IEOR 151 FALL 2018 - MIDTERM PRACTICE PROBLEMS

1. Match the applicants to the residency programs, and show intermediate steps of the algorithm.
For this problem, suppose the applicants preferences are given by:

Brady Louis Mary

City City City
General General General
Temple Temple Temple

Suppose that each residency program has only 1 open position, and that the programs pref-
erences are given by:

General City Temple

Mary Mary Mary
Louis Brady Louis
Brady Louis Brady

2. Suppose you are managing a bakery and would like to determine the number of croissants
that should be baked in the morning. Assume the selling price is $1.95(r = 1.95), the per
unit production cost is $0.75(cv = 0.75), and the holding cost is $0.25(q = 0.25).

(a) Assume the demand is distributed as a Gaussian random variable with mean 150 and
variance 5000. Use the newsvendor model to choose how many croissants should be
produced?
(b) Suppose you have measured the demand of croissants for the past 10 days, and the values
of demand sorted into ascending order are 76, 93, 119, 125, 128, 132, 163, 187, 217, 285.
Use the nonparametric newsvendor model to choose how many croissants the bakery
should produce. Explain your reasoning. Hint: Consider the sample distribution:
n
1X
F̂ (z) = 1(z ≤ Xi )
n
i=1

(c) Suppose that you also measure outside temperature Ti and rainfall Ri , which you believe
are correlated with demand for croissants. Using these measurements, you model crois-
sant demand as a linear function of outside temperature and rainfall: Xi = aTi + bRi + c,
where a, b, c are unknown constants. Furthermore, you decide to use statistical reg-
ularization to improve the results by adding the constraint |a| + |b| ≤ µ, where is a
known constant. The resulting newsvendor model with a linear demand model and
L1-constraint regularization is written as
n
1X
min (cf + cv .δi − p.(δi − Xi )− + q.(δi − Xi )+ )
n
i=1

s.t.δ1 = aTi + bRi + c

1
 |a| + |b| ≤ µ
Write this formulation as a linear program. Hint: Recall that the nonparametric
newsvendor
n
1X
min (cf + cv .δi − p.(δi − Xi )− + q.(δi − Xi )+ )
n
i=1

can be written as the following linear program

n
1X
min (cf + cv .δ − p.si + q.ti )
n
i=1

s.t.si ≥ −(δ − Xi ), ∀i
ti ≥ (δ − Xi ), ∀i
si , ti ≥ 0, ∀i
(d) Suppose you would like to perform the statistical regularization in another way. Let
λ > 0 be a known constant, and suppose you solve the following form of the newsvendor
with linear demand and L1-regularization
n
1X
min (cf + cv .δi − p.(δi − Xi )− + q.(δi − Xi )+ + λ(|a| + |b|))
n
i=1

s.t.δ1 = aTi + bRi + c

Write this formulation as a linear program.

3. Consider the following graph representation of a kidney exchange. Find the social welfare
maximizaing exchange under the constraint that all cycles can have length less than or equal
to L = 3.
3
v1 5 v2

2 5
5
4

2
v3 v4

4. Suppose an orange juice manufacturer would like to purchase oranges from a orange supplier.
√
e restaurants utility for the oranges is given by S(q) = 200 q. The fixed costs for the supplier
are 500. If the supplier is inefficient then its marginal costs are 5, and if the supplier is efficient
then its marginal costs are 4. Assume that the restaurant believes that there is a 83.33%(5/6)
chance that the food distributor is efficient.

(a) What are the first-best production levels?

(b) What are the contracts to implement the first-best production levels?

2
 (c) How much profit would the efficient distributor make if the restaurant offers a menu of
contracts corresponding to the first-best production levels of the inefficient and efficient
types?
(d) What are the second-best production levels?
(e) What is the menu of contracts for the second-best production levels?
(f) What is the information rent of the efficient distributor for the menu of contracts for the
second-best production levels?

5. Suppose you are the product manager of the Expedia, and you would like to determine
if the company should purchase a new electronic system to manage online hotel bookings.
Purchasing the system will cost $490, and similar website around the world have found that
using the electronic system leads to an average of 28 mistakes (e.g. lost sales opporunities)
per day. If the current error rate is 36 mistakes 1 per day, then purchasing the new system
will have a net savings of $360 over the span of one year. You have decided to use a minimax
hypothesis testing approach to answer this questions. As a first step, you record the number
of mistakes made over 10 days: 16, 49, 24, 24, 28, 30, 31, 38, 43, 23.

(a) Assume that the number of mistakes per day is approximated by a Gaussian random
variable with variance, σ 2 = 36. Using a binary search and z-table, compute the thresh-
old for this hypothesis test γ ∗ to within an accuracy of ±0.1 (4 points)
Hint: Use the following values for the minimax hypothesis test:n = 10, µ0 = 28, µ1 =
36, σ 2 = 36, L(µ0 , d0 ) = 0, L(µ0 , d1 ) = a = 490, L(µ1 , d0 ) = b = 360, L(µ1 , d1 ) = 0
(b) Should the manager purchase the new electronic system? Explain your answer