© CFA Institute. For candidate use only. Not for distribution.

344 Learning Module 5 Sampling and Estimation

PRACTICE PROBLEMS
1. Perkiomen Kinzua, a seasoned auditor, is auditing last year’s transactions for
Conemaugh Corporation. Unfortunately, Conemaugh had a very large number
of transactions last year, and Kinzua is under a time constraint to finish the audit.
He decides to audit only the small subset of the transaction population that is of
interest and to use sampling to create that subset.
The most appropriate sampling method for Kinzua to use is:
A. judgmental sampling.

B. systematic sampling.

C. convenience sampling.

2. Which one of the following statements is true about non-probability sampling?

A. There is significant risk that the sample is not representative of the
population.

B. Every member of the population has an equal chance of being selected for
the sample.

C. Using judgment guarantees that population subdivisions of interest are rep-

resented in the sample.

3. The best approach for creating a stratified random sample of a population in-
volves:
A. drawing an equal number of simple random samples from each
subpopulation.

B. selecting every kth member of the population until the desired sample size
is reached.

C. drawing simple random samples from each subpopulation in sizes propor-

tional to the relative size of each subpopulation.

4. Although he knows security returns are not independent, a colleague makes

the claim that because of the central limit theorem, if we diversify across a large
number of investments, the portfolio standard deviation will eventually approach
zero as n becomes large. Is he correct? false because it's dependent

5. Why is the central limit theorem important? because event pop dist. is nonnormal,sample dist. isnormal

6. What is wrong with the following statement of the central limit theorem?

Central Limit Theorem. “If the random variables X1, X2, X3, …, Xn are a ran-
dom sample of size n from any _ distribution with finite mean μ and variance
2
σ , then the distribution of ​ X ​ ​will be approximately normal, with a standard
_
deviation of σ​  / √
​  n ​. ” should be n>30 also sample mean will be X=u

7. Peter Biggs wants to know how growth managers performed last year. Biggs as-
sumes that the population cross-sectional standard deviation of growth manager
 © CFA Institute. For candidate use only. Not for distribution.
Practice Problems 345

returns is 6% and that the returns are independent across managers.

A. How large a random sample does Biggs need if he wants the standard devia-
tion of the sample means to be 1%? 1 = 6 / x^1/2 and x = 36

B. How large a random sample does Biggs need if he wants the standard devia-
tion of the sample means to be 0.25%? 0.25 = 6/ x^1/2and x = 576

8. A population has a non-normal distribution with mean µ and variance σ2. The
sampling distribution of the sample mean computed from samples of large size
from that population will have:
A. the same distribution as the population distribution.

B. its mean approximately equal to the population mean.

C. its variance approximately equal to the population variance.

9. A sample mean is computed from a population with a variance of 2.45. The sam-
ple size is 40. The standard error of the sample mean is closest to:
A. 0.039.

B. 0.247.

C. 0.387.

10. An estimator with an expected value equal to the parameter that it is intended to
estimate is described as:
A. efficient.

B. unbiased.

C. consistent.

11. If an estimator is consistent, an increase in sample size will increase the:

A. accuracy of estimates.

B. efficiency of the estimator.

C. unbiasedness of the estimator.

12. Petra Munzi wants to know how value managers performed last year. Munzi es-
timates that the population cross-sectional standard deviation of value manager
returns is 4% and assumes that the returns are independent across managers.
A. Munzi wants to build a 95% confidence interval for the population mean
return. How large a random sample does Munzi need if she wants the 95%
confidence interval to have a total width of 1%? (X + 1.96 * 4 / n^1/2) - (X - 1.96 * 4 / n^1/2) = 1 and n = 246

B. Munzi expects a cost of about $10 to collect each observation. If she has
a $1,000 budget, will she be able to construct the confidence interval she
wants? no because he need 246 * 10 = 2450 dollar for collecting samples

13. Find the reliability factors based on the t-distribution for the following confi-
dence intervals for the population mean (df = degrees of freedom, n = sample
size):
A. A 99% confidence interval, df = 20 2.845
 © CFA Institute. For candidate use only. Not for distribution.
346 Learning Module 5 Sampling and Estimation

