Exercises

Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. Descriptive Statistics
2. Probability and Expected Value
3. Covariance and Correlation
4. Normal Distribution
5. Sampling
6. Confidence Intervals
7. Hypothesis Testing
8. Regression Analysis

Exercises
Descriptive Statistics
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

# Occurrences

7
6
5
4
3
2
1
0

3
4
Values

Figure 1: Histogram for Problem 1

1. Find the median of the data in Figure 1.
2. Find the standard deviation of the data in Figure 1.
3. Five students from the 1999 MBA class took jobs in rocket science after graduation. Four
of these students reported their starting salaries: $95,000, $106,000, $106,000, $118,000.
The fth student did not report a starting salary. Choose one of the following:
(a) The median starting salary for all ve students could be anywhere between $95,000
and $118,000.
(b) The median starting salary for all ve students is $106,000.
(c) The median starting salary for all ve students is $106,500.
(d) The median starting salary for all ve students could be greater than $118,000.
4. The observations X1 , . . . , Xn have a mean of 52, a median of 52.1, and a standard deviation
of 7. Eight percent of the observation are greater than 66; 7.9% of the observations are
below 38. Based on this information, which of the following statements best describes the
data?
9

2
1.5
1
0.5
0
-2

-1

-0.5
-1
-1.5
-2

Figure 2: Scatter plot for question 5

(i) The distribution has positive skew.
(ii) The distribution has negative skew.
(iii) The distribution has high kurtosis.
(iv) The distribution conforms to a normal distribution.
5. Consider the data in the scatter plot of Figure 2. The correlation between the X and Y
values in the gure is closest to
(i) 0.2
(ii) 0.2
(iii) 1
(iv) 1
(v) 0
6. The observations X1 , . . . , Xn have a mean of 50 and a standard deviation of 7. Which of
the following statements is guaranteed to be true according to Chebyshevs rule? (Write
True or False next to each.)
(i) At least 75% of the observations are between 36 and 64
(ii) At least 80% of the observations are between 34 and 66
(iii) At least 88.9% of the observations are between 31 and 73
(iv) Fewer than 15% of the observations are below 30
7. Suppose the observations X1 , X2 , . . . , Xn have mean 10. Suppose that exactly 75% of the
observations are less than or equal to 15. According to Chebyshevs rule, what is the
smallest possible value of the population standard deviation of these observations?

10

0.25

Frequency

0.2

0.15

0.1

0.05

0
0

10

20

30

40

50

60

70

80

90

100

Price

Figure 3: Histogram of bond prices at default, 1974-1995. (Source: Moodys Investor Services.)
8. Which of the following best describes the data in Figure 3? (Base your answer on the
appearance of the histogram. You do not need to do any calculations. Select just one
statement below and complete the one you select.)
(a) The mean is greater than the median because
(b) The median is greater than the mean because
(c) The mean and median are roughly equal because
9. One proposal that has received little attention from Major League Baseball is to pay
pitchers according to the following rule: each pitcher receives a base salary of $4.25 million,
minus $0.25 million times his earned run average (ERA). (A lower ERA is associated with
better performance.) If this rule were adopted, what would be the correlation between a
pitchers earnings and ERA? (Assume that the ERA cannot exceed 17, so this rule never
results in negative earnings. You may also assume a standard deviation of 1.2 for ERA.)
10. Using the data in Figure 4, answer both (a) and (b) below, providing a numerical value
for each.
(a) The mean of the data in the histogram is
(b) The median of the data in the histogram is
11. Cluster had exams in Finance and Marketing last week. All 60 students in the cluster
took both exams. The results were as follows:
Finance: mean = 25, standard deviation = 2
Marketing: mean = 75, standard deviation = 12
Correlation between score in Finance and same students score in Marketing = 0.84
Mary, a student in Cluster , scored a 30 in Finance and a 90 in Marketing. We are
interested in comparing her performance on the two exams relative to the rest of the
class. In particular, we would like to make a statement about which of her scores ranked
higher compared to the other scores on the same exam. Select one of the choices below
and complete the statement you select.
11

20
18
16

# observations

14
12
10
8
6
4
2
0
0

10

value

Figure 4: Histogram for Problem 10

(i) Marys score in Finance probably ranks higher than her score in Marketing because
(ii) Marys score in Marketing probably ranks higher than her score in Finance because
(iii) Marys scores on the two exams probably rank about equally high because
(iv) We cannot make any comparison between the two scores because

12. Seven students from the 1998 MBA class took jobs in brain surgery after graduation. Five
of the students reported their starting salaries: $55,000, $90,250, $90,250, $95,500, and
$105,000. Choose one of the following:
(a) Based on the information given, the largest possible value of the median starting
salary for all seven students is
(b) Based on the information given, it is not possible to put an upper limit on the median
starting salary for all seven students.

12

Solutions: Descriptive Statistics

1. There are 15 data points in the histogram. Seven are smaller than 3 and seven are greater
than 3, so the median is 3.
2. We show four dierent ways of calculating the standard deviation.
Method 1. List the full set of observations in a spreadsheet, repeating values as
many times as they occur: 0, 0, 0, 0, 1, 2, 2, 3, 4, 4, 4, 5, 5, 6, 7. Apply the function
STDEVP to the observations. The result is 2.28. (The function STDEV gives a
slightly dierent answer because it calculates the sample standard deviation rather
than the population standard deviation.)
Method 2. List the full set of observations in a spreadsheet, as in the column
labeled Xi in Table 1. Calculate the average of these values; this gives 2.867.
Now make a column of the squared observations, Xi2 , as in the second column of the
table. The average of the squared observations is 13.4. The variance is the dierence
between the average of the squared observations and the square of the average:

n
1
X2
n i=1 i

n
1
Xi
n i=1

2

The standard deviation is the square root,

Mean

Xi
0
0
0
0
1
2
2
3
4
4
4
5
5
6
7
2.867

Xi2
0
0
0
0
1
4
4
9
16
16
16
25
25
36
49
13.4

= 13.4 (2.867)2 = 5.182.

5.182 = 2.28.

2
(Xi X)
8.218
8.218
8.218
8.218
3.484
0.751
0.751
0.018
1.284
1.284
1.284
4.551
4.551
9.818
17.084
5.182

Table 1: Solution to Problem 2

Method 3. As in Method 2, list the full set of observations and calculate their mean to
2 , the squared distance from
= 2.867. For each observation Xi , calculate (Xi X)
get X
13

the mean; these values are in the third column of Table 1. The average of these squared
dierences gives the variance
n
1
2 = 5.182.
(Xi X)
n i=1

The standard deviation is the square root,

5.182 = 2.28.

Method 4. List the distinct values observed without repetition, as in the second column
of Table 2. For each value, calculate what proportion of the observations had that value,
as in the rst column of the table. Calculate the weighted average of the values, using
the proportions as weights; the result is 2.867. Now calculate the weighted average of the
2
squared values
to get 13.4. The dierence 13.4 (2.867) = 5.182 is the variance. The
square root 5.182 = 2.28 is the standard deviation.
Proportion
4/15
1/15
2/15
1/15
3/15
2/15
1/15
1/15
Weighted Avg.

