Chapter 5

Sensitivity Analysis

Slide 1 of 31
 Introduction

Sensitivity analysis is used through the entire

modeling process
Purpose of sensitivity analysis:
• To analyze what really matters in the decision problem
• To construct a requisite decision model

Examples of sensitivity analysis techniques in DA:

• Determine if deterministic dominance or stochastic
dominance is present
• Identifying the important variables through tornado
diagrams
• Identify interaction effects between important variables
• Identify the importance of probability assessments
(which are also variables).
Slide 2 of 29
The Eagle Airline Case

Slide 3 of 29
The Eagle Airline Case

Slide 4 of 29
 The Eagle Airline Case

Dick Carothers wants to expand his operation.

Mid West has to offer:

• An airplane @ price $95000.
(He can probably by the plane for $85K-$90K)
• An option to buy the airplane a year later
(Cost of the option $2.5 – $4k)

Currently:
• Eagle Airlines (=Dick Carothers) owns 3 plains
• 50% of flights are chartered flights and 50% are
scheduled

Slide 5 of 29
 The Eagle Airline Case

Cost Data Mid West Plane:

• New Engines, FAA Maintained
• Contains all equipment that Eagle Airlines needs
• Has 5 seats
• Operation Cost: $245 per hour
• Fixed Cost: $20k (=Yearly Insurance) + Finance Charges

Finance Charges:
• Borrow 40% of the price at 2% above the prime rate
(=9.5%, but subject to change).

Revenue Data:
• Chartered Flights: $300 - $400 per hour
• Scheduled Flights: $100 per person per hours, plains are on
average 50% full
• Expected number of hours flown with new plane 800-1000.

Slide 6 of 29
 The Eagle Airline Case

Variables in control:
• The price he is willing to pay.
• The amount financed.

Variables not in control:

• Insurance Cost
• Operation Cost

Carothers could always invest his cash $52,500

@8% yearly interest rate, yielding an annual
interest in the first year of: $4200

What should Dick Carothers do?

Slide 7 of 29
Sensitivity Analysis: Problem Identification Level

Are we solving the right problem?

Error of the 3rd kind: Solving the wrong problem.

How to avoid this error?

• Continue to be skeptical about the problem on the
surface being the real problem

Eagle Airlines Case:

Carothers wants to expand his operation. The fact that he
owns an airline company does not mean he has to
expand by buying another plane. He could, for example,
expand by investing in computer industry.

Slide 8 of 29
 Sensitivity Analysis: Problem Structure Level

Are any of pieces of the puzzle missing?

Is this a single or multiple objective problem?

Sensitivity Analysis: Dominance Considerations

• Ask whether one alternative could end up better than
another. If not, ignore that alternative.

Eagle Airline Case:

“Buying the option” is considered never better than
“Buying the plane” alternative since asking price a
year from now will be adjusted to be similar and hence
“Buying the option” simply adds additional cost.

(You are not learning any thing new by waiting in this case)

Slide 9 of 29
 Sensitivity Analysis: Problem Structure Level

Are any of pieces of the puzzle missing?

Is this a single or multiple objective problem?

Sensitivity Analysis: Dominance Considerations

• Ask whether one alternative could end up better than
another. If not, ignore that alternative.

Eagle Airline Case:

“Buying the option” is considered never better than
“Buying the plane” alternative since asking price a
year from now will be adjusted to be similar and hence
“Buying the option” simply adds additional cost.

(You are not learning any thing new by waiting in this case)

Slide 10 of 29
 Sensitivity Analysis: Problem Structure Level
Sensitivity Analysis: Importance of variables

Back to the Eagle Airlines Case:

Objective: Maximize Profit. Consider Annual Profit,

Ignore Taxes

• Annual Profit =
Annual Total Revenue – Annual Total Cost

• Total Revenue =
Revenue from Charters + Revenue from scheduled flights

• Total Cost = Variable Cost + Fixed Cost

Slide 11 of 29
Sensitivity Analysis: Problem Structure Level
Revenue from Charters:
(Charter Ratio)*(Hours flown per year)*Charter Price

Revenue from Schedules Flights:

(1-Charter Ratio)*(Hours flown per year)*(Ticket price per
hour)*(Number of Seats)*(Average Occupancy)
Fixed Cost:
Insurance + (Purchase Price)*(% Financed)*(Interest Rate)
Variable Cost:
(Hours flown per year)*(Operating Cost)

Step 1: Determine a range for every decision variable and a best

guess (Low, Base, High) and calculate as a first cut sensitivity
analysis, the output variable using first all the low values for the
input variables and second all the high values for the input
variables.
Slide 12 of 29
Sensitivity Analysis: Problem Structure Level

Is this a worst case – best case analysis?

Slide 13 of 29
 One-Way Sensitivity Analysis

STEP 2:
1. Select a particular variable (= free variable)
2. Set all other variables to their best guesses (=base
values)
3. Set free variable to its lowest value and calculate
payoff
4. Set free variable to its highest value calculate
payoff
5. Set free variable to some intermediate values and
calculate payoff
6. Draw results in a one way sensitivity analysis graph

Eagle Airlines Case: Fix all variables, except hours flown.

