Level II CFA 2024 Slide Answers

QUANTITATIVE METHODS

CFA Level 1 Recap

EXAMPLES

SLIDE 13

Answer B1 = 102.5/80 = 1.28

Answer B0 = 2 – (1.28 x ‐1) = 3.28

Final Equation: Yhat = 3.28 + 1.28 Xi

SLIDE 14

The average of Y and Ŷi is the same.

̂i is the mean. The Y
The sum of the error terms = 0, this implies that that Y ̂i plotted in the
middle of the observed values such that the deviations from the mean sum to zero. The sum of
the squared error terms is minimized.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 1
 Level II CFA 2024 Slide Answers

SLIDE 20

Answer: T critical (at 2.5% and df=3) = 3.182

Confidence interval intercept = 3.28 + 3.182(1.92) = 9.3894 or 3.28 ‐ 3.182(1.92) = ‐2.82

Confidence interval slope = 1.28 + 3.182 (0.239) = 2.040 or 1.28 ‐ 3.182(0.239) = 0.5195

Hypothesis Testing. Set up

null b0 = 0 Alternative b0 ≠ 0

null b1 = 0 Alternative b1 ≠ 0

Test statistic is ‐ t test

Significance 5%, two tailed test, so 2.5% in each tail

Decision rule: Confidence interval, CHECK to see if the hypothesized value (0) falls within the
confidence interval. If yes then Do NOT reject the Null, if No then reject the null in favor of the
alternative hypothesis.

FOR intercept: ‐ 2.82 to 9.39, hypothesized value 0 falls within the interval, fail to reject the
null. Conclude intercept is not statistically significantly different from zero.

FOR slope coefficient: 0.52 to 2.04, hypothesized value 0 falls outside the interval, reject the null.
Conclude intercept is statistically significantly different from zero. i.e. slope is able to explain some
variation in the dependent variable.

SLIDE 21

Null Hypothesis b0 = 0 Alternative b0 ≠ 0

Null Hypothesis b1 = 0 Alternative b1 ≠ 0

T calc = (estimated value – hypothesized value) / std error

Tcalc b0 = (3.28 – 0) / 1.92 = 1.708

Tcalc b1 = (1.28 – 0) / 0.239 = 5.355

Decision rule. For b0, since t calc < t critical, we fail to reject the null… so b0 =0 i.e. not
statistically significantly different from 0

For b1, since t calc > t critical, we reject the null… so b0 ≠ 0 i.e. it is statistically significantly
different from 0

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 2
 Level II CFA 2024 Slide Answers

SLIDE 22

The beta is the slope coefficient b1

The beta of the market is =1

So the test is for:

Null b1 = 1 Alternative b1 ≠ 1

Use the t‐test

Calculate test statistic = (estimated – hypothesized)/ std error = (1.28 – 1)/0.239 = 1.17

Determine t‐critical = 2.5% in each tail, df =3 = +/‐ 3.182

Because tcalc < t critical, we fail to reject the null. We conclude that b1 is NOT statistically
different from 1 i.e. YES beta of Zeta is equal to the beta of the stock market.

Testing using confidence interval will give the interval as 1.28 +/‐ 3.182 (0.239) = 0.52 to 2.04.
Since hypothesized value (1) is within the confidence interval we once again fail to reject the
null.

SLIDE 26

Y(hat) = 9.68%, -9.52%, 16.08, 3.28, -9.52

ANOVA Table D of f Sum of Squares Mean Sum of Squares

Regression 1 525.312=RSS 525.312=MRSS

Residuals (errors) 3 54.688=SSE 18.229=MSSE

Total 4 580.000=TSS 145.000

SLIDE 27

Coefficient of Determination = RSS/TSS = 525.312/580 = 0.91

OR 1- SSE/TSS = 1 – 54.688/580 = 0.91

Correlation coefficient = sqroot(0.91) = 0.95

Standard Error of Estimate = SEE = sqroot (SSE/n-2) = sq root (MSE) = sq root(18.229) = 4.27

Unexplained variation = 1 – coefficient of determination = 1 – 0.91 = 9%

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 3
 Level II CFA 2024 Slide Answers

Learning Module 2

SLIDE 52

R2 = RSS /SST = 35,687.67/71,806.18 = 0.497

Correlation = Sq root (R2) = sq root (0.497) = 0.7049
Adjusted R2 = 1- {[(n-1)/(n-k-1)](1-R2)} = 1- {(99/96)x (1-0.497) = 0.48128

SLIDE 54

The R2 for Model 1 (CAPEX only) indicates that 87.99% of the variation of ROA is explained by
CAPEX. For Model 2 (CAPEX and ADV), the R2 increases to 88.05%. However, the adjusted R2 for
Model 2 declines to 0.8701 (87.01%) from 0.8749 for Model 1. The lower adjusted R2 is
consistent with the |t-statistic| for ADV’s coefficient < 1.0 (i.e., 0.3302) and the coefficient
not being different from zero at typical significance levels (P-value = 0.7429). To conclude,
adding the ADV variable does not improve the overall statistical performance and explanatory
power of the model.