Value
0
1
2
3
4
5
6
7
2.867

Squared Value
0
1
4
9
16
25
36
49
13.4

Table 2: Solution to Problem 2, Method 4

3. The answer is (b). No matter what the missing salary gure is, 106,000 will be the median.
For example, if the missing salary were 50,000, the ordered list would be
50, 000,

95, 000,

106, 000,

106, 000,

118, 000

and if the missing salary were 150,000 it would be

95, 000,

106, 000,

106, 000,

118, 000,

150, 000.

Either way, the median would be 160,000.

4. The answer is (iii). The data described in the question has the following characteristics:
the median is slightly larger than the mean; the right tail has slightly more data than
the left tail; both tails contain more than three times as many observations as would be
predicted by the normal distribution. (Recall that approximately 95% of the values lie
within 2 of the mean in the normal case; hence, about 2.5% should be above and
2.5% below.) The most pronounced feature is therefore the last one (heavy tails) which
is suggestive of high kurtosis.
5. The answer is (iv), because the points lie almost exactly on a line with negative slope.
14

6. The answers are T, T, F, T. (i) should be clear because the range given is exactly 2
standard deviations. (ii) First we nd the m that goes with 80%:
.80 = 1

1
m = 2.236.
m2

Thus, Chebyshev tells us that at least 80% are between 34.35 and 65.65. Clearly, at least
as many observations must lie between 34 and 66. (iii) To guarantee 88.9% we need a
range of 3 standard deviations, which would be from 29 to 71. The range given only
goes as low as 31. (iv) First nd how many standard deviations below the mean 30 is:
(50 30)/7 = 2.86.
Chebyshev guarantees that at least
1

1
= 87.8%
(2.86)2

of the observations are within 2.86 standard deviations of the mean. But then at most
12.2% can be below 30.
7. Exactly 25% are greater than 15, so 15 can be at most 2 standard deviations above the
mean of 10; i.e., the standard deviation cannot be less than 2.5.
8. (a) because the distribution is skew right.
9. The proposed rule makes Salary = 0.25 ERA + 4.25, a linear transformation with
negative slope. The correlation is therefore 1.
10. (a) The mean is [(20 0) + (2 9)]/22 = 18/22 = 0.82. (b) The median is 0. (List the
observations from smallest to largest. Since we have an even number of observations, take
the two closest to the middle the 11th and the 12th and average them. Both are 0,
so the median is 0.)
11. The answer is (i) because her score in Marketing is 2.5 standard deviations above the
mean whereas her score in Finance is only 1.25 standard deviations above the mean.
12. Even if the two unknown salaries were very large, the median could not be larger than the
fourth smallest value, $95,500. For example, if the two unknown salaries were $500,000,
the values become
55,000

90,250,

90,250,

95,500,

with 95,500 in the middle.

15

105,000,

500,000,

500,000,

Exercises
Probability and Expected Value
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. An inverse oater is a type of security whose payments move in the opposite direction of
short-term interest rates. The security is ordinarily structured so that no matter how high
interest rates rise, the payment cannot be negative, and no matter how low interest rates
drop, the payment cannot exceed some specied cap, e.g., 7%. Consider the following
specic case: if the prevailing short-term rate is X, the inverse oater pays
Y = 100 max(0.07 X, 0)
on $100 of face value. (The notation max(0.07 X, 0) means use whichever is larger,
0.07 X or 0.) Suppose the distribution of the short-term rate X at the next payment
date is given by the following table:
x
P (X = x)

0.04
0.30

0.06
0.20

0.07
0.20

0.08
0.15

0.09
0.15

Find the expected payment E[Y ].

2. A mining company plans to develop two potential gaussite reserves. Each reserve has
probability 0.30 of successfully yielding usable gaussite, and the success of each reserve
is independent of the other. If either of the two reserves is successful, it will generate $4
million in prot; if both are successful, prots will be $7 million because excess supply
will lower prices. If neither is successful, prots will be 0. Let X be the companys prot.
Find E[X].

16

3. The Gourmet Cafe serves the exotic Bernoulli Salmon at lunch and dinner. The number of
customers ordering the salmon at lunch and dinner are given by the following distributions:
Lunch demand
probability

0
0.3

1
0.5

2
0.2

Dinner demand
probability

0
0.2

1
0.4

2
0.4

Assume the lunch and dinner demands are independent of each other so the joint distribution of the lunch and dinner demands is given by the following table:

Dinner

0
1
2

0
0.06
0.12
0.12
0.3

Lunch
1
0.10
0.20
0.20
0.5

2
0.04
0.08
0.08
0.2

0.2
0.4
0.4

(Each entry of the table is just the product of the marginal probabilities at the end of the
corresponding row and the bottom of the corresponding column.)
The chef orders the sh in advance at a cost of $3.50 per serving. Any sh left over at
the end of the day is discarded.
(a) What is the expected total demand for the sh in a day?
(b) Suppose the chef orders three servings. What is the breakeven selling price (i.e.,
the price at which the expected revenue from sales of the sh equals the cost of the sh
ordered)? Assume that a customer who would have ordered the sh but nds it sold out
simply leaves rather than order something else. (Hint: Expected revenue = price times
expected number of units sold.)
4. The Uris & Warren Ratings Agency rates bonds on a simplied scale with just three
categories: A, B, and C. The Professors Pension Fund has all its money invested in two
bonds, X and Y , both of which are currently rated B. Over the course of the next year
the ratings of the bonds may change; the end-of-year value (in millions) of each bond
depends on its end-of-year rating as in the following table:
Rating
A
B
C

X Value
100
75
50

Y Value
100
75
50

The joint distribution of the end-of-year ratings of the two bonds is given by the following
table:
17

A
B
C

A
0.20
0
0

Y
B
0.05
0.40
0.05

C
0
0.10
0.20

(a) Find the probability that bond X will be rated C at the end of one year.
(b) Find the expected value of the year-end value of bond X.
(c) The Pension Fund buys an insurance contract that will pay 20 million if either bond
is downgraded to C. If both bonds are downgraded the contract still pays 20; if neither
is downgraded the contract pays nothing. Find the expected value of the payo of the
insurance contract.

18

Solutions: Probability and Expected Value

1. Using the rule Y = 100 max(0.07 X, 0), list the payments in each scenario as follows:
x
P (X = x)
y

0.04
0.30
3

0.06
0.20
1

0.07
0.20
0

0.08
0.15
0

0.09
0.15
0

Now we have E[Y ] = (.30)(3) + (.20)(1) = 1.1.