B. A 90% confidence interval, df = 20 1.725

C. A 95% confidence interval, n = 25 2.064

D. A 95% confidence interval, n = 16 2.131

14. Assume that monthly returns are normally distributed with a mean of 1% and a
sample standard deviation of 4%. The population standard deviation is unknown.
Construct a 95% confidence interval for the sample mean of monthly returns if
the sample size is 24. (-0.69 :2.69)

15. Explain the differences between constructing a confidence interval when sam-
pling from a normal population with a known population variance and sampling
from a normal population with an unknown variance. z-statistic and t-statistic

16. For a two-sided confidence interval, an increase in the degree of confidence will
result in:
A. a wider confidence interval.

B. a narrower confidence interval.

C. no change in the width of the confidence interval.

17. For a sample size of 17, with a mean of 116.23 and a variance of 245.55, the width
of a 90% confidence interval using the appropriate t-distribution is closest to:
A. 13.23.

B. 13.27.

C. 13.68.

18. For a sample size of 65 with a mean of 31 taken from a normally distributed pop-
ulation with a variance of 529, a 99% confidence interval for the population mean
will have a lower limit closest to:
A. 23.64.

B. 25.41.

C. 30.09.

19. An increase in sample size is most likely to result in a:

A. wider confidence interval.

B. decrease in the standard error of the sample mean.

C. lower likelihood of sampling from more than one population.

20. Otema Chi has a spreadsheet with 108 monthly returns for shares in Marunou
Corporation. He writes a software program that uses bootstrap resampling to
create 200 resamples of this Marunou data by sampling with replacement. Each
resample has 108 data points. Chi’s program calculates the mean of each of the
200 resamples, and then it calculates that the mean of these 200 resample means
is 0.0261. The program subtracts 0.0261 from each of the 200 resample means,
squares each of these 200 differences, and adds the squared differences together.
The result is 0.835. The program then calculates an estimate of the standard error
 © CFA Institute. For candidate use only. Not for distribution.
Practice Problems 347

of the sample mean.

The estimated standard error of the sample mean is closest to:
A. 0.0115

B. 0.0648

C. 0.0883

21. Compared with bootstrap resampling, jackknife resampling:

A. is done with replacement.

B. usually requires that the number of repetitions is equal to the sample size.

C. produces dissimilar results for every run because resamples are randomly
drawn.

22. Suppose we take a random sample of 30 companies in an industry with 200

companies. We calculate the sample mean of the ratio of cash flow to total debt
for the prior year. We find that this ratio is 23%. Subsequently, we learn that the
population cash flow to total debt ratio (taking account of all 200 companies) is
26%. What is the explanation for the discrepancy between the sample mean of
23% and the population mean of 26%?
A. Sampling error.

B. Bias.

C. A lack of consistency.

23. Alcorn Mutual Funds is placing large advertisements in several financial publi-
cations. The advertisements prominently display the returns of 5 of Alcorn’s 30 only five of them show impressive result
funds for the past 1-, 3-, 5-, and 10-year periods. The results are indeed impres- so we could not say same for rest of its
funds
sive, with all of the funds beating the major market indexes and a few beating
them by a large margin. Is the Alcorn family of funds superior to its competitors?

24. Julius Spence has tested several predictive models in order to identify un-
dervalued stocks. Spence used about 30 company-specific variables and 10
market-related variables to predict returns for about 5,000 North American use out-of-sample data and look does it
and European stocks. He found that a final model using eight variables applied make economicsense
to telecommunications and computer stocks yields spectacular results. Spence
wants you to use the model to select investments. Should you? What steps would
you take to evaluate the model?

25. A report on long-term stock returns focused exclusively on all currently publicly
traded firms in an industry is most likely susceptible to:
A. look-ahead bias.