SLIDE 58

The R2 increases or remains the same as we add variables to the model.

The adjusted R2 increases with the addition of some variables (Factors 3 and 4) but decreases
with the addition of other variables (Factors 2 and 5).
The AIC is minimized with the model using Factors 1, 2, 3, and 4.
The BIC is minimized with the model using Factor 1 only.
If we are developing a model for prediction purposes, then we would likely select the four-
factor model that AIC indicates, whereas if we are seeking the most parsimonious, best-fitting
model, we would choose the one-factor model.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 4
 Level II CFA 2024 Slide Answers

SLIDE 60

T calc for intercept = (0.157 – 0)/0.083 =1.89156

T calc for slope 1 = (‐0.38 – 0) / 0.08 = ‐4.75

T calc for slope 2 = (1.25 – 0)/1.15 = 1.0869

T calc for slope 3 = (0.05 – 0)/0.01 = 5

T critical @ 2.5% and df = n – k ‐ 1 = 100 – 3 – 1 = 96

T critical ~ 1.984

So the intercept and slope 2 are insignificant. Slope 1 and 3 are significant.

Confidence interval

Intercept 0.157 +/‐ 1.984 x 0.083 = ‐0.007672 to 0.3216

Hypothesis value 0 is within the range, DO NOT reject null, b0 is insignificant.

Slope 1 ‐0.38 +/‐ 1.984 x 0.08 = ‐0.53872 to – 0.22128

Hypothesis value 0 is not in the range, reject null, b1 is significant.

Slope 2 1.25 +/‐ 1.984 x 1.15 = ‐1.0316 to 3.5316

Hypothesis value 0 is within the range, DO NOT reject null, b2 is insignificant.

Slope 3 0.05 +/‐ 1.984 x 0.01 = 0.03016 to 0.06984

Hypothesis value 0 is not in the range, reject null, b1 is significant.

P value

P value for intercept is 0.221 > 0.025 So fail to reject null, i.e. insignificant

P value for slope 1 is 0 < 0.025 So reject null, i.e. significant.

P value for slope 2 is 0.26 > 0.025 So fail to reject null, i.e. insignificant

P value for slope 3 is 0 < 0.025 So reject null, i.e. significant.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 5
 Level II CFA 2024 Slide Answers

SLIDE 66

H0: bFactor4 = bFactor5 = 0 vs. Ha: At least one bj ≠ 0

F = (Sum of squares error restricted model − Sum of squares error unrestricted) / q / Sum of
squares error unrestricted model / (n − k − 1) , with q = 2 and n – k – 1 = 44 degrees of
freedom.
α = 1% (one-tail, right side).
Critical F-value = 5.120. Reject the null if the calculated F-statistic exceeds 5.120.
F = (64.5176 − 56.6182) / 2 / 56.6182 / 44 = 3.9497 / 1.2868 = 3.0694
Fail to reject the null hypothesis because the calculated F-statistic does not exceed the critical
F-value. There is not sufficient evidence to indicate that at least one slope coefficient among
b4 and b5 is different from zero
This joint hypothesis test indicates Factors 4 and 5 do not provide sufficient explanatory power
(i.e., SSE declines by just 7.8994 = 64.5176 – 56.6182) to compensate for the loss of two
degrees of freedom by their inclusion in the unrestricted model. Thus, we conclude the
restricted, more parsimonious model fits the data better than the unrestricted model.

SLIDE 69

Fcalc = 11895.89/376.23 = 31.62

Fcritical =2.7

Fcalc > F critical

Reject the Null, the alternative is correct. This means that…ATLEAST ONE slope coefficient is
significantly different from zero.