2. Consider the possible outcomes and their probabilities:
outcome
only 1st successful
only 2nd successful
both successful
neither successful

probability
0.30(1-0.30)
(1-0.30)0.30
(0.30)(0.30)
(1-0.30)(1-0.30)

payo
4
4
7
0

Thus,
E[X] = (2 0.3 0.7 4) + (0.30 0.30 7) = 2.31
3. (a) From the joint distribution of lunch and dinner demands we nd that the distribution
of total demand is as follows:
Demand
Prob

0
0.06

1
0.22

2
0.36

3
0.28

4
0.08

Now we have that the expected value is

0(0.06) + 1(0.22) + 2(0.36) + 3(0.28) + 4(0.08) = 2.10.
We could also get this answer by adding the expected lunch demand (0.9) and the expected
dinner demand (1.2).
(b) You cant sell sh you dont have: if four people order it, you only sell three. So, the
distribution of units sold is this:
Demand
Units sold
Prob

0
0
0.06

1
1
0.22

2
2
0.36

3
3
0.28

4
3
0.08

The expected number sold is

0(0.06) + 1(0.22) + 2(0.36) + 3(0.28) + 3(0.08) = 2.02.
To break even, the price p needs to be such that 2.02p equals the cost of the sh, namely
3 3.50 = 10.50. The price is therefore p = 10.50/2.02 = 5.198 or 5.20.
19

4. (a) The question asks for the marginal probability that X will be rated C. By summing
the third row of the table, we nd that this is 0.25.
(b) To nd E[X] we need to nd the possible values and their probabilities:
Scenario
A
B
C

Value
100
75
50

Prob.
0.25
0.50
0.25

The probabilities are obtained by summing across the rows of the table; this is the marginal distribution of X. We now nd
E[X] = 100 0.25 + 75 0.50 + 50 0.25 = 75.
(c) The insurance contract pays if either bond is downgraded; thus, the scenarios in which
the contracts pays are the ones in boxes in the following table:

A
B
C

A
0.20
0
0

Y
B
0.05
0.40
0.05

C
0
0.10
0.20

The probability that the contract pays is

0 + 0.05 + 0.20 + 0.10 + 0 = 0.35.
So, the expected payo is 0.35 20 + (1 0.35) 0 = 7.

20

Exercises
Covariance and Correlation
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. You manage a retail operation from which you sell to both walk-in and telephone customers. For a particular product, your goal is to set the inventory so that 99% of customers
looking or calling for the product nd it in stock. Consider the following two scenarios:
(i) Days with a lot of walk-in customers are also days with a lot of telephone orders. (ii)
Days with a lot of walk-in customers tend to be days with fewer telephone orders and
vice-versa. In which scenario would you expect to have to hold more total inventory to
meet your service objective? Explain your answer by making reference to the concepts of
standard deviation and correlation.
2. You invest $3 thousand in one stock and your spouse invests $2 thousand in another.
Over the next year, each dollar invested in your pick will increase by X dollars and each
dollar invested in your spouses will increase by Y dollars; X and Y are random variables
with the following properties:
X has a mean of 0.09 and a standard deviation of 0.20.
Y has a mean of 0.12 and a standard deviation of 0.27.
The correlation between X and Y is 0.6.
Your individual earnings are 3X thousand, your spouses individual earnings are 2Y
thousand and your family earnings are the sum of two.
(a) What is the expected value of your family earnings?
(b) What is the standard deviation of your family portfolio earnings?
3. Let X, Y , and Z be random variables. Consider the following statements:
(i) if Cov[X, Y ] > Cov[Y, Z] then XY > Y Z .
(ii) if XY > Y Z then X > Y .
(iii) if XY > 0 then Cov[X, Y ] > 0.
Now pick one of the following:
(a) all of the above are true
(b) (i) and (ii) are true
(c) (i) and (iii) are true
(d) (ii) and (iii) are true
21

(e) only (iii) is true

4. Suppose you borrow $1000 for one year at a variable interest rate tied to the yield on
government bonds. As a result, the total interest you will pay (in dollars) is a random
variable X1 , having mean 60 and standard deviation 2. You invest the borrowed money.
Your earnings on the investment, X2 , have mean 85 and standard deviation 8. Suppose
the correlation between your earnings and the interest you pay on the loan is 0.3.
(a) Your net earnings at the end of the year are Y = X2 X1 . Find the expected value
of your net earnings.
(b) Find the standard deviation of your net earnings.
5. *** A company knows that it will buy 1 million gallons of jet fuel in 3 months, and wants
to hedge against a possible price increase. The company chooses to hedge by buying
futures contracts on heating oil. Suppose the standard deviation of changes in the price
per gallon of jet fuel over 3 months is 0.032, the standard deviation of changes in the
futures price per gallon of heating oil is 0.040, and the correlation between the two is 0.8.
Also, each heating oil futures contract is for 42,000 gallons.
(a) What is the standard deviation of the companys unhedged exposure? (Think of the
company as holding 1, 000, 000 gallons of jet fuel, because of its anticipated purchase.)
(b) A simple gallon-for-gallon hedging rule would suggest that the company should buy
1, 000, 000/42, 000 = 23.8 24 contracts. What is the standard deviation of the companys exposure under this strategy? (Hint: Let X = change in price of jet fuel and Y =
change in futures price (per gallon) of heating oil. Write the exposure in terms of X and
Y .)
(c) Find the number of contract that minimizes the standard deviation of the companys
exposure.

22

Solutions: Covariance and Correlation

1. To meet a 99% service objective, you would need to set the inventory at roughly 2 or
3 standard deviations above the mean. (Mean and standard deviation here refer to the
demand for the product.) The standard deviation of the total demand is given by
total =

2
2
walkin
+ phone
+ 2walkin phone ,

where walkin , phone are the standard deviations for the individual types of demands and
is their correlation. If the demand streams are positively correlated, the variability of
total demand will be greater, resulting in a higher inventory requirement. If the demands
are negatively correlated, total variability and required inventory will be lower.
2. (a) E[3X + 2Y ] = 3E[X] + 2E[Y ] = 3(.09) + 2(.12) = .51.
2 + 4 2 + 2(3)(2) = 1.04. The standard deviation is the
(b) V ar[3X + 2Y ] = 9X
X Y
Y
square root, 1.02.

3. The answer is (e). In more detail:

(i) False, because X could be larger than Z .
(ii) False, because we dont know anything about the covariances.
(iii) True, because standard deviations are positive so the correlation and covariance
always have the same sign.
4. (a) E[Y ] = E[X2 X1 ] = E[X2 ] E[X1 ] = 85 60 = 25.
2 + 2 2 = 58.4. So, =
(b) V ar[Y ] = V ar[X2 +(1)X1 ] = X
X1 X2
Y
X2
1

58.4 = 7.64.

5. (a) The standard deviation of the price change for one gallon is 0.032 so the standard
deviation for 1, 000, 000 gallons is 32,000.
(b) After buying 24 contracts, the company in eect has a position of 1, 000, 000X +
24 42, 000Y , using the notation in the hint. The resulting standard deviation is


2 + (24 42, 000)2 2 + 2(1, 000, 000)(24)(42, 000)

(1, 000, 000)2 X
X Y
Y

(1, 000, 000)2 (.032)2 + (24 42, 000)2 (.040)2 + 2(1, 000, 000)(24)(42, 000)(.032)(.040)(0.8)

= 24, 193.
This is lower than the unhedged standard deviation.
(c) A simple way to nd the mimimum is to repeat the calculation in (b) with 24 replaced
by several dierent values and then graph the results; see Figure 5. From this it is clear
that the answer is about 15.
For a more systematic approach, let b denote the number of contracts and notice that we
want to nd the value of b that minimizes
2
+ (b 42, 000)2 Y2 + 2(1, 000, 000)b(42, 000)X Y .
(1, 000, 000)2 X

23

0.04

Standard Deviation (in millions)

0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
0

10

15
20
Number of Contracts

25

30

35

Figure 5: Hedge Eectiveness

This is a quadratic function of b, hence the shape of the graph in Figure 5. Setting the
derivative with respect to b equal to zero, we get
2b (42, 000)2 Y2 + 2(1, 000, 000)(42, 000)X Y = 0,
so the optimal number is
b =

X
1, 000, 000
,

Y
42, 000

which works out to be 15.2, or about 15 contracts.

