SlideShare a Scribd company logo
1 of 17
A Resource Allocation Method
using a Non-parametric Approach
ICIBM2018
Tomohiro Noguchic,
Nobuyuki Tachibanaa,
Susumu Kadoyab,
Takashi Namatamed
a,b,c BrainPad Inc., Analytics Service Division
d Department of Indastrial and Systems Engineering, Chuo University
2018/06/14
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
1
Outline
1. Background & Objectives
2. Budget Allocation
3. Non Parametric Models
4. Proposed Method (iFDH)
5. Empirical Test
6. Conclusion
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
2
Background & Objectives
• Since the difficulty of specifying advertising contribution structure, we studied non
parametric method, Data Envelopment Analysis (DEA) , to apply the problem.
• Then, we propose a new MMM methodology that can be applied to a non-convex
shape frontier.
• Efficient budget plans are increasingly becoming important in mature developed
countries, and the market mix model (MMM) plays an important role.
• When products are advertised, in many cases, multiple advertising media are used.
However, the sales contributions cannot be decomposed into individual advertising
media effect since we do not directly know the relationship between the product
sales and the investment costs in each media.
• Despite these limitations, to allocate the budget based on the media contribution,
we must assume the relationship between the individual media investments and
the product sales.
• Conveniently, we use parametric formula to estimate the relationships. (such like
regression model) However, cross effects exist among a plurality of mediums
make difficult to estimate the accurate parameters (Even when there is no cross
effect it is difficult to estimate parametrically though).
BackgroundObjectives
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
3
Budget Allocation
Sales cannot be decomposed into individual advertising media based on their contributions.
Therefore, it is difficult for decision-makers to decide their budget.
Decision Makers
Our Total Budget is 15M$
TV
WEB
Media
Call Center
WEB Page
EC
Sales
Medius Budget
TV 10M?
WEB 5M?
TV
Sales
Sales
Cost of TV Cost of WEB Total Cost
Media A
Unobservable
Medea B
Unobservable
A + B
Observable
① ② ③WEB
Sales Basically using a parametric
formula to estimate the model
parameters like regressions.
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
4
It is common that sales response to advertisement costs eventually saturates.
Considering that BCC and FDH are available for budget allocation
Non-Parametric Approach | DEA (Data Envelopment Analysis)
• DEA is a multi-criteria evaluation method that can select the most favorable alternatives
from a large set when there is no parametric assumption among variables. It uses a
mathematical programming technique where the most favorable alternatives form an
effective frontier
• Outline the model properties for DEA(CCR and BCC) and FDH(Free Disposal Hull), one-
input one-output case
CCR
The initial DEA is a CCR model, which
assumes constant RTS
Evaluate all DMUs based on the most
effective DMU.
But this methodology is not applicable to
the case of decreasing returns to scale
(DRS)
Sales
Media Cost
BCC
BCC is more suitable in capturing the
frontier line in the case of a concave
type.(We assume that the sales
response function to advertisement
costs can be a sigmoid or concave
shape based on numerous marketing
theories)
FDH
Free Disposal Hull
model assumes the free
disposability relaxing the convexity
assumption in defining
the production possibility set from
the observations. FDH is more
suitable in capturing the frontier
line in the case of a sigmoid
Media CostMedia Cost
Sales Sales
DMU : Decision Making Unit
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
5
Problem of BCC and FDH
【BCC】
•BCC can not correspond to Non Convex. When Sales response is Sigmoid, there
is a Gap
【FDH】
•The frontier is represented as a step-like form. This means that there is a region
where the output does not increase even if the input does. The step-like frontier
shape is inadequate except in the case where sales are saturated
Input 1
Input2
Media Cost
Sales
Gap
Input2 > Input1. But got same output
Media Cost
Sales
In low cost area,There is a GAP
between estimation frontier
andrreal sales response curve
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
6
Figures of procedure | Proposed Method
x2
y
x1
x2
y
x1
x2
y
x1
x2
y
x1
DMU A DMU B DMU C DMU D
Input 1 80 30 90 50
Input 2 20 70 10 50
DMU A DMU B DMU C DMU D
Input 1 0.8 0.3 0.9 0.5
Input 2 0.2 0.7 0.1 0.5
DMU A DMU B DMU C DMU D
Input 1 160 60 180 100
Input 2 40 140 20 100
x2
y
x1
Step0:Observe DMUs Step1: Calculate FDH Step2:Delaunay triangulation.(Smoothing)
Step3: The individual media investment set
in monetary amounts
Step4: assigned to the regions Step5: The achievable maximum
sales are calculated
y
xView from the side
✖✖ M$
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
7
Propose a method to compensate for the disadvantages(step like frontier) of FDH
which can express non-convexity
Our proposed procedure is as follows:
1. Calculate the FDH frontier using the observed DMU set
2. Select only the efficient DMUs by FDH. Then, applying Delaunay triangulation to their
inputs as a vertex set, divide the frontier area into triangles (higher-dimensional simplex
such as tetrahedron). Here, we do not consider output values
3. Optimal allocation candidates are the observed DMU allocations. An advertising budget
is given, the candidate allocation can be transferred to the individual media investment
set in monetary amounts
4. The allocation candidates are assigned to the regions where they belong, which is the
divided frontier by the triangulation
5. In terms of the assigned region in Step 4, estimate the hyperplane, which includes the
