Business Statistics Final Exam

Spring 2018

This is a closed-book, closed-notes exam. You may use a calculator.

Please answer all problems in the space provided on the exam.
Read each question carefully and clearly present your answers.

Here are some useful formulas:

• E(aX + bY ) = aE(X) + bE(Y )

• V ar(aX + bY ) = a2 V ar(X) + b2 V ar(Y ) + 2ab × Cov(X, Y )

• The standard error for the difference in the averages between groups a and b is
defined as: s
s2a s2
s(X̄a −X̄b ) = + b
na nb
where s2a denotes the sample variance of group a and na the number of observations
in group a.

Good Luck!

Honor Code Pledge: “I pledge my honor that I have not violated the Honor Code
during this examination.”

Signed:

Name:

1
Problem 1: Who’s to blame? (10 points)
In manufacturing its iPhone, Apple buys a particular kind of microchip from 3 suppliers:
30% from Freescale, 20% from Texas Instruments and 50% from Samsung.

Apple has extensive histories on the reliability of the chips and knows that 3% of the
chips from Freescale are defective; 5% from Texas Instruments are defective and 4%
from Samsung are defective.

In testing a newly assembled iPhone, Apple found the microchip to be defective. What
provider is the likely culprit?

P (defective | Freescale) = 0.03

P (defective | Texas) = 0.05
P (defective | Samsung) = 0.04
(1)
P (Freescale) = 0.3
P (Texas) = 0.2
P (Samsung) = 0.5

P (Freescale and defective) = P (defective | Freescale) × P (Freescale) = 0.03 × 0.3 = 0.009

P (Texas and defective) = P (defective | Texas) × P (Texas) = 0.05 × 0.2 = 0.01
P (Samsung and defective) = P (defective | Samsung) × P (Samsung) = 0.04 × 0.5 = 0.02
(2)
0.009
P (Freescale | defective) = = 0.23
0.009 + 0.01 + 0.02
0.01
P (Texas | defective) = = 0.26 (3)
0.009 + 0.01 + 0.02
0.02
P (Samsung | defective) = = 0.51
0.009 + 0.01 + 0.02
Given the chip is defective, it’s most likely to be Samsung.

Page 2
Problem 2: Breaking Bad... (10 points each)

Two chemists working for a chicken fast food company, have been
producing a very popular sauce. Let’s call then Jesse and Mr. White. Gus, their
boss, is tired of Mr. White’s negative attitude and is thinking about “firing” him and
keeping only Jesse on payroll. The problem, however, is that Mr. White seems to
produce a higher quality sauce whenever he is in charge of production if compared to
Jesse. Before making a final decision, Gus collected some data measuring the quality
of different batches of sauce produced by Mr. White and Jesse. The results, measured
on a quality scale, are listed below:

average std. deviation sample size

Mr. White 97 1 7
Jesse 94 3 10

Two questions:

1. Based in this data, can we tell for sure which one is the better chemist?

Two ways to solve the question:

(a) Confidence interval

• Mr. White: sX̄ = √sn = √17 = 0.378, 95% confidence interval is 97 ± 2 ×
0.378 = [96.244, 97.756].
• Jesse: sX̄ = √sn = √3
10
= 0.949, 95% confidence interval is 94±2×0.949 =
[92.102, 95.898].
Two confidence intervals do not overlap, the difference is significant and Mr.
White is better.

Page 3
 (b) Hypothesis testing
Null hypothesis : mean of two companies are the same. H0 : µ1 = µ2 , which
is equivalent to H0 : µ1 − µ2 = 0.
Difference of the mean is 97 − 94 = 3. Standard deviation of the difference
of the mean is s r
s21 s22 1 32
+ = + = 1.02.
n1 n2 7 10
3
t statistic is 1.02 = 2.94 > 2. We reject the null hypothesis at 95% level. Mr.
White is better.

2. Gus wants to keep the mean quality score for the sauce above 90. In this case, can
he can rid of Mr. White, i.e., is Jesse good enough to run the sauce production?
Yes, 95% confidence interval of mean quality score for Jesse is [92.102, 95.898],
which is above 90.