The factor 1,000,000/42,000 is just to get the units right (it converts gallons to contracts).
The more interesting factor is X /Y . This is the minimum variance hedge ratio and
applies whenever we want to hedge X using Y , whatever X and Y may be. If we plug b
back into the formula for the variance, we get
Variance at optimal hedge
= 1 2 .
Variance with no hedge
Thus, the correlation between the X and Y determines how eectively we can hedge X
using Y . If is close to 1, we can eliminate almost all the risk in one by hedging with
the other.

24

Exercises
Normal Distribution
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. Daily demand for widgets is normally distributed with a mean of 100 and a standard
deviation of 15.
(a) What is the probability that the demand in a day will exceed 125?
(b) What is the probability that demand will be less than 75? Less than 70?
(c) How many widgets should be stocked to ensure that with 95% probability all demands
will be met?
2. The plant manager of a manufacturing facility is concerned about drug use among plant
workers and plans to implement random drug testing. One of the tests to be applied
measures the level of factor-X in blood samples. Among recent users of cocaine, the level
of factor-X is normally distributed with a mean of 10.0 and a standard deviation of 1.3.
Among non-users, the level is normally distributed with a mean of 6.75 and a standard
deviation of 1.5.
The employer plans to send a warning letter to all employees with a factor-X level of x
or greater, with x to be determined.
(a) Find the value of x that will ensure that 90% of recent cocaine users will be sent a
warning letter.
(b) If your answer to part (a) is adopted, what proportion of non-users will also be sent
warning letters?
3. A bank nds that the one-day increase in the dollar value of its foreign exchange portfolio
is normally distributed with a mean of $1.5 million and a standard deviation of $9.7
million. (A negative increase is a loss.)
(a) Find the value x such that the probability that the portfolio will lose more than x
dollars in one day is 5%.
(b) For the x you found in part (a), what is the probability that the portfolio will increase
in value by more than x dollars in one day?
4. Mogul Magazine has recently completed an analysis of its customer base. It has determined that 75% of the issues sold each month are subscriptions and the other 25% are
sold at newsstands. It has also determined that the ages of its subscribers are normally
distributed with a mean of 44.5 and a standard deviation of 7.42 years, whereas the ages
of its newsstand customers are normally distributed with a mean of 36.1 and a standard
deviation of 8.20 years.
25

(a) Mogul would like to make the following statement to its advertisers: 80% of our
and
. Your job is to ll in the blanks,
subscribers are between the ages of
choosing a range that is symmetric around the mean. (In other words, the mean age of
subscribers should be the midpoint of the range.)
(b) What proportion of Moguls newsstand customers have ages in the range you gave in
(a)?
(c) What proportion of all of Moguls customers have ages in the range you gave in (a)?

26

Solutions: Normal Distribution

1. (a) Demand X N (100, (15)2 ). Standardizing, we get P (X > 125) = P (Z > [125
100]/15) = P (Z > 1.67) = 1 0.9525 = 0.0475. (b) By symmetry of the normal, the
probability below 75 is the same as the probability above 125 (both are 25 away from
the mean of 100) so the answer is again 0.0475. For 70, standardize to get P (X < 70) =
P (Z < [70 100]/15) = P (Z < 2) = P (Z > 2) = 1 P (Z < 2) = 1 0.9772 = 0.0228.
(c) The z-value for 0.95 is 1.645 so we need to set the inventory level 1.645 standard
deviations above the mean, at 100 + 1.645 15 = 124.67.
2. (a) Need to nd x so that P (X > x) = 0.90, where X N (10, 1.32 ). Standardizing, we
nd that this is the same as P (Z > (x 10)/1.3) = 0.90. From the table, we nd that
(x 10)/1.3 must be 1.28. Thus, x = 10 (1.28)(1.3) = 8.336.
(b) Now we need to nd P (Y > x), with x = 8.336 and Y N (6.75, 1.52 ); i.e., P (Z >
(8.336 6.75)/1.5), which is P (Z > 1.0573) = 1 P (Z < 1.0573) = 1 .8554 = .1446.
3. (a) Let X be change in portfolio value, X N (1.5, (9.7)2 ). Positive changes are gains,
negative changes are losses. Need to nd x so that P (X < x) = 0.05. Standardizing,
we nd that x = 1.5 1.645(9.7) = 14.456, so x = 14.456. There is only a 5% chance
that the portfolio will lose more than 14.456 million dollars.
(b) Now we need to nd P (X > 14.456). Standardizing, this becomes P (Z > 1.337) =
1 0.909 = .091.
4. (a) For the range from x to + x to contain the middle 80% of the subscribers, 10%
must be abve + x so 90% must be below + x. From the normal table, we nd that
the z-value for 90% is 1.28. Thus, x must be 1.28 standard deviations or 1.28 7.42 = 9.5.
The resulting range is from 35 to 54.
(b) Let Y be N (36.1, (8.2)2 ). We need to nd
P (35 < Y < 54) = P (Y < 54) P (Y < 35).
Standardizing, we nd that P (Y < 54) P (Y < 35) equals
P (Z <

35 36.1
54 36.1
)P (Z <
) = P (Z < 2.18)P (Z < 0.13) = 0.9850.447 = 0.538.
8.2
8.2

(c) Of the 75% in the subscriber group, 80% are in the range. Of the remaining 25%,
53.8% are in the range. Thus, the overall fraction is (0.75)(0.80) + (0.25)(.538) = 0.734.

27

Exercises
Sampling
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. Let X1 , . . . , X10 be a random sample from a population with mean 50 and standard
be the sample mean. Find the expected value and standard deviation
deviation 4. Let X

of X.
2. Each visit to Uris.com has a 2% chance of turning into a purchase. Let p denote the
proportion of visits that turn into purchases from a random sample of 100 visits. Find
the expected value and standard deviation of p.
3. (a) The starting salary among 1999 graduates of the Evian Business School has a standard
deviation of $17,000. If you randomly survey 40 students and average their starting
salaries, what is the probability that the average among these students will be greater
than the average among all students?
(b) What is the probability that the average in the sample will exceed the average among
all students by more than $5,000?
(c) What is the probability that the average in the sample will dier from the average
among all students by more than $5,000?
4. A traders annual bonus is normally distributed with a mean of $650,000 and standard
deviation $125,000. Her bonuses are independent from year to year. Find the probability
that her average bonus over a ve-year period will be less than $500,000.
5. Fifty-seven percent of students at Calabria Business School support making Introduction
to Philanthropy a required course. The school plans to survey 100 students to gauge
opinion on this issue. What is the probability that fewer than half of those surveyed will
say they support requiring the course?
6. A wireless communication company is considering switching from a per-minute charge to
a at monthly fee for unlimited service. It anticipates that this will result in greater usage
and would like to estimate what the average number of minutes per month per customer
will be under the new plan. To do this, it oers 1000 customers the at-fee plan for one
month and tracks their usage. From past data, the company knows that the standard
deviation of monthly minutes is about 120 minutes; the company expects that this gure
will not change much under the new plan. Find the probability that the average among
the 1000 test customers will dier from the true mean by more than 5 minutes.