vertex of the triangulation. In this hyperplane estimation process, not only the inputs, but
also the output values are considered. Consequently, the FDH frontier is smoothed by
this hyperplane.
The achievable maximum sales are calculated on the hyperplane for the allocation
candidates in their assigned areas
Our Proposed Method
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
8
Empirical Tests 1/2
Test purpose :
• Compare the four estimation errors of the model(Parametric, BCC, FDH,
Proposed method).
• Through this test, To show that nonparametric is appropriate when the structure
is unknown and that the proposed method gives a more accurate estimate
Procedure of test :
① Define the true relationships between the advertising costs and sales by a set of
equations that are simplified versions of the ADBUDG* formula (call true model)
② Generate scenarios based on this true model(make a test datasets)
③ Outputs can be calculated based on following Tested models .”Parametric Model”, “DEA
model”, “FDH model” ,and ”Our proposed model(Interpolated FDH | iFDH)”
④ Comparing the model accuracy
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
9
Empirical Tests 2/2
Test overview :
• There are two advertising media
• Assume following cases:
• The interaction between the two media and without any interaction
• The individual response function of sales to the advertisement costs is considered to be a sigmoid
shape or concave
Media A Media B
True
Media A Media B
False
Media BMedia A
sigmoid & concave w/ DRS
Media BMedia A
sigmoid & sigmoid
Considering the six cases (three function combinations & interaction)
Have interaction or Not :
x1 x2 x1 x2
Media BMedia A
concave w/ DRS & concave w/ DRS
x1 x2
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
10
Empirical Tests | Result
Parametric DEA FDH IFDH
a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009
2 1.040 0.005 0.038 0.006
3 0.571 0.002 0.022 0.001
4 0.694 0.001 0.021 0.002
5 0.633 0.002 0.026 0.003
Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115
2 1.138 0.278 0.929 0.192
3 1.062 0.035 0.152 0.035
4 1.372 0.004 0.044 0.008
5 1.639 0.002 0.030 0.003
Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589
2 0.894 0.921 0.792 0.348
3 1.068 0.190 0.248 0.057
4 0.959 0.032 0.217 0.019
5 0.739 0.009 0.101 0.009
a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020
2 1.051 0.007 0.056 0.008
3 0.819 0.002 0.041 0.002
4 1.041 0.003 0.022 0.003
5 0.950 0.003 0.033 0.005
Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133
2 2.341 0.224 0.881 0.132
3 2.996 0.018 0.129 0.022
4 3.813 0.011 0.105 0.016
5 4.573 0.003 0.048 0.003
Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585
2 1.337 1.594 0.705 0.322
3 2.647 0.299 0.470 0.069
4 3.188 0.055 0.326 0.041
5 2.204 0.021 0.199 0.021
Average 1.563 0.215 0.299 0.093
Model Type
BudgetResponse Function Type CombinationInteraction
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
11
Empirical Tests | Result
Parametric DEA FDH IFDH
a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009
2 1.040 0.005 0.038 0.006
3 0.571 0.002 0.022 0.001
4 0.694 0.001 0.021 0.002
5 0.633 0.002 0.026 0.003
Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115
2 1.138 0.278 0.929 0.192
3 1.062 0.035 0.152 0.035
4 1.372 0.004 0.044 0.008
5 1.639 0.002 0.030 0.003
Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589
2 0.894 0.921 0.792 0.348
3 1.068 0.190 0.248 0.057
4 0.959 0.032 0.217 0.019
5 0.739 0.009 0.101 0.009
a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020
2 1.051 0.007 0.056 0.008
3 0.819 0.002 0.041 0.002
4 1.041 0.003 0.022 0.003
5 0.950 0.003 0.033 0.005
Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133
2 2.341 0.224 0.881 0.132
3 2.996 0.018 0.129 0.022
4 3.813 0.011 0.105 0.016
5 4.573 0.003 0.048 0.003
Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585
2 1.337 1.594 0.705 0.322
3 2.647 0.299 0.470 0.069
4 3.188 0.055 0.326 0.041
5 2.204 0.021 0.199 0.021
Average 1.563 0.215 0.299 0.093
Model Type
BudgetResponse Function Type CombinationInteraction
1
1. The parametric models’
accuracies are lower than those
of the non-parametric model.
Parametric models is
remarkable in the case where
interaction exists between the
two media. This means that the
structure complexity causes
difficulties in calibration of
parameters in the parametric
model
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
12
Empirical Tests | Result
Parametric DEA FDH IFDH
a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009
2 1.040 0.005 0.038 0.006
3 0.571 0.002 0.022 0.001
4 0.694 0.001 0.021 0.002
5 0.633 0.002 0.026 0.003
Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115
2 1.138 0.278 0.929 0.192
3 1.062 0.035 0.152 0.035
4 1.372 0.004 0.044 0.008
5 1.639 0.002 0.030 0.003
Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589
2 0.894 0.921 0.792 0.348
3 1.068 0.190 0.248 0.057
4 0.959 0.032 0.217 0.019
5 0.739 0.009 0.101 0.009
a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020
2 1.051 0.007 0.056 0.008
3 0.819 0.002 0.041 0.002
4 1.041 0.003 0.022 0.003
5 0.950 0.003 0.033 0.005
Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133
2 2.341 0.224 0.881 0.132
3 2.996 0.018 0.129 0.022
4 3.813 0.011 0.105 0.016
5 4.573 0.003 0.048 0.003
Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585
2 1.337 1.594 0.705 0.322
3 2.647 0.299 0.470 0.069
4 3.188 0.055 0.326 0.041
5 2.204 0.021 0.199 0.021
Average 1.563 0.215 0.299 0.093
Model Type
BudgetResponse Function Type CombinationInteraction
2
2
2. The DEA model is more accurate than the original FDH. But some case of
sigmoid type combination, FDH is more accurate.
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
13
Empirical Tests | Result
Parametric DEA FDH IFDH
