Square Regression (MSR) ( MSR = SSR / k ) Explained variation per variable
Mean Squared Error (MSE) ( MSE = SSE / (n - k - 1) ) Average unexplained variation
SEE (Standard Error of Estimate) ( SEE = \sqrt{MSE} ) Standard deviation of residuals

For simple linear regression (k = 1):

→ Degrees of freedom (df):

 Regression: 1
 Error: ( n - 2 )
 Total: ( n - 1 )

4. Coefficient of Determination (R²)

[
R^2 = \frac{SSR}{SST} = 1 - \frac{SSE}{SST}
]

 Measures the % of total variation in Y explained by X.

 Example: ( R^2 = 0.63 ) → 63% of variation in Y explained by X.
 For simple regression, ( R^2 = r^2 ) (the square of the correlation coefficient).

5. Example — Using the ANOVA Table

Source of Variation df Sum of Squares Mean Squares

Regression (Explained) 1 0.0076 0.0076
Error (Unexplained) 34 0.0406 0.0012
Total 35 0.0482 —

Compute:
[
R^2 = \frac{0.0076}{0.0482} = 0.158 = 15.8%
]
[
SEE = \sqrt{MSE} = \sqrt{0.0012} = 0.035
]
✅ So, the model explains 15.8% of the variation, with a residual standard deviation of 3.5%.
6. F-Test (Overall Significance of Regression)

Used to test whether the model as a whole explains a significant portion of Y’s variation.

Formula:
[
F = \frac{MSR}{MSE} = \frac{SSR/k}{SSE/(n - k - 1)}
]

Hypotheses:
[
H_0: b_1 = 0 \quad (\text{no linear relationship})
]
[
H_1: b_1 \neq 0 \quad (\text{significant relationship})
]

Decision Rule:
Reject ( H_0 ) if ( F > F_{crit} ) at chosen α level.
It’s always a one-tailed test.

Degrees of freedom:

 Numerator = k (number of Xs = 1)
 Denominator = n – k – 1 = n – 2

7. Example — F-Test Calculation

From the earlier table:

[
F = \frac{MSR}{MSE} = \frac{0.0076}{0.0012} = 6.33
]
df₁ = 1, df₂ = 34

At 5% significance → Critical ( F_{0.05,1,34} ≈ 4.1 )

→ Since 6.33 > 4.1, Reject ( H_0 ).
✅ The slope (b₁) is significantly different from zero — meaning S&P 500 returns do explain
ABC stock returns.
8. t-Test (Significance of Individual Coefficient)

Tests whether the slope ( b_1 ) differs significantly from a hypothesized value (often 0).

Formula:
[
t = \frac{b_1 - b_{1,0}}{SE(b_1)}
]
Degrees of freedom: ( n - 2 )

Decision Rule:
Reject ( H_0: b_1 = 0 ) if (|t| > t_{crit}).

Example:

 ( b_1 = 0.64 ), ( SE(b_1) = 0.26 ), ( n = 36 )

[
t = \frac{0.64 - 0}{0.26} = 2.46
]
From table: ( t_{crit} (df = 34, 5%) = 2.03 )
✅ ( 2.46 > 2.03 ) → Reject ( H_0 ), slope is significant.

🔹 In simple regression, F-test ≡ t-test²

(they test the same null: b₁ = 0)

9. Exam Tips

 ( SEE ) → Lower = better fit.

 ( R^2 ) → Higher = stronger explanatory power.
 ( F )-stat significant → regression model explains Y overall.
 ( t )-stat significant → specific slope (β) is meaningful.
 For CFA Level I, interpretation > calculation.

🧠 Module Quiz 10.2

Question Correct Answer Explanation

1 A df = n – 2 for both parameter & predicted Y confidence intervals
2 A To test intercept significance → ( H_0: a_0 = 0 )
3 C Variation explained by X = Regression Sum of Squares (SSR)
That’s Module Ⅱ (ANOVA and Goodness of Fit) — complete, clean, and exam-focused ✅

Would you like me to continue with Module Ⅲ (10.3 — Predicted Values and Functional
Forms of Regression) next?

Excellent 🌟
Now let’s finish strong with the last part — Module Ⅲ (10.3): Predicted Values &
Functional Forms of Regression.
This one focuses on forecasting (predicted Y values), confidence intervals, and different
regression models (log-linear, lin-log, etc.).

We’ll go point by point — all formulas, interpretations, and examples from your CFA text are
included.

Ⅲ. MODULE 10.3 — PREDICTED VALUES &

FUNCTIONAL FORMS OF REGRESSION
🎯 Learning Outcomes

LOS 10.e: Calculate and interpret predicted values and prediction intervals for a dependent
variable.
LOS 10.f: Describe different functional forms of simple linear regressions.

1. Predicted Value (Forecast)

 The regression equation can be used to forecast Y, given a forecast for X.