28

7. A company believes that roughly 35% of consumers rate its brand rst in quality. It
would like to estimate the proportion more accurately and retains a market research rm
to survey n consumers. Using the initial estimate of 35% as a guide, determine how large
n must be to ensure that there is only a 5% chance that the estimate from the survey will
dier from the true value by more than 3 percentage points.

29

Solutions: Sampling

has expected value 50 and standard deviation /n = 4/ 10 = 1.26.

1. X
2. p has expected value 0.02 and standard deviation
0.014.

p(1 p)/n =

.02(1 .02)/100 =

denote the average for the sample and the average among all students. Then X

3. Let X
2
is (approximately) normally distributed with a mean of and a variance of /n, where
is given as $17,000 and n = 40.
> ) = 1/2 because the mean of a normal distribution is also its median.
(a) P (X

> 5000) = P (Z > 5000/(17000/ 40)) = P (Z > 1.86) = 10.9686 = 0.0314.

(b) P (X
(c) By symmetry, this is twice as large as the answer to (b) and thus 0.0628.
denote the ve-year average. Then X
is normal with mean 650000 and standard de4. Let X

< 500000) = P (Z < [500000650000]/55902) =

viation 125000/ 5 = 55902. Thus, P (X
P (Z < 2.68) = 0.0037.
5. Let p denote the proportion in favor out of the 100 surveyed. Then p is approximately

normally distributed with a mean of p = 57% and a standard deviation of p(1 p)/n,
n = 100. Thus,

P (
p < 0.50) = P

0.50 p
p p

<
p(1 p)/n
p(1 p)/n

.07
= P Z<
.57(1 .57)/100
= P (Z < 1.41) = 0.0793.

denote the sample mean among the 1000 test customers and let denote the true
6. Let X
N (, 2 /n), with given as 120 and n = 1000. We
mean under the new plan. Then X
need to nd
< 5 or X
> + 5).
P (X
By symmetry of the normal, this is the same as
> + 5).
2P (X
Standardizing, this becomes

2P

5
X
>
/ n
/ n

= 2P

Z>
120/ 1000

= 2P (Z > 1.32) = 2(10.9066) = 18.68%.

7. Let p denote the estimate obtained by the market research rm and let p denote the true
proportion. Even though we dont know p, p, or n, we do know that the error in the

30

estimate, p p is approximately normally distributed with mean of 0 and a standard

deviation of

p(1 p)
.
=
n
It follows that there is only a 5% chance that the error will be greater than 1.96. We
would therefore like to choose n so that 0.03 = 1.96; i.e.,


0.03 = 1.96
Solving for n, we get

p(1 p)
.
n

n = (p(1 p)) (1.96/0.03)2 .

Of course, we still dont know p, but heres where we use the rough prior estimate of 35%
as a guide. Substituting this value, we nd
n (0.35(1 0.35)) (1.96/0.03)2 = 971.

31

Exercises
Confidence Intervals
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. Let X1 , . . . , X10 be a random sample from a normal population with mean and standard
be the sample mean and suppose X
= 48. Find a 95% condence
deviation 4. Let X
interval for .
2. Each visit to Uris.com has a probability p of resulting in a purchase. Out of a random
sample of 500 visits, 15 result in purchases. Use this to nd a 95% condence interval for
p.
3. In a random sample of 40 students from the 1999 class at Evian Business School, the average staring salary is $110 thousand and the sample standard deviation is $16 thousand.
Find a 95% condence interval for the population mean.
4. A wireless communication company is considering switching from a per-minute charge to
a at monthly fee for unlimited service. It anticipates that this will result in greater usage
and would like to estimate the change in the average number of minutes per month per
customer resulting from the new plan. To do this, it oers 850 customers the at-fee plan
for one month and tracks their usage. It also tracks a control group of 700 customers under
the old plan. Among the 850 test customers, it nds an average usage of 227 minutes
and a sample standard deviation of 135 minutes; in the control group, the corresponding
values are 163 and 85. Find a 95% condence interval for the increase in mean usage.
5. For a t-based condence interval, what multiplier would you use with a sample size of
15 and a condence level of 95%? 90%? What condence level would you get with a
multiplier of 2.624?
6. Six months after elections in the Democratic Republic of Urisia, a newspaper reports that
two-thirds of French companies with major government contracts have enjoyed increases
in the value of their contracts since the election and accuses the government of proFrench policies. Table 3 shows the contract values (in millions of dollars) for the 12
French companies with a major presence in Urisia, before and after the election. Using
the t distribution, calculate a 95% condence interval to assess the dierence in average
contract size before and after the election. What does this condence interval suggest?
7. (a) A recent study compared HMOs that provide nancial incentives for physicians to
reduce referrals with HMOs that do not. The study found that in the rst group 12 out
of 52 physicians surveyed said they would refer a patient with chest pain upon waking

32

Average
Std Dev

Before
68.3
37.0
121.4
113.1
117.3
47.7
85.7
99.4
108.8
92.0
63.3
33.9
82.3
31.6

After
71.5
32.5
122.1
119.8
111.2
44.0
94.6
103.6
129.4
74.7
75.1
39.2
84.8
33.8

Dierence
3.2
-4.6
0.7
6.6
-6.1
-3.6
8.8
4.2
20.6
-17.2
11.8
5.3
2.5
9.7

Table 3: Data for Problem 6

to a cardiologist; in the second group, 18 out of 45 physicians surveyed said they would
make the referral. Find a 95% condence interval for the dierence.
(b) Repeat the calculation now assuming the proportions are 120 out of 520 and 180 out
of 450. Comment on the dierence between this interval and the one in part (a).

33

Solutions: Confidence Intervals

1.96/n; i.e., 48 1.96(4/ 10), which is 48 2.48.

1. X
2. The sample proportion is p = 15/500 = .03. Condence interval:


p 1.96

p(1 p)
= .03 1.96
n

(.03)(1 .03)
= .03 .015.
500

= 110, s = 16, and n = 40, gives 110 4.96.