SLIDE 71

Y = 0.157 – 0.38 X1 + 1.25 X2 + 0.05 X3

Y = 0.157 – 0.38(10) + 1.25(15) + 0.05(185)

Y=24.357

SLIDE 75

Y(hat) = 4.7022 + 1.2302 (1.5) – 0.0371(0.8) + 0.1029 (0.3) = 6.54869

Confidence Interval = 6.54869 + 1.96 x 1.146 = 8.79485
Confidence Interval = 6.54869 − 1.96 x 1.146 = 4.30253

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 6
 Level II CFA 2024 Slide Answers

Learning Module 3
SLIDE 88

T calculated = (coefficient value – hypothesis value) / white std error. = (0.22 – 0) / 0.19 = 1.15
T critical = df =22, significance = 5%/2 = 0.025 = 2.074

SLIDE 93

Question 1: Answer C
Statement A is incorrect. Since the p value of the F statistic (0.000) is less than alpha (5%) we can
reject the null (all slopes are zero) in favour of the alternative hypothesis (at least one slope is not
zero). Statement B is incorrect. R 2 is not a measure of prediction accuracy of regression models.
Prediction capability is an assessment of future actual Y vs. predicted Y. Instead, R 2 measures how
well changes in the past independent variables (X) can be used to explain changes in past Y variables.
R2 is a measure of goodness of fit among the data.
Statement C is correct. The statement correctly describes R 2

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 7
 Level II CFA 2024 Slide Answers

SLIDE 94

Question 1: Answer C

Question 2: Answer C
The calculated DW = 1.89 (provided in the question)
The four critical values of DW test are:
0 dl = 1.63 du = 1.72 4 – du = 4 – 1.72 = 2.28 4 – dl = 4 – 1.63 = 2.37

Since DW is between du and 4 – du, the test result is ‘do-not reject the null hypothesis’ i.e. we fail to
reject the null hypothesis. This implies that there is no serial correlation among the dataset.

Question 3: DW = 2(1-r)
DW = 1.89 = 2 x (1 – r)
R = 0.055

SLIDE 102

Answer D
This is an example of multicollinearity, which arises when one of the regressors is very highly correlated
with the other regressors. In this case, all three regressors are highly correlated with each other, so
multicollinearity exists between all three. Since the variables are not perfectly correlated with each other this
is a case of imperfect, rather than perfect, multicollinearity.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 8
 Level II CFA 2024 Slide Answers

EXAM STYLE QUESTIONS

Question 1
Answer B
Answer A is incorrect – both conditional and unconditional heteroskedasticity violate regression
assumptions. The only difference is, unconditional heteroskedasticity has less severe
complications.
Answer C is incorrect – conditional heteroskedasticity presents more problems in being able to
make accurate inferences.

Question 2
Answer A
Serial correlation causes regression coefficients to be biased. Heteroskedasticity does not affect
the coefficients, heteroskedasticity affects the standard errors. Which may cause biased
inferences during hypothesis testing.
Answer B is incorrect – Serial correlation is more serious violation because of its ability to bias
regression coefficients.
Answer C is incorrect – First half of the sentence is correct, but second half is incorrect.
Conditional heteroskedasticity also violates an assumption.

Question 3
Answer C
Answer A is incorrect – unadjusted R2 gives a more aggressive (higher value) than the adjusted
R2. The ADJUSTED R2 is a conservative measure.
Answer B is incorrect – unadjusted R2 will increase each time new variable is added to the
equation.

Question 4
Answer C is correct. This is serial correlation. The test for serial correlation is Breusch
Godfrey.
Answer A is incorrect: This is heteroskedasticity and the test is Breusch Pagan.
Answer B is incorrect: This is multicollinearity and the test is VIF or correlation matrix
OR contradiction between t and F test.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education Pa g e | 9
 Level II CFA 2024 Slide Answers

Learning Module 4
SLIDE 121

A and C are correct.

A is correct because the coefficient on REG is –0.5.
C is correct because the sum of coefficients is –0.3 = –0.5REG + (0.4MKTSH + –0.2REG_MKTSH).
B is not correct because the coefficient on MKTSH is positive and the coefficient on REG is negative.

SLIDE 123

Q1: answer C

Q2: Answer D

SLIDE 124

Q3: answer C

Q4. R = 0.522 + 0.046(MKT) + 0.7102(IND) + 0.9(FORT)

8.3624% = 0.522 + 0(.046 x 120) + (0.7102 x 2) + 0.9 x 1

SLIDE 133

Q1. The intercept of –0.3785 is the log odds of the probability of defaulting if DE, CASH, EBIT
are all zero.
The odds are 𝑒 −0.3785 = 0.68488797
The probability of default is then: P/(1 – P) = 0.68488797/(1+0.68488797) = 0.4065

Q2. A slope coefficient is interpreted as the expected change in the log odds for a one -unit
change in the variable.
The interpretation of DE coefficient is that a one-unit change in debt-to-equity ratio results in
a 43.37% increase in the probability of default holding all other variables constant.