a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009
2 1.040 0.005 0.038 0.006
3 0.571 0.002 0.022 0.001
4 0.694 0.001 0.021 0.002
5 0.633 0.002 0.026 0.003
Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115
2 1.138 0.278 0.929 0.192
3 1.062 0.035 0.152 0.035
4 1.372 0.004 0.044 0.008
5 1.639 0.002 0.030 0.003
Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589
2 0.894 0.921 0.792 0.348
3 1.068 0.190 0.248 0.057
4 0.959 0.032 0.217 0.019
5 0.739 0.009 0.101 0.009
a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020
2 1.051 0.007 0.056 0.008
3 0.819 0.002 0.041 0.002
4 1.041 0.003 0.022 0.003
5 0.950 0.003 0.033 0.005
Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133
2 2.341 0.224 0.881 0.132
3 2.996 0.018 0.129 0.022
4 3.813 0.011 0.105 0.016
5 4.573 0.003 0.048 0.003
Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585
2 1.337 1.594 0.705 0.322
3 2.647 0.299 0.470 0.069
4 3.188 0.055 0.326 0.041
5 2.204 0.021 0.199 0.021
Average 1.563 0.215 0.299 0.093
Model Type
BudgetResponse Function Type CombinationInteraction
3
3. The best model is the IFDH. Therefore, the inferiority of the FDH against the
DEA may be caused by the stair-like surface of the frontier.
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
14
Additional Test ~ Effectiveness of smoothing technique
Improvement
DEA (A) FDH (B) Diff1 (A)-(B) DEA (C) FDH (D) Diff2 (C)-(D) Diff2 - Diff1
a=0 Concave with DRS & Concave with DRS 1 0.01 0.05 -0.04 0.00 0.03 -0.03 0.01
2 0.00 0.04 -0.03 0.00 0.02 -0.02 0.01
3 0.00 0.02 -0.02 0.00 0.02 -0.02 0.01
4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01
5 0.00 0.03 -0.02 0.00 0.01 -0.01 0.02
Concave with DRS & Sigmoid 1 0.21 0.42 -0.21 0.12 0.15 -0.03 0.18
2 0.28 0.93 -0.65 0.07 0.32 -0.25 0.40
3 0.03 0.15 -0.12 0.01 0.08 -0.07 0.05
4 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02
5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02
Sigmoid & Sigmoid 1 0.99 1.16 -0.18 0.52 0.66 -0.14 0.04
2 0.92 0.79 0.13 0.79 0.48 0.31 0.18
3 0.19 0.25 -0.06 0.27 0.11 0.16 0.21
4 0.03 0.22 -0.19 0.02 0.06 -0.04 0.15
5 0.01 0.10 -0.09 0.01 0.04 -0.03 0.06
a=1 Concave with DRS & Concave with DRS 1 0.02 0.06 -0.05 0.00 0.05 -0.05 0.00
2 0.01 0.06 -0.05 0.00 0.02 -0.02 0.03
3 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02
4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01
5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02
Concave with DRS & Sigmoid 1 0.24 0.47 -0.23 0.11 0.21 -0.10 0.13
2 0.22 0.88 -0.66 0.06 0.36 -0.30 0.35
3 0.02 0.13 -0.11 0.01 0.08 -0.07 0.04
4 0.01 0.11 -0.09 0.01 0.06 -0.05 0.04
5 0.00 0.05 -0.05 0.00 0.02 -0.02 0.02
Sigmoid & Sigmoid 1 1.26 1.16 0.10 1.39 0.68 0.71 0.61
2 1.59 0.70 0.89 1.53 0.45 1.08 0.19
3 0.30 0.47 -0.17 0.20 0.26 -0.06 0.11
4 0.05 0.33 -0.27 0.03 0.09 -0.07 0.21
5 0.02 0.20 -0.18 0.01 0.06 -0.05 0.13
Average 0.21 0.30 -0.08 0.17 0.15 0.02 0.11
Interaction Response Function Type Combination Budget
200 DMUs 600 DMUs
We increased the number of DMUs to reduce the step level differences in the FDH
frontier, and investigate the improvement in performance before and after
increasing the DMUs
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
15
Additional Test ~ Effectiveness of smoothing technique
Improvement
DEA (A) FDH (B) Diff1 (A)-(B) DEA (C) FDH (D) Diff2 (C)-(D) Diff2 - Diff1
a=0 Concave with DRS & Concave with DRS 1 0.01 0.05 -0.04 0.00 0.03 -0.03 0.01
2 0.00 0.04 -0.03 0.00 0.02 -0.02 0.01
3 0.00 0.02 -0.02 0.00 0.02 -0.02 0.01
4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01
5 0.00 0.03 -0.02 0.00 0.01 -0.01 0.02
Concave with DRS & Sigmoid 1 0.21 0.42 -0.21 0.12 0.15 -0.03 0.18
2 0.28 0.93 -0.65 0.07 0.32 -0.25 0.40
3 0.03 0.15 -0.12 0.01 0.08 -0.07 0.05
4 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02
5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02
Sigmoid & Sigmoid 1 0.99 1.16 -0.18 0.52 0.66 -0.14 0.04
2 0.92 0.79 0.13 0.79 0.48 0.31 0.18
3 0.19 0.25 -0.06 0.27 0.11 0.16 0.21
4 0.03 0.22 -0.19 0.02 0.06 -0.04 0.15
5 0.01 0.10 -0.09 0.01 0.04 -0.03 0.06
a=1 Concave with DRS & Concave with DRS 1 0.02 0.06 -0.05 0.00 0.05 -0.05 0.00
2 0.01 0.06 -0.05 0.00 0.02 -0.02 0.03
3 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02
4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01
5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02
Concave with DRS & Sigmoid 1 0.24 0.47 -0.23 0.11 0.21 -0.10 0.13
2 0.22 0.88 -0.66 0.06 0.36 -0.30 0.35
3 0.02 0.13 -0.11 0.01 0.08 -0.07 0.04
4 0.01 0.11 -0.09 0.01 0.06 -0.05 0.04
5 0.00 0.05 -0.05 0.00 0.02 -0.02 0.02
Sigmoid & Sigmoid 1 1.26 1.16 0.10 1.39 0.68 0.71 0.61
2 1.59 0.70 0.89 1.53 0.45 1.08 0.19
3 0.30 0.47 -0.17 0.20 0.26 -0.06 0.11
4 0.05 0.33 -0.27 0.03 0.09 -0.07 0.21
5 0.02 0.20 -0.18 0.01 0.06 -0.05 0.13
Average 0.21 0.30 -0.08 0.17 0.15 0.02 0.11
Interaction Response Function Type Combination Budget
200 DMUs 600 DMUs
The results indicate that surface
smoothing improves the model
performance.
Analytics Innovation Company
©BrainPad Inc.
Strictly Confidential
16
Conclusion
• We propose a non-parametric model, which can be applied to a non-convex shape frontier
for MMM
• From the result of these tests, our proposed model, IFDH, shows the highest performance
from the perspective of minimizing the error
• The technique that we have introduced to smooth the FDH frontier surface seems to work
well because the DEA model performance is better than the original FDH
• To confirm this, we increased the number of DMUs to reduce the step level differences in
the FDH frontier, and investigate the improvement in performance before and after
increasing the DMUs
Some future work:
Our model has been proven to work on artificial data, it should be tested in a real-world
context. Furthermore, the robustness of the model is still subject to be confirmed.