1.96s/n, with X
3. X
= 227, sX = 135, nX = 850. Control: Y = 163, sY = 85, nY = 700.
4. Test customers: X
Condence interval


Y ) 1.96
(X

s2
s2X
+ Y = 164 11.04.
nX
nY

5. Look up tn1,/2 with n = 15 and = 5%; i.e., t14,.025 which is 2.145. For 90% its
1.761. A multiplier of 2.624 leaves 1% probability in each tail and thus results in a 98%
condence level.
6. This is a matched-pairs setting because we are comparing the same companies before and
after the election. The average of the dierences is 2.5 and their standard deviation is
9.7. The sample size is 12, so we use the t-multiplier for 11 degrees of freedom and 95%
condence (right-tail probability 0.025) which is 2.201. Putting the pieces together we
get a condence interval of
9.5
2.5 2.201 = 2.5 6.0.
12
Comment: This condence interval is very wide. The uncertainty (as measured by the
halfwidth of 6.0) is very large compared with the average increas of 2.5 million. Indeed,
we cannot even be condent that the mean changed after the election because the interval
crosses zero. (Note that for this interpretation of the condence interval we need to view
the 12 companies as a random sample from a larger hypothetical population.)
7. (a) No incentive: pX = 18/45 = 0.40, nX = 45. With incentive: pY = 12/52 = 0.23,
nY = 52. Condence interval:


pX pY 1.96

pX (1 pX ) pY (1 pY )
+
= 0.17 0.18.
nX
nY

(b) Now we have pX = 0.40, nX = 450, pY = 0.23, nY = 520. The same formula gives a
condence interval of 0.17 0.058. We can make two comments
comparing (a) and (b).
First, the condence interval in (b) is narrower by a factor of 10 because both sample
sizes in (b) are 10 times larger. Second, even though the dierence in the two proportions
is the same in both cases (0.17), this dierence is more convincing in (b) because of the
narrower interval. This is again a consequence of the larger sample size.

34

Exercises
Hypothesis Testing
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

1. Let X1 , . . . , X10 be a random sample from a normal population with mean and standard
be the sample mean and suppose X
= 48. Let denote the unknown
deviation 4. Let X
population mean. Test H0 : = 45 vs. H1 :
= 45 at the 5% level. Test H0 : 45 vs.
H1 : > 45 at the 5% level.
2. Each visit to Uris.com has a probability p of resulting in a purchase. Out of a random
sample of 500 visits, 15 result in purchases. Is this statistically signicant evidence that
p
= 2%?
3. In a random sample of 40 students from the 1999 class at Evian Business School, the average staring salary is $110 thousand and the sample standard deviation is $16 thousand.
How signicant is the evidence that the population mean is greater than $100 thousand?
Give a p-value.
4. A wireless communication company is considering switching from a per-minute charge to
a at monthly fee for unlimited service. It anticipates that this will result in greater usage
and would like to estimate the change in the average number of minutes per month per
customer resulting from the new plan. To do this, it oers 850 customers the at-fee plan
for one month and tracks their usage. It also tracks a control group of 700 customers under
the old plan. Among the 850 test customers, it nds an average usage of 227 minutes
and a sample standard deviation of 135 minutes; in the control group, the corresponding
values are 163 and 85. Is the increase in mean usage statistically signicant?
5. Six months after elections in the Democratic Republic of Urisia, a newspaper reports that
two-thirds of French companies with major government contracts have enjoyed increases
in the value of their contracts since the election and accuses the government of proFrench policies. Table 4 shows the contract values (in millions of dollars) for the 12
French companies with a major presence in Urisia, before and after the election. Using
the t distribution, test whether the dierence in average contract size before and after is
statistically signicant at the 5% level.
6. (a) A recent study compared HMOs that provide nancial incentives for physicians to
reduce referrals with HMOs that do not. The study found that in the rst group 12 out
of 52 physicians surveyed said they would refer a patient with chest pain upon waking
to a cardiologist; in the second group, 18 out of 45 physicians surveyed said they would
make the referral. Is the dierence statistically signicant?
(b) Repeat the test now assuming the proportions are 120 out of 520 and 180 out of 450.
Comment on the dierence between this and part (a).

35

Average
Std Dev

Before
68.3
37.0
121.4
113.1
117.3
47.7
85.7
99.4
108.8
92.0
63.3
33.9
82.3
31.6

After
71.5
32.5
122.1
119.8
111.2
44.0
94.6
103.6
129.4
74.7
75.1
39.2
84.8
33.8

Dierence
3.2
-4.6
0.7
6.6
-6.1
-3.6
8.8
4.2
20.6
-17.2
11.8
5.3
2.5
9.7

Table 4: Data for Problem 5

Solutions: Hypothesis Testing

1. Test statistic:
Z=

45
X

= 2.37.
4/ 10

The rejection criterion at 5% is Z > 1.96 or Z < 1.96, so we reject the null hypothesis.
In the second test, the rejection criterion is Z > 1.645 so we again reject the null.
2. Set up the test as H0 : p = .02 vs. H1 : p
= .02. Test statistic:
p .02
Z=

.02(1.02)
500

= 1.60,

with p = .03. Note that we use .02 (the value under the null) in calculating the standard
error. At 5% signicance level, we would reject if Z > 1.96 or Z < 1.96; clearly, the
evidence is therefore not signicant at 5%. To measure how signicant it is we nd a pvalue. This is the area outside the range (1.60, 1.60) under the standard normal, which
is 11%.
3. For H0 : 100 vs. H1 : > 100 the test statistic is
Z=

100
110 100
X

= 3.95.
=
s/ n
16/ 40

The p-value is the area to the right of 3.95 under the standard normal curve which is zero
to four decimal places. The evidence is therefore extremely signicant.

36

4. Set up the test as H0 : X Y 0 vs. H0 : X Y > 0, with X the new-plan customers

and Y the control group. The test statistics is
Y 0
X
.
Z=
s2X
s2Y
nX + nY
= 227, sX = 135, nX = 850, and Y = 163, sY = 85, nY = 700, this evaluates to
Using X
227 163

1352
850

852
700

= 11.35.

This is a very large Z value; the null can be rejected at almost signicance level; the
evidence is highly signicant.
5. This is a matched-pairs setting because we are comparing the same companies before and
after the election. We need to test H0 : X Y = 0 vs. H0 : X Y
= 0, with Y =
Before and X = After. The average of thedierences is 2.5 and their standard deviation
is 9.7. The test statistic is t = 2.5/(9.7/ 12) = 0.89. To reject at 5% signifcance we
would need t > t11,.025 or t < t11,.025 . But t11,.025 = 2.201 so we cannot reject the null.
The test statistic 0.89 tells us that 2.5 is not statistically dierent from 0 because it is
only 0.89 standard errors away from 0.
6. We need to test H0 : pX pY = 0 vs. H0 : pX pY
= 0 with X = no incentive and
Y = with incentive. (a) We are given pX = 18/45 = 0.40, nX = 45. With incentive:
pY = 12/52 = 0.23, nY = 52. Under the null hypothesis, the two proportions are equal
so we rst need to calculate the pooled estimate
p0 =

18 + 12
= 0.31.
45 + 52

The standard error under the null is therefore



p0 (1 p0 ) p0 (1 p0 )
+
= 0.094.
nX
nY

This results in a test statistic of

Z=

0.40 0.23
pX pY
=
= 1.80.
0.094
0.094

Because this fails to exceed 1.96, it is not signicant at 5% (though it would be signicant
at 10%).
(b) With the larger sample sizes p0 is unchanged but the standard error becomes


p0 (1 p0 ) p0 (1 p0 )
+
= 0.030.
450
520

The test statistic becomes Z = 0.17/0.030 = 5.71 which now looks extremely signicant.
Thus, the dierence of 17% looks signicant in (b) but not in (a) because (b) is based on
a much larger sample size.