Page 4
Problem 3: Portfolios (5 points each)
We’re considering building a portfolio from three investments: a fund tracking the
SP500, a bond fund, and a fund of large cap stocks. The portfolios under consideration
are:

• Portfolio A: 50% SP500, 50% bonds

• Portfolio B: 50% SP500, 50% large-cap

Returns on the large cap fund and the bond fund have the same expected value and
standard deviation. Historically, there is a small negative correlation between the bond
and SP500 funds, and a small positive correlation between the large cap and SP500
funds. The returns on each investment have normal distributions.
Using only the information given above, choose the single correct response to each
question below:
(a) (4 points) What is the relationship between the expected returns for each portfo-
lio?
Portfolio A has higher expected returns
Portfolio B has higher expected returns
Both portfolios have the same expected returns correct
Impossible to say without more information

(b) (4 points) If we want the portfolio with the largest Sharpe ratio, which portfolio
should we choose?
Portfolio A correct
Portfolio B
Either one; their Sharpe ratios are the same
Impossible to say without more information

(c) (4 points) If we want the portfolio with the most potential for growth (say, the
portfolio that is most likely to generate returns greater than its average plus 2%),
which portfolio should we choose?
Portfolio A
Portfolio B correct
Either one; they are equally likely to generate returns greater than their
average plus 2%
Impossible to say without more information

Page 5
Problem 4 (2 points each)

Assume the model: Y = 5 + 2X1 + 3X2 + ε, ε ∼ N (0, 81)

1. What is E[Y |X1 = 1, X2 = 0] ?

(a) 5
(b) 9
(c) 7 correct
(d) 8

2. What is the V ar[Y |X1 = 0, X2 = 4]?

(a) 9
(b) 81 correct
(c) 3
(d) 6

3. What is the P r(Y > 5), given X1 = 0.5 and X2 = 3 ?

(a) 15%
(b) 68%
(c) 98%
(d) 87% correct

4. What is the P r(28 < Y < 35), given X1 = 4 and X2 = 4 ?

(a) 5%
(b) 23% correct
(c) 2.5%
(d) 34%

Page 6
Problem 5 (5 points each)

“ProShares UltraShort S&P500 (SDS) seeks daily investment results, before

fees and expenses, that correspond to two times the inverse (−2×) of the
daily performance of the S&P 500”

The above quote is from ProShares’ website, the manager of SDS.

In trying to validate their claim and make sure that SDS is a good fund that appro-
priately tracks its target, I decided to collect data on monthly returns (in percentage
terms) of SDS and the S&P500 Index since 2009 and run the following regression:

SUMMARY  OUTPUT SDS = β0 + β1 SP 500 +   ∼ N (0, σ 2 )

Regression  Statistics
Multiple  R 0.994
R  Square 0.989
Adjusted  R  Square 0.988
Standard  Error 0.760
Observations 62.000

ANOVA
df SS MS F Significance  F
Regression 1.000 3024.488 3024.488 5242.184 0.000
Residual 60.000 34.617 0.577
Total 61.000 3059.106

Coefficients Standard  Error t  Stat P-­‐value Lower  95% Upper  95%

Intercept -­‐0.437 0.103 -­‐4.252 0.000 -­‐0.642 -­‐0.231
SP500 -­‐1.867 0.026 -­‐72.403 0.000 -­‐1.918 -­‐1.815

Answer the following questions:

1. In trying to evaluate the claim made by ProShares, test the appropriate hypothe-
ses about β0 . What is your conclusion?

t statistic of intercept (β0 ) is -4.252, which is greater than 2 in absolute value. We

reject the null hypothesis that β0 = 0 at 95% level.

Page 7
2. In trying to evaluate the claim made by ProShares, test the appropriate hypothe-
ses about β1 . What is your conclusion?

We test the null hypothesis β1 = −2 which is equivalent to β1 + 2 = 0. the t

statistic is
−1.867 + 2
= 5.12 > 2 (4)
0.026
Reject the null hypothesis at 95% level. Proshares claims that SDS has −2 times
daily performance of S&P 500, which is rejected.

