Week 4: Decisions in Settings with High Uncertainty

u Session 1 Decision Trees

m Example: Furniture maker IDEA Chooses a Supplier
u Session 2 Using Simulation within Decision Trees
m Example: More Complex Demand Distributions for IDEA
u Session 3 Using Optimization Together with Simulation
m Example: IDEA Chooses Order Quantities
u Session 4 Wrap Up
m Example: Back to the Newsvendor Problem

OPS Analytics MOOC, Week 4 Session 4

Remember Senthils newsvendor problem from week 1

u You are selling wodgets
m Unit cost = 3 talers
m Sales price = 12 talers
m Salvage value = 0 talers
u Historically, demand has been variable, uncertain
100

90
80
70

60
50
40
30

20
10
0
0

10

20

30

40

50

60

70

80

90

100

u How many wodgets, Q, should you order to maximize expected profit?

OPS Analytics MOOC, Week 4 Session 4

We know how to find a good order quantity Q

u Use the historical data to forecast future demand
m Normally distributed with mean of 52.81 and standard deviation of 15.10
u Use the demand forecast to drive a simulation
m If we order Q and demand is D, then profit p = 12 * min{D , Q} 3 * Q
m For a given Q simulate samples of D and calculate a p for each sample
m Calculate the average of the ps
u Use optimization for find an average-profit-maximizing Q for the sample
m Objective to maximize the average profit
m Decision variable is Q
m Constraints on minimum and maximum order quantity
u The optimal Q maximizes average profit for the sample
m Its a good estimate of the optimal Q for the demand forecast
OPS Analytics MOOC, Week 4 Session 4

Q* 60 maximizes expected profit for this sample

A

1 Newsvendor Optimization.xlsx
2
3
Sales Price =
12
Decision
4
Unit Cost =
3
variable5
Order Quantity =
59.5
6
7
8
Simulated Sales
9
Sample
Value
Revenue
10
1
35.4
424.5
11
2
15.3
183.7
12
3
48.6
582.8
13
4
43.4
521.2
14
5
60.9
714.4
15
6
41.9
503.1
16
7
51.5
618.3
17
8
62.9
714.4
58
49
65.6
714.4
59
50
52.0
624.3
60
61
62
OPS Analytics MOOC, Week 4 Session 4

=$C$3*MIN(B10,$C$5)

=$C$4*$C$5

Unit
Cost
178.6
178.6
178.6
178.6
178.6
178.6
178.6
178.6
178.6
178.6

Total
Profit
245.9
5.1
404.2
342.6
535.8
324.5
439.7
535.8
535.8
445.7

Average =
Standard Deviation =
4

401.7
157.9

=C10-D10

Maximization
Objective

=AVERAGE(E10:E1009)
=STDEV(E10:E1009)

We can also include constraints on risk

u Recall Sergeis session on risk and reward
m Often decision makers must trade off risk and reward
m A common measure of reward is expected valueits risk neutral
m One common measure of risk (of many) is the standard deviation
u Suppose we want to limit the newsvendors risk
m E.g., standard deviation of the profit should be no more than 125 talers
u We can update our optimization problem to limit that risk
m Use Excel to calculate the standard deviation of the profit
m Add a constraint that limits the standard deviation for any Q
u Note: the standard deviation function is also not linear
m But it is quadratic which is a well-behaved kind of nonlinear function
m For example, see J. R. Evans, Business Analytics, Pearson, 2013.
OPS Analytics MOOC, Week 4 Session 4

For standard deviation 125 on Q* 50

A

1 Newsvendor Optimization.xlsx
2
3
Sales Price =
12
4
Unit Cost =
3
5
Order Quantity =
50.3
6
7
8
Simulated Sales
Decision
9
Sample
Value
Revenue
10
1variable35.4
424.5
11
2
15.3
183.7
12
3
48.6
582.8
13
4
43.4
521.2
14
5
60.9
603.3
15
6
41.9
503.1
16
7
51.5
603.3
Constraint
17
8
62.9
603.3
58
49
65.6
603.3
59
50
52.0
603.3
60
61
62
OPS Analytics MOOC, Week 4 Session 4

=$C$3*MIN(B10,$C$5)

=$C$4*$C$5

Unit
Cost
150.8
150.8
150.8
150.8
150.8
150.8
150.8
150.8
150.8
150.8

Average =
Standard Deviation =

Total
Profit
273.7
32.8
432.0
370.4
452.5
352.3
452.5
452.5
452.5
452.5
381.9
125.0
6

=C10-D10

Maximization
Objective

=AVERAGE(E10:E1009)
=STDEV(E10:E1009)

Wrap-up for Week 4 Session 4

u We came back to the newsvendor, a fundamental problem in operations
m Given the price, cost, and a demand history choose a good order quantity Q
u We used the demand forecast from week 1, along with simulation and

optimization from weeks 2 and 3 to find a good Q

u With these tools, finding an effective Q was easy
m Simulate a large number of demands
m Optimize to find a Q that maximizes average profit for the sample
m Use the optimal Q for the sample as an estimate of the optimal Q
u Adding a constraint on the standard deviation of the profit was also easy
u Two caveats
m The problems min {D,Q} is not linear see the optional advanced session
m The standard deviation is also non-linear but thats not a problem
OPS Analytics MOOC, Week 4 Session 4