37

Exercises
Regression Analysis
Fall 2001
B6014: Managerial Statistics

Professor Paul Glasserman

403 Uris Hall

The raw data for all of these exercises can be downloaded in the le Exercises Regression.xls.
It is not necessary to download this le to solve these exercises but you may nd it helpful.
The data for Exercises 1-4 is from Managerial Statistics, by S.C. Albright, W. Winston, and
C. Zappe, published by Duxbury Press.
1. A drugstore chain has collected data on of 50 of its stores. For each store it has values of
the following two variables:
PROMOTE: Store promotional expenditure as a percentage of expenditure
by leading competitor in the stores area
SALES: Store sales as a percentage of leading competitors
Figure 6 shows part of the output from a regression of SALES against PROMOTE.
ANOVA
df
Regression
Residual
Total

Intercept
Promote

1
48
49
Coefficients
25.126
0.762

SS
2172.88
2624.74

Std Err
11.883
0.121

Figure 6: Drugstore Regression

(a) Interpret the coecient for PROMOTE in words.
(b) Find the R2 and standard error of the estimate se .
(c) Find the t-statistic for PROMOTE. What can you conclude from this value?
(d) Find a 95% condence interval the coecient for PROMOTE.
2. Figure 7 shows results of a regression of the US minimum wage during 1950-1994 against
time. The time variable is measured in years since 1950 and therefore runs from 0 to 44.
The wage variable is in dollars per hour and increased from 0.75 to 4.25 over this period.
(a) Interpret the coecient for Time in words.
(b) Find a 95% condence interval for this coecient.
(c) Calculate the R2 . What does this say about the growth in the minimum wage over
time?
38

ANOVA
df
Regression
Residual
Total

Intercept
Time

1
43
44

SS
57.0874
3.9456
61.0330

MS
F
57.0874 622.1441
0.0918

Coefficients
0.219
0.087

Std Err
0.09
0.00

t Stat
P-value
2.46
0.018
24.94
0.000

Figure 7: Regression of minimum wage against time

3. A bank is facing a sex discrimination suit in which it is accused of paying men more than
women.1 Consider the following analysis of 208 employees. For each employee we have
the following information:
SALARY: annual salary in thousands of dollars
FEMALE: 1 for women, 0 for men
YrsPrior: years of experience in banking prior to hiring
YrsExper: years working at this bank
Figures 8 and 9 show results of regressions based on this data. The bank also has information about the positions held and education backgrounds of the employees, but that is
not used in these regressions.
Regression Statistics
Multiple R
0.347
R Square
0.120
Adjusted R Square
0.116
Standard Error
10.584
Observations
208

Intercept
Female

Coefficients
45.505
-8.296

Std Err
1.284
1.564

t Stat
35.453
-5.302

P-value Lower 95% Upper 95%

0.000
42.975
48.036
0.000
-11.380
-5.211

Figure 8: Regression of bank salaries

Regression Statistics
Multiple R
R Square
Adjusted R Square
Standard Error
Observations

Intercept
YrsPrior
Female
YrsExper

0.702
0.492
0.485
8.079
208

Coefficients
35.492
0.131
-8.080
0.988

Std Err
1.341
0.181
1.198
0.081

t Stat
26.466
0.726
-6.744
12.208

P-value Lower 95% Upper 95%

0.000
32.848
38.136
0.469
-0.225
0.488
0.000
-10.443
-5.718
0.000
0.828
1.148

Figure 9: Regression of bank salaries

1

Albright, Winston, and Zappe state that this is real data taken from an actual law suit.

39

(a) Interpret the coecient for Female in Figure 8. Interpret the p-value that goes with
it.
(b) Interpret the coecient for Female in Figure 9. Interpret the p-value that goes with
it.
(c) The coecient for Female has larger magnitude in Figure 8 than Figure 9. Based on
information in the regression outputs, can you suggest an explanation for this? What
other information in the outputs might temper your conclusion?
(d) You have been hired to help defend the bank. What argument could you oer to
defend the bank against the evidence in the gures? What additional regressions might
support your argument?
4. Figure 10 shows the output from a regression of the following variables
QUANTITY: number of cars sold in US, in thousands
PRICE: index of ination-adjusted car prices
INCOME: ination-adjusted measure of disposable income, in thousands
INTEREST: prime rate of interest
We have one value of each of these variables for each year running from 1970 to 1987.
QUANTITY is the dependent variable.
Regression Statistics
Multiple R
0.843
R Square
0.711
Adjusted R Square
0.649
Standard Error
Observations
18
ANOVA
df
Regression
Residual
Total

Intercept
Price
Income
Interest

SS
3 15980400
14 6494658
17 22475058

Coefficients
1332.062
-62.400
8.290
-116.554

Std Err
2854.15
18.54
2.54
57.62

MS

F
P-value
11.48
0.000

t Stat
P-value Lower 95% Upper 95%
0.47
0.648
-4789.5
7453.6
-3.36
0.005
-102.2
-22.6
3.27
0.006
2.8
13.7
-240.1
7.0

Figure 10: Regression of car sales

(a) Find se , the standard error of the estimate. Interpret the value you get.
(b) Find the p-value for INTEREST. (The Excel function TDIST will give the exact value;
altneratively, use the z or t table to get an approximate value.) Relate the value you get
to the condence interval for INTEREST.
(c) Use the regression model to forecast car sales in 2001, assuming a PRICE index
level of 285, and INCOME index of 3,110, and prime rate of 9.50% (corresponding to an
INTEREST value of 9.5, not 0.095).
5. Figure 11 shows the results of regressing daily returns on the Australian dollar
against daily returns on the British Pound, Japanese Yen, and Canadian dollars. The
40

results are based on 250 days from July 1, 1999 to June 30, 2000. For each currency, the
data consists of the daily percentage changes in the number of US dollars per unit of the
foreign currency. We may think of these as the percentage changes in the price (in US
dollars) of the foreign currency.
Regression Statistics
Multiple R
R Square
Adjusted R Square
Standard Error
Observations

0.425
0.181
0.171
0.006
250

ANOVA
df
Regression
Residual
Total

Intercept
BritPound
JapanYen
CanDollar

3
246
249

SS
0.0018
0.0081
0.0099

Coefficients
-0.0004
0.2235
0.0298
0.7293

Std Err
0.0004
0.0749
0.0473
0.1126

MS
0.0006
0.0000

F
P-value
18.1213
0.0000

t Stat
P-value Lower 95% Upper 95%
-0.98
0.329
-0.0011
0.0004
2.98
0.003
0.0759
0.3710
0.63
0.530
-0.0634
0.1229
6.48
0.000
0.5076
0.9511