Q3. The log-likelihood statistics from the logistic regression is:

Restricted: Intercept only –314.68
Unrestricted: Intercept, CASH, DE, NPM –301.20
Step 1: State the hypothesis Null: all slopes = 0 Alternative: atleast
one slope is not zero.
Step 2: Identify the appropriate test statistic Chi-squared test with 3 restrictions
Step 3: Specify the significance alpha = 5%
Step 4: State the decision rule. Critical chi-square-value = 7.8150, at
3 degrees of freedom and 𝑎𝑙𝑝ℎ𝑎 =5%
Reject the null hypothesis if the calculated LR is greater than 7.8150
Step 5: Compute the test statistic LR = −2(−314.68 − −301.20) = 26.960
Step 6: Make a decision Since LR calc 26.96 >chi critical 7.815,
we reject the null and we conclude that the unrestricted model fits the data better than the
restricted model.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education P a g e | 10
 Level II CFA 2024 Slide Answers

Learning Module 5

EXAMPLES
SLIDE 151

Answer C

R2 is a good sign… high R2 means the independent variables explain a significant variation in the
dependent variable.

Ignore DW = not used in AR models.

All the autocorrelations are statistically insignificant at the 5% level of confidence ‐ Good

The t statistic for intercept and slope coefficient is statistically significantly different from zero ‐
Good again

Mean reverting level = 0.025 / (1 – 0.014) = 2.53%

SLIDE 158

Answer Perform a simple Hypothesis test using the DF t values.

Null = g =0 Alternative g ≠ 0

Because DF t value = 1.89 and this is less than critical value of 2.35. We are unable to reject the
null.

Which means g = 0 is correct. Which means b1 – 1 = 0, this implies B1 = 1, a unit root. Because
this model has a unit root, the model is NOT covariance stationary.

Note: Even though the coefficient is 1.04, and not a perfect 1, we still conclude that B1 = 1, ie. It
is not statistically significantly different from zero.

Note: R2 is high… means the model is good. High number of observations, which means the
accuracy should be good.

DW is useless in this case… because this is an AR model. DW tests serial correlation in Multiple
regression only.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education P a g e | 11
 Level II CFA 2024 Slide Answers

Question 1
Answer C

Serial correlation, biased the standard errors downwards and the t‐statistics upwards.
Because the residuals are correlated… they are non‐random.

Question 2
Answer B

If the linear equation has residuals with serial correlation, then Paula will use log‐linear
model to reduce/remove the serial correlation.
Log‐linear model assumes dependent variable grows at a constant RATE
Linear model assumes dependent variable grows at a constant Amount

Question 3
Answer B

Answer A is incorrect – DW has nothing to do with AR models. DW is only useful for Multiple
regression models.
Answer C is ALMOST correct – Large t statistics implies that constant and lag are both
significant. So the model has some predictive capability. BUT the model suffers from serial
correlation – as observed from the autocorrelations of the errors. SO even if the model has
statistically significant coefficients… it is invalid due to error correlations.

Question 4
Answer C

Autocorrelation measures the correlation between the residuals… when there is no relation
between one observation to the next, the error (residual) term will also be random…
however… when a season approaches (seasonal sale). There is a gradual build‐up… towards
the sale… a consistent positive (or negative) pattern…this causes correlation between the
residual terms… and indicates ‘seasonality’.

Question 5
Answer C

Add a lagged variable to correct for seasonality. Exponential smoothing is used to deal with
noisy data ‐‐‐ data that is too erratic. It is used to suppress the noise. Should not be used to
suppress seasonal patterns.

Log transformation is good for data that grows at a constant growth rate.

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education P a g e | 11
 Level II CFA 2024 Slide
Answers

Question 6
Answer A
RMSE muse be calculated on out of sample data. The lower the errors in prediction (lower
RMSE), the better our model.

Learning Module 6
EXAMPLES
Question 1
Answer C

Contact us:
Tel: 04‐3688966 Website: www.edge‐education.com
Post: P.O. Box 182080 EDGE Education
Email: info@edge‐education.com
Dubai, U.A.E. Excellence Demands Great Education P a g e | 12