B. survivorship bias.

C. intergenerational data mining.

26. Which sampling bias is most likely investigated with an out-of-sample test?
A. Look-ahead bias

B. Data-mining bias
 © CFA Institute. For candidate use only. Not for distribution.
348 Learning Module 5 Sampling and Estimation

C. Sample selection bias

27. Which of the following characteristics of an investment study most likely indi-
cates time-period bias?
A. The study is based on a short time-series.

B. Information not available on the test date is used.

C. A structural change occurred prior to the start of the study’s time series.
 © CFA Institute. For candidate use only. Not for distribution.
Solutions 349

SOLUTIONS
1. A is correct. With judgmental sampling, Kinzua will use his knowledge and
professional judgment as a seasoned auditor to select transactions of interest
from the population. This approach will allow Kinzua to create a sample that is
representative of the population and that will provide sufficient audit coverage.
Judgmental sampling is useful in cases that have a time constraint or in which the
specialty of researchers is critical to select a more representative sample than by
using other probability or non-probability sampling methods. Judgement sam-
pling, however, entails the risk that Kinzua is biased in his selections, leading to
skewed results that are not representative of the whole population.

2. A is correct. Because non-probability sampling is dependent on factors other

than probability considerations, such as a sampler’s judgment or the convenience
to access data, there is a significant risk that non-probability sampling might
generate a non-representative sample

3. C is correct. Stratified random sampling involves dividing a population into

subpopulations based on one or more classification criteria. Then, simple random
samples are drawn from each subpopulation in sizes proportional to the relative
size of each subpopulation. These samples are then pooled to form a stratified
random sample.

4. No. First the conclusion on the limit of zero is wrong; second, the support cited
for drawing the conclusion (i.e., the central limit theorem) is not relevant in this
context.

5. In many instances, the distribution that describes the underlying population is

not normal or the distribution is not known. The central limit theorem states that
if the sample size is large, regardless of the shape of the underlying population,
the distribution of the sample mean is approximately normal. Therefore, even in
these instances, we can still construct confidence intervals (and conduct tests of
inference) as long as the sample size is large (generally n ≥ 30).

6. The statement makes the following mistakes:

_
■ Given the conditions in the statement, the distribution of ​ X ​ ​will be approxi-
mately normal only for large sample sizes.
■ The statement omits _ the important element of the central limit theorem that
the distribution of ​ X ​ ​will have mean μ.

7.
_
A. The standard deviation or standard error of the sample mean is ​σ​ ​ X ​ ​  =  σ / ​
_ _ _   _
√ n ​.  Substituting in the values for σ
​ ​ ​ X ​​  and σ, we have 1% = 6​ % / √
​  n ​,  or ​√ n ​ = 6.
Squaring this value, we get a random sample of n = 36.
_ _
B. As in Part A, the standard deviation of sample mean is ​σ​ ​ X ​ ​   =  σ / √ ​  n ​​. 
_ _ _
Substituting in the values for ​σ​ ​ X ​
​  and σ, we have 0.25% = 6​ % / √​  n ​,  or ​√ n ​ = 24.
Squaring this value, we get a random sample of n = 576, which is substan-
tially larger than for Part A of this question.

8. B is correct. Given a population described by any probability distribution (nor-

mal or non-normal) with finite variance, the central limit theorem states that the
sampling distribution of the sample mean will be approximately normal, with the
 © CFA Institute. For candidate use only. Not for distribution.
350 Learning Module 5 Sampling and Estimation

mean approximately equal to the population mean, when the sample size is large.

9. B is correct. Taking the square root of the known population variance to deter-
mine the population standard deviation (σ) results in
_
​σ  =  ​√ 2.45  
​  =  1.565​
The formula for the standard error of the sample mean (σX), based on a known
sample size (n), is
σ
​​σ​ X​  =  _
​ ​√ _n   ​ ​ 
Therefore,
1.565
​​σ​ X​  =  _
​  _ ​ 
  =  0.247​
​  40  ​
√

10. B is correct. An unbiased estimator is one for which the expected value equals
the parameter it is intended to estimate.

11. A is correct. A consistent estimator is one for which the probability of estimates
close to the value of the population parameter increases as sample size increases.
More specifically, a consistent estimator’s sampling distribution becomes concen-
trated on the value of the parameter it is intended to estimate as the sample size
approaches infinity.