3. What is your final evaluation? Is SDS a good ETF? Justify your answer (and
don’t forget to address the estimate of σ 2 ).

It’s not a good ETF. Standard error is 0.760, which means that the ETF cannot
keep track of S&P 500 index well.

Page 8
 Problem 6: Crime data from our homework
(5 points each)
Let’s recall the “Crime” vs. “Police” example from our homework. There, we were
trying to understand the effect of more police on crime and we couldn’t just get data
from a few different cities and run the regression of “Crime” on “Police”. The problem
here is that data on police and crime cannot tell the difference between more police
leading to crime or more crime leading to more police... in fact I would expect to see a
potential positive correlation between police and crime if looking across different cities
as mayors probably react to increases in crime by hiring more cops. Again, it would be
nice to run an experiment and randomly place cops in the streets of a city in different
days and see what happens to crime. Obviously we can’t do that!
The researchers from UPENN mentioned in the homework were able to estimate this
effect by using what we call a natural experiment. They were able to collect data on
crime in DC and also relate that to days in which there was a higher alert for potential
terrorist attacks. Why is this a natural experiment? Well, by law the DC mayor has
to put more cops in the streets during the days in which there is a high alert. That
decision has nothing to do with crime so it works essentially as a experiment.
Here’s is the main table displaying the results from the analysis:

effect of police on crime 271

TABLE 2
Total Daily Crime Decreases on High-Alert Days

(1) (2)
High Alert !7.316* !6.046*
(2.877) (2.537)
Log(midday ridership) 17.341**
(5.309)
R2 .14 .17

Note.—The dependent variable is the daily total number of crimes

(aggregated over type of crime and district where the crime was
Figure 1: Thein Washington,
committed) dependentD.C., variable
during theisperiod
the March
daily12,total
2002–number of crimes in D.C. This
July 30, 2003. Both regressions contain day-of-the-week fixed effects.
table The
present
numberthe estimatedis coefficients
of observations and their
506. Robust standard standard
errors are in errors in parenthesis. The
parentheses.
first column refers different
* Significantly to a model
from zerowhere
at the the onlylevel.
5 percent variable used in the High Alert dummy
whereas**theSignificantly
model different
in columnfrom zero
(2)atcontrols
the 1 percent
formlevel.the METRO ridership. * refers to a

significant coefficient at the 5% level, ** at the 1% level.

local officials. In addition to increasing its physical presence, the police
department increases its virtual street presence by activating a closed-circuit
camera system that covers sensitive areas of the National Mall. The camera
system is not permanent; it is activated only during heightened terror alert
periods or during major events such as presidential inaugurations.10

IV. Results
Page 9
The results from our most basic regression are presented in Table 2, where
we regress daily D.C. crime totals against the terror alert level (1 p high,
Answer the following questions:

1. Why it was not enough to present the results from column (1) in the table? Why
did they have to include the METRO ridership variable?

There are confounding effects that may induce omitted variable bias. For example,
METRO ridership may be correlated with crime, because on the day with high
alert, people get out less, so there are less targets for crimes and thus daily crime
goes down. This effect is not caused by increasing police.

2. Can you explain why the estimates of the impact of police on crime from the
columns are different?

The direction of change is just as we expected. That is, some part of the effect in
column (1) can be explained by the fact that high alert will cause people to go out
less, so the coefficient for High Alert in column (2) is smaller in absolute value,
and the coefficient for Log(midday ridership) has significant effect on crime.

Page 10
Problem 7: House Prices (2 points each)
Let’s go back to the Midcity housing prices dataset from our homework... For simplicity
I have combined the two cheap neighborhoods into one group so we are left with only
two neighborhoods.

Let’s start by looking at the following model:

Model 1: P rices = β0 + β1 Size + β2 N BH + β3 BRICK × N BH + 

where N BH is a dummy variable that takes the value 1 if the house is in neighborhood
2 and BRICK is a dummy variable that equals 1 if the house is made out of brick. The
figure below displays the results from the regression. This is a graphical representation
of of the estimates of all coefficients in this regression.