More Related Content

Recently uploaded

Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
shivangimorya083
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
amitlee9823
 
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
amitlee9823
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
shivangimorya083
 

Recently uploaded (20)

Smarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptxSmarteg dropshipping via API with DroFx.pptx
Smarteg dropshipping via API with DroFx.pptx
 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signals
 
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls CP 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
 
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
 
Zuja dropshipping via API with DroFx.pptx
Zuja dropshipping via API with DroFx.pptxZuja dropshipping via API with DroFx.pptx
Zuja dropshipping via API with DroFx.pptx
 
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
 
BigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptxBigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptx
 
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
 
Call me @ 9892124323 Cheap Rate Call Girls in Vashi with Real Photo 100% Secure
Call me @ 9892124323  Cheap Rate Call Girls in Vashi with Real Photo 100% SecureCall me @ 9892124323  Cheap Rate Call Girls in Vashi with Real Photo 100% Secure
Call me @ 9892124323 Cheap Rate Call Girls in Vashi with Real Photo 100% Secure
 
Introduction-to-Machine-Learning (1).pptx
Introduction-to-Machine-Learning (1).pptxIntroduction-to-Machine-Learning (1).pptx
Introduction-to-Machine-Learning (1).pptx
 
BabyOno dropshipping via API with DroFx.pptx
BabyOno dropshipping via API with DroFx.pptxBabyOno dropshipping via API with DroFx.pptx
BabyOno dropshipping via API with DroFx.pptx
 
Edukaciniai dropshipping via API with DroFx
Edukaciniai dropshipping via API with DroFxEdukaciniai dropshipping via API with DroFx
Edukaciniai dropshipping via API with DroFx
 
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdfMarket Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
Market Analysis in the 5 Largest Economic Countries in Southeast Asia.pdf
 
Ravak dropshipping via API with DroFx.pptx
Ravak dropshipping via API with DroFx.pptxRavak dropshipping via API with DroFx.pptx
Ravak dropshipping via API with DroFx.pptx
 
Midocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFxMidocean dropshipping via API with DroFx
Midocean dropshipping via API with DroFx
 