12.

A. Assume the sample size will be large and thus the

_ 95% confidence interval
_ _
for the population mean of manager returns is ​ X ​  ± 1.96 ​s​ ​ X ​
​,  where ​s​ ​ X ​
​   =  s / ​
_
√ n ​.  Munzi wants the distance between the upper limit and lower limit in the
confidence interval to be 1%, which is
_ _
​​ (​ ​ X ​  + 1.96 ​s​ ​ X ​
_
​)  ​ ​ − ​ (​ ​ X ​  − 1.96 ​s​ ​ X ​
_
​)  ​ ​  =  1%​
_ _
Simplifying this equation, we get 2​ ​ (​ ​1.96 ​s​ ​ X ​ ​)  ​ ​= 1%. Finally, we have 3​ .92 ​s​ ​ X ​​ ​ =
_
1%, which gives us the standard deviation of the sample mean, s​ ​ ​ X ​​  = 0.255%.
_ _
The distribution of sample means is s​ ​ ​ X ​ ​   =  s / √ ​  n ​.  Substituting in the values
_ _ _
for ​s​ ​ X ​
​​  and s, we have 0.255% = 4​ % / √
​  n ​,  or ​√ n ​ = 15.69. Squaring this value,
we get a random sample of n = 246.
B. With her budget, Munzi can pay for a sample of up to 100 observations,
which is far short of the 246 observations needed. Munzi can either proceed
with her current budget and settle for a wider confidence interval or she can
raise her budget (to around $2,460) to get the sample size for a 1% width in
her confidence interval.

13.

A. For a 99% confidence interval, the reliability factor we use is t0.005; for df =
20, this factor is 2.845.
B. For a 90% confidence interval, the reliability factor we use is t0.05; for df =
20, this factor is 1.725.
C. Degrees of freedom equals n − 1, or in this case 25 − 1 = 24. For a 95% con-
fidence interval, the reliability factor we use is t0.025; for df = 24, this factor
is 2.064.
D. Degrees of freedom equals 16 − 1 = 15. For a 95% confidence interval, the
reliability factor we use is t0.025; for df = 15, this factor is 2.131.
 © CFA Institute. For candidate use only. Not for distribution.
Solutions 351

14. Because this is a small sample from a normal population and we have only the
sample standard deviation, we use the following model to solve for the confi-
dence interval of the population mean:
_ s
​​ X ​  ± ​t​ α/2​ _
​ ​√ _n    ​ ​ 
where we find t0.025 (for a 95% confidence interval) for
_ df = n − 1 = 24 − 1 = 23;
this value is 2.069. Our solution is 1% ± 2.069(4%)/​√
​  24 ​ = 1% ± 2.069(0.8165) = 1%
± 1.69. The 95% confidence interval spans the range from −0.69% to +2.69%.

15. If the population variance is known, the confidence interval is

_ σ
​​ X ​  ± ​z​ α/2​ _
​ ​√ _n   ​ ​ 
The confidence_ interval for the population mean is centered at the sample
mean, X ​ ​  ​.  The population standard deviation is σ, and the sample size is n. The
population standard deviation divided by the square root of n is the standard
error of the estimate of the mean. The value of z depends on the desired degree
of confidence. For a 95% confidence interval, z0.025 = 1.96 and the confidence
interval estimate is
_ σ
​​ X ​  ± 1.96​_  ​√ _n   ​ ​ 
If the population variance is not known, we make two changes to the technique
used when the population variance is known. First, we must use the sample stan-
dard deviation instead of the population standard deviation. Second, we use the
t-distribution instead of the normal distribution. The critical t-value will depend
on degrees of freedom n − 1. If the sample size is large, we have the alternative of
using the z-distribution with the sample standard deviation.

16. A is correct. As the degree of confidence increases (e.g., from 95% to 99%), a
given confidence interval will become wider. A confidence interval is a range for
which one can assert with a given probability 1 – α, called the degree of confi-
dence, that it will contain the parameter it is intended to estimate.