Nbhd = 1
200

Nbhd = 2
Nbhd = 2 and Brick = 1
180
160
Price

140
120
100
80

1.6 1.8 2.0 2.2 2.4 2.6

Size

Based on the figure, answer the following questions:

1. What is the estimated value for the effect of Size on P rices for houses in neigh-
borhood 1?

(a) 65.32
(b) 30.45
(c) 17.98
(d) 49.85 - correct

Page 11
2. What is the estimated value for the effect of Size on P rices for houses in neigh-
borhood 2?

(a) 65.32
(b) 49.85 - correct
(c) 20.31
(d) 12.67

3. In neighborhood 2, what is the estimated premium for brick houses compared to

other houses?

(a) 15.76
(b) 38.61
(c) 26.08 - correct
(d) 52.10

4. What is the estimated average difference between a 1,800 sqft wood house in
neighborhood 2 and neighborhood 1?

(a) 25.09 - correct

(b) 39.78
(c) 48.90
(d) 13.94

Page 12
Problem 8: House Prices again! (2 points each)
Continuing in analyzing the MidCity data (same as the previous question), I now de-
cided to investigate whether or not the effect of Size on P rices changes in the different
neighborhoods. To this end, I worked with the following model:

Model 2: P rices = β0 + β1 Size + β2 N BH + β3 BRICK × N BH + β4 Size × N BH + 

The results are summarized in the figure below:

Nbhd = 1
200

Nbhd = 2
Nbhd = 2 and Brick = 1
180
160
Price

140
120
100
80

1.6 1.8 2.0 2.2 2.4 2.6

Size

Based on the figures, answer the following questions:

1. In model 2, what is the estimated value for the effect of Size on P rices for houses
in neighborhood 1?

(a) 71.30
(b) 30.45
(c) 17.98
(d) 51.27 - correct

Page 13
2. In model 2, what is the estimated value for the effect of Size on P rices for houses
in neighborhood 2?

(a) 75.23
(b) 46.67 - correct
(c) 20.31
(d) 51.27

3. In model 2, what is the estimate for β4 ?

(a) 46.67
(b) 51.27
(c) 13.15
(d) -4.60 - correct

4. What is the t-stat for the difference between the slope for Size in the two neigh-
borhoods?

(a) 2.15
(b) -4.44
(c) -0.35 - correct
(d) 5.63

Page 14
Problem 9: Medal Count
(3 points each)
Using data from Beijing 2008 and London 2012 I run a regression trying to understand
the impact of GDP (gross domestic product measured in billions of US$) and Popu-
lation (in millions of people) on the total number of medals won by a country in
SUMMARY OUTPUT
the summer Olympics. The results are
Regression Statistics
Multiple R 0.82488
R Square 0.68043
Adjusted R Square0.67660
Standard Error 10.83097
Observations 170.00000

ANOVA
df SS MS F Significance F
Regression 2.00000 41712.86080 20856.43040 177.78909 0.00000
Residual 167.00000 19590.76273 117.30996
Total 169.00000 61303.62353

Coefficients Standard Error t Stat P-value Lower 95% Upper 95%

Intercept 4.77423 0.90407 5.28082 0.00000 2.98935 6.55911
Population 0.01267 0.00467 2.71239 0.00738 0.00345 0.02189
GDP 0.00778 0.00050 15.67150 0.00000 0.00680 0.00876

(a) Is the intercept interpretable in this regression? Why?

No. The intercept corresponds to the case where population and GDP are both
zero, while we can’t have a country with zero population, in which case it’s not a
country.

Page 15
(b) Provide an interpretation for the coefficients associated with Population and
GDP?

Coefficient of population: Given GDP fixed, a one-million increase in population

is associated with 0.01267 more medal.
Coefficient of GDP: Given population fixed, a one-billion increase in GDP is
associated with 0.00778 more medal.

(c) What is the t-stat for Population telling you? Clearly explain the hypothesis
being tested and your conclusion.

The null hypothesis is H0 : β1 = 0, where β1 is the true coefficient of population.