Cheap Rate Call girls Sarita Vihar Delhi 9205541914 shot 1500 night
Cheap Rate Call girls Sarita Vihar Delhi 9205541914 shot 1500 nightCheap Rate Call girls Sarita Vihar Delhi 9205541914 shot 1500 night
Cheap Rate Call girls Sarita Vihar Delhi 9205541914 shot 1500 night
 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptx
 
Sampling (random) method and Non random.ppt
Sampling (random) method and Non random.pptSampling (random) method and Non random.ppt
Sampling (random) method and Non random.ppt
 

Featured

Social Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie InsightsSocial Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie Insights
Kurio // The Social Media Age(ncy)
 
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them wellGood Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
Saba Software
 
Introduction to C Programming Language
Introduction to C Programming LanguageIntroduction to C Programming Language
Introduction to C Programming Language
Simplilearn
 

Featured (20)

How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024
 
Social Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie InsightsSocial Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie Insights
 
Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024
 
5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary
 
ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd
 
Getting into the tech field. what next
Getting into the tech field. what next Getting into the tech field. what next
Getting into the tech field. what next
 
Google's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search IntentGoogle's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search Intent
 
How to have difficult conversations
How to have difficult conversations How to have difficult conversations
How to have difficult conversations
 
Introduction to Data Science
Introduction to Data ScienceIntroduction to Data Science
Introduction to Data Science
 
Time Management & Productivity - Best Practices
Time Management & Productivity -  Best PracticesTime Management & Productivity -  Best Practices
Time Management & Productivity - Best Practices
 
The six step guide to practical project management
The six step guide to practical project managementThe six step guide to practical project management
The six step guide to practical project management
 
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
 
Unlocking the Power of ChatGPT and AI in Testing - A Real-World Look, present...
Unlocking the Power of ChatGPT and AI in Testing - A Real-World Look, present...Unlocking the Power of ChatGPT and AI in Testing - A Real-World Look, present...
Unlocking the Power of ChatGPT and AI in Testing - A Real-World Look, present...
 
12 Ways to Increase Your Influence at Work
12 Ways to Increase Your Influence at Work12 Ways to Increase Your Influence at Work
12 Ways to Increase Your Influence at Work
 
ChatGPT webinar slides
ChatGPT webinar slidesChatGPT webinar slides
ChatGPT webinar slides
 
More than Just Lines on a Map: Best Practices for U.S Bike Routes
More than Just Lines on a Map: Best Practices for U.S Bike RoutesMore than Just Lines on a Map: Best Practices for U.S Bike Routes
More than Just Lines on a Map: Best Practices for U.S Bike Routes
 
Ride the Storm: Navigating Through Unstable Periods / Katerina Rudko (Belka G...
Ride the Storm: Navigating Through Unstable Periods / Katerina Rudko (Belka G...Ride the Storm: Navigating Through Unstable Periods / Katerina Rudko (Belka G...
Ride the Storm: Navigating Through Unstable Periods / Katerina Rudko (Belka G...
 
Barbie - Brand Strategy Presentation
Barbie - Brand Strategy PresentationBarbie - Brand Strategy Presentation
Barbie - Brand Strategy Presentation
 
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them wellGood Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
Good Stuff Happens in 1:1 Meetings: Why you need them and how to do them well
 
Introduction to C Programming Language
Introduction to C Programming LanguageIntroduction to C Programming Language
Introduction to C Programming Language
 