Slide 14 of 29
 One-Way Sensitivity Analysis
Sensitivity of Profit on Hours Flown
(Keeping Everything Else Fixed)
$35,000

$30,000

$25,000

$20,000

$15,000
Profit

$10,000

$5,000 $4200
$0

-$5,000

-$10,000

-$15,000
500 600 700 800 900 1,000
Hours Flown
Hours Flown Money Market

Slide 15 of 29
 One-Way Sensitivity Analysis

STEP 3: Perform a one-way sensitivity analysis for all variables

and plot results in a Spider Diagram.
$32,500
Hours Flown
$27,500

$22,500
Occupancy
$17,500 Rate

$12,500
Operating
Cost
$7,500

$2,500 Charter Price

-$2,500

-$7,500 Money
Market
-$12,500
-40% -30% -20% -10% 0% 10% 20% 30%

Slide 16 of 29
 One-Way Sensitivity Analysis

STEP 4: Calculated payoff range is a measure of uncertainty in payoff

due to uncertainty in the free variable. Plot the payoff ranges in a
Tornado Diagram and visually determine the important variables.

5 Occupancy
Rate

4
Operating Cost
3
Rank

2 Hours Flown

1
Charter Price

0
$0
-$5,000

$5,000

$10,000

$15,000

$20,000

$25,000

$30,000

$35,000
-$15,000

-$10,000

Money Market

Profit

Slide 17 of 29
 Two-Way Sensitivity Analysis

One-way sensitivity analysis ignores the effect of

changing multiple variables at the same time

"One Way" under estimates sensitivity due to:

• Additive effects of varying more than one variable
• Multiplicative effects of varying more than one
variable
Eagle Airlines Case:
Tornado Diagram indicates that Occupancy
Rate (OR) and Operating Cost (OC) on
scheduled flight are critical (suggesting a two-
way sensitivity analysis of these variables)

Slide 18 of 29
 Two-Way Sensitivity Analysis

Eagle Airlines Case:

1. Determine Annual Profit (AP) as a function of OR and
OC:

AP = R*H*CP + (1-R)*H*TP*NPS*OR-H*OC-I-PP*F*IR

2. Set all other parameters at their base values, yielding

AP(OR,OC) = $130000 + $200000*OR-800*OC-$24025

3. For what values of OR and OC is "buying the plane"

worse than "putting money in the savings account".
Hence for what values of OR and OC is the following
true?
Slide 19 of 29
 Two-Way Sensitivity Analysis

AP < $4200 

$130000 + $200000*OR-800*OC-$24025 < $4200 

$200000*OR < $800*OC -$101775 

• First, draw graph of values of OR and OC such that

one is indifferent between "buying the plane" and "the
saving account".
OR = 0.004*OC - 0.509
• Second, determine which alternative is preferred
above and below the indifference curve.

Slide 20 of 29
 Two-Way Sensitivity Analysis

Two Way Strategy Region of Eagle Airline Case for Operating

Cost and Occupancy Rate

54%

52%

50% AP(OC,OP)>$4200 $9975

Occupancy Rate

48%

46%

44%
AP(OC,OP)<$4200
42%

40%
$230 $235 $240 $245 $250 $255 $260
Operating Cost

Indifference Line Base Case Values

Slide 21 of 29
Two-Way Sensitivity Analysis: Texaco - Pennzoil
Max Settlement Amount
Accept $2 Billion
($ Billion )
2

Texaco Accepts $5 Billion (0.17)

5

High (p)
10.3
Counteroffer Texaco Refuses (0.50) Medium (q)
Final Court 5
Decision
$5 Billion Counteroffer
Low (1-p-q)
0

High (p)
10.3

Refuse Final Court Medium (q) 5

Decision
Texaco Low (1-p-q)
(0.33) 0
Counter -
offers $3 Billion
Accept $3 Billion
3

Liedtke is unsure of court probabilities. If Liedtke thinks that p

must be more than 0.15 and q must be more than 0.35 can he
make the decision without further probability assessment?
Slide 22 of 29
 Two-Way Sensitivity Analysis: Texaco - Pennzoil

Step 1: Create a two-way sensitivity graph that shows optimal

strategies for Liedtke for all possible values of p and q.
Strategy A = Accept $2 billion.

Strategy B = Counteroffer $5 billion, then refuse if Texaco offers $3 billion.

Strategy C = Counteroffer $5 billion, then accept if Texaco offers $3 billion.

EMV(A) = 2

EMV(B) = 0.17 (5) + 0.5 [p 10.3 + q 5 + (1-p - q) 0] + 0.33 [p 10.3 + q 5 + (1-p - q) 0]

= 0.85 + 8.549 p + 4.15 q.

EMV(C) = 0.17 (5) + 0.5 [p 10.3 + q 5 + (1-p - q) 0] + 0.33 (3)

= 1.85 + 5.15 p + 2.5 q.