17. B is correct. The confidence interval is calculated using the following equation:
_ s
​​ X ​  ± ​t​ α/2​ _
​ ​√ _n   ​  ​ 
_
Sample standard deviation (s) = √​  245.55  
​= 15.670.
For a sample size of 17, degrees of freedom equal 16, so t0.05 = 1.746.
The confidence interval is calculated as
15.67
​116.23 ± 1.746​ _
_ ​ 
  =  116.23 ± 6.6357​
​  17  ​
√

Therefore, the interval spans 109.5943 to 122.8656, meaning its width is equal to
approximately 13.271. (This interval can be alternatively calculated as 6.6357 × 2.)

18. A is correct. To solve, use the structure of Confidence interval = Point estimate ±
Reliability factor × Standard error, which, for a normally distributed population
with known variance, is represented by the following formula:
_ σ
​​ X ​  ± ​z​ α/2​ _
​ ​√ _n   ​ ​ 
For a 99% confidence
_
interval, use z0.005 = 2.58.
Also, σ = √
​  529 ​ = 23.
Therefore, the lower limit = ​31 − 2.58​_   23
_  ​   =  23.6398​.
​√ 65  ​
19. B is correct. All else being equal, as the sample size increases, the standard error of
the sample mean decreases and the width of the confidence interval also decreases.
 © CFA Institute. For candidate use only. Not for distribution.
352 Learning Module 5 Sampling and Estimation

20. B is correct.
The estimate of the standard error of the sample mean with bootstrap resampling
is calculated as follows:
_______________ _____________________ _
_2
√ √
1 B ˆ 1 200 ˆ 2
      
_
​s​ ​ X ​
​​
​   _
=  ​    
​B − 1 ∑ 
   ​ 
 b=1
  ​(  ​θ​​ b​ − ​θ​)  ​  ​   =  ​  _
​     
200 − 1  
​
 ∑ 
b=1
  
​ √
(​θ​​ b​ − 0.0261)​​   
​
  =  ​ _
  1
​199   ​  × 0.835 ​ 
​
_
​s​ ​ X ​​   =  0.0648

21. B is correct. For a sample of size n, jackknife resampling usually requires n repeti-
tions. In contrast, with bootstrap resampling, we are left to determine how many
repetitions are appropriate.

22. A is correct. The discrepancy arises from sampling error. Sampling error exists
whenever one fails to observe every element of the population, because a sample
statistic can vary from sample to sample. As stated in the reading, the sample
mean is an unbiased estimator, a consistent estimator, and an efficient estimator
of the population mean. Although the sample mean is an unbiased estimator of
the population mean—the expected value of the sample mean equals the popu-
lation mean—because of sampling error, we do not expect the sample mean to
exactly equal the population mean in any one sample we may take.

23. No, we cannot say that Alcorn Mutual Funds as a group is superior to compet-
itors. Alcorn Mutual Funds’ advertisement may easily mislead readers because
the advertisement does not show the performance of all its funds. In particular,
Alcorn Mutual Funds is engaging in sample selection bias by presenting the in-
vestment results from its best-performing funds only.

24. Spence may be guilty of data mining. He has used so many possible combina-
tions of variables on so many stocks, it is not surprising that he found some
instances in which a model worked. In fact, it would have been more surprising
if he had not found any. To decide whether to use his model, you should do two
things: First, ask that the model be tested on out-of-sample data—that is, data
that were not used in building the model. The model may not be successful with
out-of-sample data. Second, examine his model to make sure that the relation-
ships in the model make economic sense, have a story, and have a future.

25. B is correct. A report that uses a current list of stocks does not account for firms
that failed, merged, or otherwise disappeared from the public equity market in
previous years. As a consequence, the report is biased. This type of bias is known
as survivorship bias.

26. B is correct. An out-of-sample test is used to investigate the presence of

data-mining bias. Such a test uses a sample that does not overlap the time period
of the sample on which a variable, strategy, or model was developed.

27. A is correct. A short time series is likely to give period-specific results that may
not reflect a longer time period.