A Resource Allocation Method using a Non-parametric Approach

  • 1. A Resource Allocation Method using a Non-parametric Approach ICIBM2018 Tomohiro Noguchic, Nobuyuki Tachibanaa, Susumu Kadoyab, Takashi Namatamed a,b,c BrainPad Inc., Analytics Service Division d Department of Indastrial and Systems Engineering, Chuo University 2018/06/14
  • 2. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 1 Outline 1. Background & Objectives 2. Budget Allocation 3. Non Parametric Models 4. Proposed Method (iFDH) 5. Empirical Test 6. Conclusion
  • 3. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 2 Background & Objectives • Since the difficulty of specifying advertising contribution structure, we studied non parametric method, Data Envelopment Analysis (DEA) , to apply the problem. • Then, we propose a new MMM methodology that can be applied to a non-convex shape frontier. • Efficient budget plans are increasingly becoming important in mature developed countries, and the market mix model (MMM) plays an important role. • When products are advertised, in many cases, multiple advertising media are used. However, the sales contributions cannot be decomposed into individual advertising media effect since we do not directly know the relationship between the product sales and the investment costs in each media. • Despite these limitations, to allocate the budget based on the media contribution, we must assume the relationship between the individual media investments and the product sales. • Conveniently, we use parametric formula to estimate the relationships. (such like regression model) However, cross effects exist among a plurality of mediums make difficult to estimate the accurate parameters (Even when there is no cross effect it is difficult to estimate parametrically though). BackgroundObjectives
  • 4. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 3 Budget Allocation Sales cannot be decomposed into individual advertising media based on their contributions. Therefore, it is difficult for decision-makers to decide their budget. Decision Makers Our Total Budget is 15M$ TV WEB Media Call Center WEB Page EC Sales Medius Budget TV 10M? WEB 5M? TV Sales Sales Cost of TV Cost of WEB Total Cost Media A Unobservable Medea B Unobservable A + B Observable ① ② ③WEB Sales Basically using a parametric formula to estimate the model parameters like regressions.
  • 5. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 4 It is common that sales response to advertisement costs eventually saturates. Considering that BCC and FDH are available for budget allocation Non-Parametric Approach | DEA (Data Envelopment Analysis) • DEA is a multi-criteria evaluation method that can select the most favorable alternatives from a large set when there is no parametric assumption among variables. It uses a mathematical programming technique where the most favorable alternatives form an effective frontier • Outline the model properties for DEA(CCR and BCC) and FDH(Free Disposal Hull), one- input one-output case CCR The initial DEA is a CCR model, which assumes constant RTS Evaluate all DMUs based on the most effective DMU. But this methodology is not applicable to the case of decreasing returns to scale (DRS) Sales Media Cost BCC BCC is more suitable in capturing the frontier line in the case of a concave type.(We assume that the sales response function to advertisement costs can be a sigmoid or concave shape based on numerous marketing theories) FDH Free Disposal Hull model assumes the free disposability relaxing the convexity assumption in defining the production possibility set from the observations. FDH is more suitable in capturing the frontier line in the case of a sigmoid Media CostMedia Cost Sales Sales DMU : Decision Making Unit
  • 6. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 5 Problem of BCC and FDH 【BCC】 •BCC can not correspond to Non Convex. When Sales response is Sigmoid, there is a Gap 【FDH】 •The frontier is represented as a step-like form. This means that there is a region where the output does not increase even if the input does. The step-like frontier shape is inadequate except in the case where sales are saturated Input 1 Input2 Media Cost Sales Gap Input2 > Input1. But got same output Media Cost Sales In low cost area,There is a GAP between estimation frontier andrreal sales response curve
  • 7. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 6 Figures of procedure | Proposed Method x2 y x1 x2 y x1 x2 y x1 x2 y x1 DMU A DMU B DMU C DMU D Input 1 80 30 90 50 Input 2 20 70 10 50 DMU A DMU B DMU C DMU D Input 1 0.8 0.3 0.9 0.5 Input 2 0.2 0.7 0.1 0.5 DMU A DMU B DMU C DMU D Input 1 160 60 180 100 Input 2 40 140 20 100 x2 y x1 Step0:Observe DMUs Step1: Calculate FDH Step2:Delaunay triangulation.(Smoothing) Step3: The individual media investment set in monetary amounts Step4: assigned to the regions Step5: The achievable maximum sales are calculated y xView from the side ✖✖ M$