Slide 23 of 29
Two-Way Sensitivity Analysis: Texaco - Pennzoil

Now construct three inequalities:

• EMV(A) > EMV(B) 
2 > 0.85 + 8.549 p + 4.15 q 
0.135 - 0.485 q > p . (1)

• EMV(A) > EMV(C) 

2 > 1.85 + 5.15 p + 2.5 q 
0.03 - 0.485 q > p . (2)

• EMV(B) > EMV(C) 

0.85 + 8.549 p + 4.15 q > 1.85 + 5.15 p + 2.5 q 
0.294 - 0.485 q < p. (3)

Plot three indifference lines on a graph with p on the

vertical axis and q on the horizontal axis. Note that only
the region below the line p + q = 1 is feasible because
p + q must be less than or equal to one.
Slide 24 of 29
 Two-Way Sensitivity Analysis: Texaco - Pennzoil

• These three lines 1.0

divide the graph into 0.9

four separate 0.8
regions, labeled I,
II, III, and IV. 0.7

0.6

0.5 p=1-q

0.4
p = 0.294 - 0.485 q
0.3 I
p = 0.135 - 0.485 q 0.2

p = 0.03 - 0.485 q II
0.1
IV III
0 q
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Slide 25 of 29
Two-Way Sensitivity Analysis: Texaco - Pennzoil
p

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3 I

A-C 0.2 B
B
II
0.1
IV III
0 q
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
A
B-C
A-B
Slide 26 of 29
 Two-Way Sensitivity Analysis: Texaco - Pennzoil

Inequality (3) divides regions I and II. For points above this line,
p > 0.294 - 0.485 q, and so EMV(B) > EMV (C).
Inequality (1) divides regions II and III. For points above this line,
p > 0.135 - 0.485 q, and EMV(B) > EMV(A). As a result of this, we
know that B is the preferred choice in region I and that C is the
preferred choice in region II [where EMV(C) > EMV (B) > EMV(A)].
Inequality (2) divides regions III and IV. For points above this
line, p > 0.03 - 0.485 q, and EMV(C) > EMV (A). Thus, we now know
that C is the preferred choice in region III [where EMV(C) > EMV(A)
and EMV(C) > EMV(B)], and A is preferred in region IV.

Thus, we can redraw the graph, eliminating

the line between regions II and III

Slide 27 of 29
 Two-Way Sensitivity Analysis: Texaco - Pennzoil

1.0 • The shaded area in

the figure represents
0.9
those points for which
0.8 p > 0.15 and q > 0.35.
0.7
p > 0.15 • Note that all of these
0.6
q > 0.35 points fall in the
0.5 “Choose B” region.
0.4
B • Thus, Liedtke should
0.3
adopt strategy B:
0.2

0.1 C Counteroffer $5 billion,

A then refuse if Texaco
0 q
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
offers $3 billion.

Slide 28 of 29
 Uncertainty Analysis

• Thus far, one-way sensitivity analysis and two-way sensitivity

analysis acknowledge that output parameters are uncertain
by indicating a range for the different input variables.
• Of course, we can conduct a three-way sensitivity analysis,
a four- way etc.?
• Perhaps we should vary all the parameters at the same time.
The latter is called an: Uncertainty Analysis (Chapter 11).
• Parameters are uncertain as indicated by assessing a range. By
specifying probability distributions for uncertainty of input
parameters we assess how uncertain these parameters are.
• Given the uncertainty distribution of the input parameters and
the calculation model the uncertainty distribution of the output
parameters is fixed.

Slide 29 of 29
 Uncertainty Analysis

3.50

3.00

2.50

2.00

1.50

1.00

0.50
INPUT UNCERTAINTY

0.00
0.00 0.20 0.40 0.60 0.80 1.00

OC

3.50
MODEL OUTPUT
3.00

2.50
UNCERTAINTY
2.00 BLACK 1.40

BOX
1.50
1.20
1.00
1.00
0.50

0.00
MODEL 0.80

0.00 0.20 0.40 0.60 0.80 1.00 0.60

OR 0.40

0.20

0.00
0.00 0.20 0.40 0.60 0.80 1.00

AP
3.50

3.00

2.50

2.00

1.50

1.00

0.50

0.00
0.00 0.20 0.40 0.60 0.80 1.00

H
Slide 30 of 29
 Uncertainty Analysis

3.50
Calculate
3.00
Sample X1,Y1,Z 1 O1

STATISTICS
2.50

2.00

1.50

1.00

Calculate
Sample X2,Y2,Z 2 O2
0.50

0.00
INPUT UNCERTAINTY

OC
0.00 0.20 0.40 0.60 0.80 1.00

Calculate
3.50

3.00 Sample X3,Y3,Z 3 O3

2.50

2.00

1.50

1.00

0.50
ETC ...
0.00
0.00 0.20 0.40 0.60 0.80 1.00

OR
OUTPUT
2.50 UNCERTAINTY
2.00
1.40

1.50 1.20

1.00
MODEL = 1.00

0.80
0.50
F(X,Y,Z) 0.60

0.00 0.40
0.00 0.20 0.40 0.60 0.80 1.00

0.20

H 0.00
0.00 0.20 0.40 0.60 0.80 1.00 AP
Slide 31 of 29