Escolar Documentos
Profissional Documentos
Cultura Documentos
Fixed Costs
John Howell and Greg Allenby
The Ohio State University
Our Problem
Agricultural Equipment Manufacturer
Developed a new machine for detecting diseases
Two components: a test reader and disposable
test strips
Three different test strips to detect three
different diseases
Study is real client study, but attributes and
levels have been disguised
Primary Objective
Estimate the volume of both readers and test
strips sold under different pricing scenarios
CBC not sufficient as it does not directly allow
for volume estimates
Volumes should depend on both the reader
price and the test strip prices
Example Screen
Please select the option you would chose and indicate how many tests you would
purchase for system you selected.
Reader 1
Reader 2
Reader 3
Feature Level 1
Feature Level 2
Feature Level 1
Service Level 1
Service Level 2
Service Level 4
Disease 1: $10/Test
Disease 2: $7/Test
Disease 3: $15/Test
Disease 1: $12/Test
Disease 2: $6/Test
Disease 3: $10/Test
Disease 1: $6/Test
Disease 2: $9/Test
Disease 3: $5/Test
Prior Work
Tom Eagle presented a number of models at
previous conference
Regression based approach
Choice Modeling approaches
Volume = Share * MEV
Whats New?
Fixed Cost: A one time cost that a consumer
must incur in order to participate in a market
Need not be monetary
Examples
Examples
Examples
Examples
Examples
Examples
Fixed Cost
Make purchasing small quantities too costly
leading to no purchases with small quantities
Expect to observe no data with small purchase
quantities
Intuition
A consumer will buy a product as long as the
benefit outweighs the cost
With Fixed Costs
2
1
0
Product 2
4
Product 1
11
. .
12
Product of Interest
Outside Good
. .
12
. .
Expenditure
12
. .
Fixed cost only applies if x is positive
12
Preference Parameter
d
13
Preference Parameter
d
Part-worths
13
Preference Parameter
d
13
14
13
12
11
10
Outside Good
9
8
7
6
x = .4
4
3
Minimum Demand non-zero demand for inside good
x = .8
Bu
dg
et
Co
nst
rai
nt
1
0
Inside Good
14
Example Screen
Please select the option you would chose and indicate how many tests you would
purchase for system you selected.
Reader 1
Reader 2
Reader 3
Feature Level 1
Feature Level 2
Feature Level 1
Service Level 1
Service Level 2
Service Level 4
Disease 1: $10/Test
Disease 2: $7/Test
Disease 3: $15/Test
Disease 1: $12/Test
Disease 2: $6/Test
Disease 3: $10/Test
Disease 1: $6/Test
Disease 2: $9/Test
Disease 3: $5/Test
15
Bayesian Estimation
Bayesian modeling makes this easier
Bayesian Data Augmentation
Estimate so and E easily estimated
16
Sample Characteristics
3 categorical attributes plus 4 continuous
prices
220 respondents from population of 3500
Only 147 were usable for estimation
12 choice tasks
Difficult task
Respondents asked to make product choice and
then provide volume estimates
17
600
400
200
0
Number of Observations
800
Raw Data
11000
14000
18000
21000
18
600
400
0
10
20
30
40
50
200
0
Number of Observations
800
Raw Data
60
80
100
125
143
160
190
11000
14000
18000
21000
18
10
20
30
40
50
Raw Data
g
e
t
o
d
n
i
p
n
a
m
60
80
100
125
143
160
190
19
Model Validation
Compared fit of two holdout tasks
3 Products + None
3 Volume estimates for chosen product
20
25
0
Choice 1
Predicted
*RMSE = 5.93%
Choice 2
Choice 3
Actual
21
25
0
Choice 1
Predicted
*RMSE = 3.55%
Choice 2
Choice 3
Actual
22
Comparison to CBC
CBC does not natively handle volume
estimation
Procedure
Calculate the max test volume for each
respondent (MEV) for each test
Weight the share predictions by individuals
average volume
Volume = ShareReader*MEVDisease
23
40
60
80
100
120
140
20
g
o
N
d
n
i
p
n
a
m
23
50
75
100
125
150
175
200
24
RMSE
FCCA
CBC
Holdout 1
5.93%
10.31%
Holdout 2
3.6%
5.0%
25
RMSE
FCCA
CBC
Holdout 1
768.24
1375.66
Holdout 2
642.89
687.27
26
Example Simulation
Fixed costs screen out low volume
purchasers
Low volume customer may be unprofitable
Optimize allocation of sales, service, and
support resources
27
Example Simulation
Screening Respondents
Base Case
$0 Reader
Purchase Structure 1
Purchase Structure 1
$X,000 Reader
$0 Reader
Feature Level 1
Feature Level 1
Service Level 3
Service Level 3
28
Results
Base Case
$0 Reader
%Change
Customers
17.21
71.3
314%
Disease 1
25,115
66,765
166%
Disease 2
10,309
29,566
187%
Disease 3
1,047
4,692
348%
Revenue
49%
Revenue/
Customer
-64%
29
Summary
Pros:
Theoretically Sound
Accurate volume estimates
Automatically enforces economic constraints
Volumes sensitive to product configuration and all
prices
Allows wider range of possible simulations
Cons:
Estimation more difficult
Simulation more difficult
No packaged software yet
30
Thank You
31