  • 8. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 7 Propose a method to compensate for the disadvantages(step like frontier) of FDH which can express non-convexity Our proposed procedure is as follows: 1. Calculate the FDH frontier using the observed DMU set 2. Select only the efficient DMUs by FDH. Then, applying Delaunay triangulation to their inputs as a vertex set, divide the frontier area into triangles (higher-dimensional simplex such as tetrahedron). Here, we do not consider output values 3. Optimal allocation candidates are the observed DMU allocations. An advertising budget is given, the candidate allocation can be transferred to the individual media investment set in monetary amounts 4. The allocation candidates are assigned to the regions where they belong, which is the divided frontier by the triangulation 5. In terms of the assigned region in Step 4, estimate the hyperplane, which includes the vertex of the triangulation. In this hyperplane estimation process, not only the inputs, but also the output values are considered. Consequently, the FDH frontier is smoothed by this hyperplane. The achievable maximum sales are calculated on the hyperplane for the allocation candidates in their assigned areas Our Proposed Method
  • 9. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 8 Empirical Tests 1/2 Test purpose : • Compare the four estimation errors of the model(Parametric, BCC, FDH, Proposed method). • Through this test, To show that nonparametric is appropriate when the structure is unknown and that the proposed method gives a more accurate estimate Procedure of test : ① Define the true relationships between the advertising costs and sales by a set of equations that are simplified versions of the ADBUDG* formula (call true model) ② Generate scenarios based on this true model(make a test datasets) ③ Outputs can be calculated based on following Tested models .”Parametric Model”, “DEA model”, “FDH model” ,and ”Our proposed model(Interpolated FDH | iFDH)” ④ Comparing the model accuracy
  • 10. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 9 Empirical Tests 2/2 Test overview : • There are two advertising media • Assume following cases: • The interaction between the two media and without any interaction • The individual response function of sales to the advertisement costs is considered to be a sigmoid shape or concave Media A Media B True Media A Media B False Media BMedia A sigmoid & concave w/ DRS Media BMedia A sigmoid & sigmoid Considering the six cases (three function combinations & interaction) Have interaction or Not : x1 x2 x1 x2 Media BMedia A concave w/ DRS & concave w/ DRS x1 x2
  • 11. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 10 Empirical Tests | Result Parametric DEA FDH IFDH a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009 2 1.040 0.005 0.038 0.006 3 0.571 0.002 0.022 0.001 4 0.694 0.001 0.021 0.002 5 0.633 0.002 0.026 0.003 Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115 2 1.138 0.278 0.929 0.192 3 1.062 0.035 0.152 0.035 4 1.372 0.004 0.044 0.008 5 1.639 0.002 0.030 0.003 Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589 2 0.894 0.921 0.792 0.348 3 1.068 0.190 0.248 0.057 4 0.959 0.032 0.217 0.019 5 0.739 0.009 0.101 0.009 a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020 2 1.051 0.007 0.056 0.008 3 0.819 0.002 0.041 0.002 4 1.041 0.003 0.022 0.003 5 0.950 0.003 0.033 0.005 Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133 2 2.341 0.224 0.881 0.132 3 2.996 0.018 0.129 0.022 4 3.813 0.011 0.105 0.016 5 4.573 0.003 0.048 0.003 Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585 2 1.337 1.594 0.705 0.322 3 2.647 0.299 0.470 0.069 4 3.188 0.055 0.326 0.041 5 2.204 0.021 0.199 0.021 Average 1.563 0.215 0.299 0.093 Model Type BudgetResponse Function Type CombinationInteraction
  • 12. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 11 Empirical Tests | Result Parametric DEA FDH IFDH a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009 2 1.040 0.005 0.038 0.006 3 0.571 0.002 0.022 0.001 4 0.694 0.001 0.021 0.002 5 0.633 0.002 0.026 0.003 Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115 2 1.138 0.278 0.929 0.192 3 1.062 0.035 0.152 0.035 4 1.372 0.004 0.044 0.008 5 1.639 0.002 0.030 0.003 Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589 2 0.894 0.921 0.792 0.348 3 1.068 0.190 0.248 0.057 4 0.959 0.032 0.217 0.019 5 0.739 0.009 0.101 0.009 a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020 2 1.051 0.007 0.056 0.008 3 0.819 0.002 0.041 0.002 4 1.041 0.003 0.022 0.003 5 0.950 0.003 0.033 0.005 Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133 2 2.341 0.224 0.881 0.132 3 2.996 0.018 0.129 0.022 4 3.813 0.011 0.105 0.016 5 4.573 0.003 0.048 0.003 Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585 2 1.337 1.594 0.705 0.322 3 2.647 0.299 0.470 0.069 4 3.188 0.055 0.326 0.041 5 2.204 0.021 0.199 0.021 Average 1.563 0.215 0.299 0.093 Model Type BudgetResponse Function Type CombinationInteraction 1 1. The parametric models’ accuracies are lower than those of the non-parametric model. Parametric models is remarkable in the case where interaction exists between the two media. This means that the structure complexity causes difficulties in calibration of parameters in the parametric model