[
\hat{Y} = b_0 + b_1 X_f
]
Where:

 ( \hat{Y} ): predicted (forecasted) value of dependent variable

 ( b_0, b_1 ): estimated regression coefficients
 ( X_f ): forecasted value of the independent variable

2. Example — Forecasting
Regression equation for ABC stock:
[
ABC = -2.3% + 0.64(\text{S&P 500 excess return})
]

If forecasted S&P 500 excess return = 10%,

then:
[
\hat{Y} = -2.3 + 0.64(10) = 4.1%
]

✅ Predicted ABC excess return = 4.1%

3. Confidence (Prediction) Interval for Forecasts

We don’t just want a single forecast — we want a range that shows how confident we are in that
prediction.

Formula:
[
\hat{Y} \pm t_{\alpha/2, n-2} \times s_f
]

Where:

 ( t_{\alpha/2, n-2} ) = two-tailed critical t-value at desired confidence level

 ( s_f ) = standard error of forecast

4. Standard Error of Forecast (sₓ)

Variance formula:
[
s_f^2 = SEE^2 \left(1 + \frac{1}{n} + \frac{(X_f - \bar{X})^2}{\sum (X_i - \bar{X})^2}\right)
]

Where:

 ( SEE ) = standard error of estimate

 ( X_f ) = forecasted X value
 ( \bar{X} ) = mean of X
 ( n ) = number of observations
⚠️CFA Level I exams rarely ask you to calculate this — the standard error of the forecast (sₓ)
is usually given.

5. Example — Prediction Interval

Given:

 Regression: ( \hat{Y} = -2.3 + 0.64X )

 Forecast ( X = 10% )
 ( s_f = 3.67 )
 ( n = 36 ) → ( df = 34 )
 ( t_{0.025,34} = 2.03 )

Then:
[
\hat{Y} = -2.3 + 0.64(10) = 4.1%
]
[
\text{Prediction interval: } 4.1 \pm (2.03)(3.67)
]
[
4.1 \pm 7.5 \Rightarrow [-3.4%, 11.6%]
]

✅ Interpretation: We are 95% confident that ABC’s excess return will lie between –3.4% and
+11.6% when S&P 500 excess return = 10%.

6. Functional Forms of Regression

Linear regression assumes the relationship between X and Y is linear.

If it’s not, we can transform variables (using logarithms) to make it linear.

7. Common Functional Forms

Model Type Equation Interpretation of ( b_1 )

Linear–Linear ( Y = b_0 + b_1 X + \ Absolute change in Y for one-unit change in
(Lin–Lin) varepsilon ) X
Log–Linear (Log– ( \ln(Y) = b_0 + b_1 X + \ Relative (%) change in Y for an absolute
Lin) varepsilon ) change in X
Linear–Log (Lin– ( Y = b_0 + b_1 \ln(X) + \ Absolute change in Y for a relative (%)
 Model Type Equation Interpretation of ( b_1 )
Log) varepsilon ) change in X
Log–Log (Double ( \ln(Y) = b_0 + b_1 \ln(X) Relative (%) change in Y for a relative (%)
Log) + \varepsilon ) change in X (elasticity)

8. Example of Nonlinear Relationship

Suppose EPS of a company grows 10% annually (nonlinear trend).

Plotting EPS vs. Time gives a curved pattern → violates linearity assumption.

✅ Solution → Apply log transformation:

 Take natural log of EPS → makes the relationship linear.

 Regression: ( \ln(EPS) = b_0 + b_1(\text{Time}) )

This gives a log–lin model, where ( b_1 ) ≈ growth rate.

9. Visual Example (Conceptually)

 Original EPS vs Time: Curved upward (exponential growth).

 Log(EPS) vs Time: Straight line (linear relationship).
→ Now OLS regression can be applied correctly.

🧠 Module Quiz 10.3

Correct
Question Explanation
Answer
90% CI → ( \hat{Y} = 40 ), SE = 2.5, ( z = 1.65 ): ( 40 ± 1.65(2.5) =
1 C
35.9 \text{ to } 44.1 )
Lin–Log model: absolute change in Y for a relative (% log) change in
2 B
X

🧾 Key Concepts Summary

Concept Formula / Description

Regression Equation ( Y = b_0 + b_1 X + \varepsilon )
Least Squares Criterion Minimizes ( SSE = \sum (Y - \hat{Y})^2 )
Coefficient of Determination ( R^2 = SSR/SST = 1 - SSE/SST )
Standard Error of Estimate ( SEE = \sqrt{MSE} = \sqrt{SSE/(n-2)} )
 Concept Formula / Description
Predicted Value ( \hat{Y} = b_0 + b_1 X_f )
Confidence Interval ( \hat{Y} \pm t_{\alpha/2,n-2}(s_f) )
Log–Lin Model Relative ΔY / Absolute ΔX
Lin–Log Model Absolute ΔY / Relative ΔX
Log–Log Model Relative ΔY / Relative ΔX

✅ End of Module Ⅲ (10.3) Summary

You now have all three modules (Ⅰ, Ⅱ, Ⅲ) of CFA Level I – Regression Analysis fully
explained and organized — nothing missed.

Would you like me to now make a 1-page condensed revision sheet (cheat sheet) with only
formulas, key points, and exam triggers from all 3 modules?