The t-statistic is
0.01267
t= = 2.71 > 1.96,
0.00467
so we reject the null hypothesis and conclude that population has a significant
impact on the total medal.

(d) From the results, give a 95% prediction interval for the total number of medals for
the U.S. in the Rio 2016 Olympics, given that the U.S. current GDP is of 18.5
trillion of dollars and population is 300 million?

expected medal = 4.77423+0.01267*300+0.00778*18500=152.50

standard error = 10.83
95% prediction interval = [152.50−2×10.83, 152.50+2×10.83] = [130.84, 174.16]

Page 16
The following table shows the total medal count for a few countries in Rio 2016 Olympics
along with their current GDP and Population:

Country Total Medals GDP (in US$ billions) Population (in millions)
U.S. 121 18,500 300
Great Britain 67 2,800 64
China 70 11,300 1,357
Brazil 19 1,600 200
India 2 1,877 1,250
Holland 19 853 16.8
Fiji 1 3.8 0.881

(e) Using the results from the regression, which of these countries performance in the
Rio 2016 is not surprising? Why?
2 standard error = 21.66

Country Total Medals Predicted Medals Within 2 s.e.?

U.S. 121 152.50 NO
Great Britain 67 27.37 NO
China 70 109.88 NO
Brazil 19 19.76 YES
India 2 35.21 NO
Holland 19 11.62 YES
Fiji 1 4.82 YES

So Brazil, Holland, and Fiji are not surprising.

(f ) Based on the regression results, rank the performance of these countries in the Rio
Olympics. Explain your ranking methodology.

Ranking Country Total Medals - Predicted Medals

1 Great Britain 39.63
2 Holland 7.38
3 Brazil -0.76
4 Fiji -3.82
5 U.S. -31.51
6 India -33.21
7 China -39.88

Rank according to (Total Medals-Predicted Medals).

Page 17
I proceeded to add a dummy variable for the host country into the regression... I also
ran a regression with only GDP and Host. The results are below:
Regression Statistics
Multiple R 0.8639
R Square 0.7462
Adjusted R Square0.7417
Standard Error 9.6805
Observations 170.0000

ANOVA
df SS MS F Significance F
Regression 3.0000 45747.2827 15249.0942 162.7214 0.0000
Residual 166.0000 15556.3409 93.7129
Total 169.0000 61303.6235

Coefficients Standard Error t Stat P-value Lower 95% Upper 95%

Intercept 4.8246 0.8081 5.9705 0.0000 3.2292 6.4200
Population 0.0034 0.0044 0.7763 0.4387 -0.0053 0.0121
GDP 0.0077 0.0004 17.4626 0.0000 0.0069 0.0086
Host 48.3225 7.3648 6.5613 0.0000 33.7819 62.8632

Regression Statistics
Multiple R 0.86332
R Square 0.74532
Adjusted R Square0.74227
Standard Error 9.66902
Observations 170.00000

ANOVA
df SS MS F Significance F
Regression 2.00000 45690.80714 22845.40357 244.36222 0.00000
Residual 167.00000 15612.81639 93.48992
Total 169.00000 61303.62353

Coefficients Standard Error t Stat P-value Lower 95% Upper 95%

Intercept 4.92223 0.79728 6.17374 0.00000 3.34817 6.49628
GDP 0.00789 0.00041 19.46333 0.00000 0.00709 0.00869
Host 50.15148 6.96945 7.19590 0.00000 36.39190 63.91107

Page 18
(h) Of the 3 models presented, which one is the best in your opinion? Carefully explain
why?

The model with GDP and host dummy included is the best one. One can see
from the standard error of the models.

(i) In the last model presented, provide an interpretation for the coefficient associated
with Host.

Given GDP fixed, being host country is associated with 50.15148 more medals than
not being host country.

(j) Using your chosen model, evaluate Brazil’s performance in the Rio Olympics. Com-
pare and explain the difference in the results if you were to talk about Brazil’s
performance based on the first regression.

Using the third model,

predicted medal = 4.9222 + 0.0079× 1600 + 50.1515 = 67.71.

So using this model, Brazil has bad performance in the Rio Olympics, given it
being the host country.

Page 19