Figure 11: Regression of currencies

(a) State in words the meaning of the coecients in the regression output.
(b) Suppose you hold 100 million US dollars worth of Australian currency and you want
to hedge the possible change in value of this position over the next day. Suppose you can
hedge by buying or short selling British Pounds, Japanese Yen, and Canadian dollars.
What does the regression suggest you should do, buy or sell? How much of each? [Think
about your answer to (a).]
(c) How much risk is left after you hedge as in (b)? Be precise about the meaning of
your answer, not just the numerical value. What percent is this of the unhedged risk?
(d) Which of the following statements is best supported by the regression output:
(i) Yen should be omitted from the hedge because its coecient is very small
(ii) Yen should be omitted from the hedge because its p-value is rather large
(iii) Yen should be omitted from the hedge both because its coecient is very small and
because its p-value is rather large
(iv) Yen should not be omitted from the hedge

41

Solutions: Regression Analysis

1. The full output is displayed in Figure 12.
Regression Statistics
Multiple R
0.673
R Square
0.453
Adjusted R Squ
0.442
Standard Error
7.395
Observations
50
ANOVA
df
Regression
Residual
Total

Intercept
Promote

1
48
49

SS
2172.88
2624.74
4797.62

Coefficients
25.126
0.762

Std Err
11.883
0.121

MS
2172.88
54.68

F
39.74

P-value
0.000

t Stat
P-value Lower 95% Upper 95%
2.115
0.040
1.235
49.018
6.304
0.000
0.519
1.005

Figure 12: Drugstore Regression

(a) An increase in promotional expenditures of 1% of competitors promotional expenditures brings an increase of 0.762% in sales as a percentage of competitors sales.


(b) R2 = SSR/(SSR+SSE) = 2172.88/4797.62 and se = M SE = SSE/(n k 1) =

54.68 = 7.395. See Figure 12.

= 6.304. Because this is a large t value, we can say that the evidence of a
(c) t = /s

relation between SALES and PROMOTE is statistically signicant.

(d) tnk1,.025 s = 0.762 (2.01)(0.121). See Figure 12.

2. (a) The coecient for Time measures the annual rate of increase in the minimum wage.
According to the regression output, this rate is estimated at $0.0867 per year.
(b) t43,.025 s ; i.e., 0.0867 (2.32)(.0035); i.e., (.0797 , .0937).

(c) The

R2

is
SSR
SSR + SSE
57.0874
=
57.0874 + 3.9456
= 93.54%

R2 =

This is a high R2 indicating a good linear t. This means that the minimum wage grew
roughly linearly (i.e., at a constant rate) during 1950-1994.
3. (a) According to Figure 8, the average salary among female employees is $8,296 less
than amonge male employees. The very small p-value indicates that this dierence is
statistically signicant.

42

(b) According to Figure 9, the average salary among female employees is $8,080 less than
amonge male employees, once dierences in prior years of experience and years with the
bank have been accounted for. The very small p-value indicates that the dierence between
men and women remains statistically signicant even once these other dierences have
been accounted for.
(c) The coecients for both YrsPrior and YrsExper are positive, indicating that employees
with more years of experience tend to have higher salaries. The fact that the dierence
between average salaries for men and women is smaller when the experience variables are
included suggests that women may have lower values of YrsPrior and YrsExper, on average, so that part of the dierence observed in Figure 8 is due to dierences in experience,
rather than dierences in sex.
Two considerations temper this argument. First, the condence intervals for the Female
coecients are very wide compared with the dierence across the two regressions; so, the
dierence many not be meaningful. Second, the p-value for YrsPrior suggests that this
variable does not have a statistically signicant eect on salary.
(d) To defend the bank, you could argue that these regressions do not take account of
other dierences between the men and women employed at the bank. For example, women
may have been hired with fewer years of education and/or hired into lower-level positions.
To support this argument, you would want to run a regression that includes variables for
education and job grade to determine whether the salary dierence remains signicant
even after these other factors are considered. (Of course, this would not address the
question of whether the initial hiring was discriminatory; rather, it would check whether
men and women in comparable positions earn equal pay.)
4. (a) Recall that
se =

MSE =

SSE
,
nk1

so

6494658.12
= 681.
14
Thus, the se is 681,000 cars, and this is roughly the standard error in any forecast we try
to make from the model.
se =

(b) First we need the t-stat, which is

t=

116.554
0
= 2.02.
=
s
57.62

For the p-value, we use TDIST(2.02,14,2) because we have 14 degrees of freedom and we
want a 2-sided p-value. The answer is 0.063. Notice that this just fails to be signicant
at the 5% level; similarly (and equivalently) the 95% condence interval given in the
regression output just straddles 0.
(c) The estimated regression equation is
QUANTITY = 1332.062 62.400PRICE +8.290INCOME 116.554INTEREST
43

so our forecast is
QUANTITY = 1332.062 62.400 285 + 8.290 3110 116.554 9.50
which is 8223, or 8,223,000 cars.
5. (a) Interpret the coecient on British Pound as follows: for each percentage point change
in the pound, the Australian dollar changes by 0.2235. The interpretation for the others
is analogous.
(b) The coecients are all positive, which suggests that all four currencies tend to move
in the same direction. To hedge a position in Australian dollars, we should therefore take
an opposite position in the others. So, we should sell short.
The interpretation in (a) suggests that we should sell 22.35 million US dollars worth of
British pounds, 2.98 million US dollars worth of Yen, 72.93 million US dollars worth of
Canadian dollars. If we do this and if, for example, the dollar/pound rate increases by 1%
then we lose .2235 million US dollars on the short pound position. The regression predicts
that when the US dollar/pound rate goes up by 1%, the US dollar/Australian dollar rate
goes up by 0.2235%. So, we earn 0.2235 million on the Australian dollar position, which
is oset by the gains on the pound. Of course, in practice these would not oset each
other exactly because the regression equation does not hold exactly.
(c) The residuals of this regression are the hedging errors. Recall that se estimates the
standard error of the residuals. So, se estimates the residual risk. More precisely, the
percentage change of the hedged position has a standard deviation of 0.006. On a position
of $100 million, 0.006% means $6,000. This is the estimated standard deviation of the
one-day change in the value of the position.
Recall that the R2 gives the ratio of the explained variance to the total variance, and 1R2
gives the ratio of unexplained variance to the total variance. In the current setting, the
unexplained variance is how much variance remains after we hedge, and the total variance
is the variance without the hedge; similarly,
unexplained variance
Var[hedged position]
=
= 1 R2 .
Var[original position]
total variance

The ratio of the standard deviations is therefore 1 R2 = 1 0.180 = 0.906. Hence,

we have managed to hedge only about 10% of the risk.
The spreadsheet Exercises Regression.xls contains an illustration of this. Using the hedge
ratio suggested in (b), it calculates the daily returns on the hedged position. From these
values we can calculate the standard deviation over the 250 days to conrm that it is
0.006. We can them compare this with the standard deviation of the unhedged position.
(d) The best answer is (ii). The magnitude of the coecient is not relevant to this decision;
we could make the coecient as large or small as we like by changing units (dollars to
pennies for example). The p-value suggests that the relation between the Yen and the
Australian dollar is not statistically signicant once the eect of the other currencies is
considered.

44