  • 13. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 12 Empirical Tests | Result Parametric DEA FDH IFDH a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009 2 1.040 0.005 0.038 0.006 3 0.571 0.002 0.022 0.001 4 0.694 0.001 0.021 0.002 5 0.633 0.002 0.026 0.003 Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115 2 1.138 0.278 0.929 0.192 3 1.062 0.035 0.152 0.035 4 1.372 0.004 0.044 0.008 5 1.639 0.002 0.030 0.003 Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589 2 0.894 0.921 0.792 0.348 3 1.068 0.190 0.248 0.057 4 0.959 0.032 0.217 0.019 5 0.739 0.009 0.101 0.009 a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020 2 1.051 0.007 0.056 0.008 3 0.819 0.002 0.041 0.002 4 1.041 0.003 0.022 0.003 5 0.950 0.003 0.033 0.005 Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133 2 2.341 0.224 0.881 0.132 3 2.996 0.018 0.129 0.022 4 3.813 0.011 0.105 0.016 5 4.573 0.003 0.048 0.003 Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585 2 1.337 1.594 0.705 0.322 3 2.647 0.299 0.470 0.069 4 3.188 0.055 0.326 0.041 5 2.204 0.021 0.199 0.021 Average 1.563 0.215 0.299 0.093 Model Type BudgetResponse Function Type CombinationInteraction 2 2 2. The DEA model is more accurate than the original FDH. But some case of sigmoid type combination, FDH is more accurate.
  • 14. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 13 Empirical Tests | Result Parametric DEA FDH IFDH a=0 Concave with DRS & Concave with DRS 1 1.528 0.011 0.048 0.009 2 1.040 0.005 0.038 0.006 3 0.571 0.002 0.022 0.001 4 0.694 0.001 0.021 0.002 5 0.633 0.002 0.026 0.003 Concave with DRS & Sigmoid 1 0.989 0.212 0.420 0.115 2 1.138 0.278 0.929 0.192 3 1.062 0.035 0.152 0.035 4 1.372 0.004 0.044 0.008 5 1.639 0.002 0.030 0.003 Sigmoid & Sigmoid 1 0.851 0.985 1.162 0.589 2 0.894 0.921 0.792 0.348 3 1.068 0.190 0.248 0.057 4 0.959 0.032 0.217 0.019 5 0.739 0.009 0.101 0.009 a=1 Concave with DRS & Concave with DRS 1 1.484 0.017 0.064 0.020 2 1.051 0.007 0.056 0.008 3 0.819 0.002 0.041 0.002 4 1.041 0.003 0.022 0.003 5 0.950 0.003 0.033 0.005 Concave with DRS & Sigmoid 1 1.675 0.239 0.469 0.133 2 2.341 0.224 0.881 0.132 3 2.996 0.018 0.129 0.022 4 3.813 0.011 0.105 0.016 5 4.573 0.003 0.048 0.003 Sigmoid & Sigmoid 1 1.600 1.261 1.162 0.585 2 1.337 1.594 0.705 0.322 3 2.647 0.299 0.470 0.069 4 3.188 0.055 0.326 0.041 5 2.204 0.021 0.199 0.021 Average 1.563 0.215 0.299 0.093 Model Type BudgetResponse Function Type CombinationInteraction 3 3. The best model is the IFDH. Therefore, the inferiority of the FDH against the DEA may be caused by the stair-like surface of the frontier.
  • 15. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 14 Additional Test ~ Effectiveness of smoothing technique Improvement DEA (A) FDH (B) Diff1 (A)-(B) DEA (C) FDH (D) Diff2 (C)-(D) Diff2 - Diff1 a=0 Concave with DRS & Concave with DRS 1 0.01 0.05 -0.04 0.00 0.03 -0.03 0.01 2 0.00 0.04 -0.03 0.00 0.02 -0.02 0.01 3 0.00 0.02 -0.02 0.00 0.02 -0.02 0.01 4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01 5 0.00 0.03 -0.02 0.00 0.01 -0.01 0.02 Concave with DRS & Sigmoid 1 0.21 0.42 -0.21 0.12 0.15 -0.03 0.18 2 0.28 0.93 -0.65 0.07 0.32 -0.25 0.40 3 0.03 0.15 -0.12 0.01 0.08 -0.07 0.05 4 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02 5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02 Sigmoid & Sigmoid 1 0.99 1.16 -0.18 0.52 0.66 -0.14 0.04 2 0.92 0.79 0.13 0.79 0.48 0.31 0.18 3 0.19 0.25 -0.06 0.27 0.11 0.16 0.21 4 0.03 0.22 -0.19 0.02 0.06 -0.04 0.15 5 0.01 0.10 -0.09 0.01 0.04 -0.03 0.06 a=1 Concave with DRS & Concave with DRS 1 0.02 0.06 -0.05 0.00 0.05 -0.05 0.00 2 0.01 0.06 -0.05 0.00 0.02 -0.02 0.03 3 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02 4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01 5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02 Concave with DRS & Sigmoid 1 0.24 0.47 -0.23 0.11 0.21 -0.10 0.13 2 0.22 0.88 -0.66 0.06 0.36 -0.30 0.35 3 0.02 0.13 -0.11 0.01 0.08 -0.07 0.04 4 0.01 0.11 -0.09 0.01 0.06 -0.05 0.04 5 0.00 0.05 -0.05 0.00 0.02 -0.02 0.02 Sigmoid & Sigmoid 1 1.26 1.16 0.10 1.39 0.68 0.71 0.61 2 1.59 0.70 0.89 1.53 0.45 1.08 0.19 3 0.30 0.47 -0.17 0.20 0.26 -0.06 0.11 4 0.05 0.33 -0.27 0.03 0.09 -0.07 0.21 5 0.02 0.20 -0.18 0.01 0.06 -0.05 0.13 Average 0.21 0.30 -0.08 0.17 0.15 0.02 0.11 Interaction Response Function Type Combination Budget 200 DMUs 600 DMUs We increased the number of DMUs to reduce the step level differences in the FDH frontier, and investigate the improvement in performance before and after increasing the DMUs
  • 16. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 15 Additional Test ~ Effectiveness of smoothing technique Improvement DEA (A) FDH (B) Diff1 (A)-(B) DEA (C) FDH (D) Diff2 (C)-(D) Diff2 - Diff1 a=0 Concave with DRS & Concave with DRS 1 0.01 0.05 -0.04 0.00 0.03 -0.03 0.01 2 0.00 0.04 -0.03 0.00 0.02 -0.02 0.01 3 0.00 0.02 -0.02 0.00 0.02 -0.02 0.01 4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01 5 0.00 0.03 -0.02 0.00 0.01 -0.01 0.02 Concave with DRS & Sigmoid 1 0.21 0.42 -0.21 0.12 0.15 -0.03 0.18 2 0.28 0.93 -0.65 0.07 0.32 -0.25 0.40 3 0.03 0.15 -0.12 0.01 0.08 -0.07 0.05 4 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02 5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02 Sigmoid & Sigmoid 1 0.99 1.16 -0.18 0.52 0.66 -0.14 0.04 2 0.92 0.79 0.13 0.79 0.48 0.31 0.18 3 0.19 0.25 -0.06 0.27 0.11 0.16 0.21 4 0.03 0.22 -0.19 0.02 0.06 -0.04 0.15 5 0.01 0.10 -0.09 0.01 0.04 -0.03 0.06 a=1 Concave with DRS & Concave with DRS 1 0.02 0.06 -0.05 0.00 0.05 -0.05 0.00 2 0.01 0.06 -0.05 0.00 0.02 -0.02 0.03 3 0.00 0.04 -0.04 0.00 0.02 -0.02 0.02 4 0.00 0.02 -0.02 0.00 0.01 -0.01 0.01 5 0.00 0.03 -0.03 0.00 0.01 -0.01 0.02 Concave with DRS & Sigmoid 1 0.24 0.47 -0.23 0.11 0.21 -0.10 0.13 2 0.22 0.88 -0.66 0.06 0.36 -0.30 0.35 3 0.02 0.13 -0.11 0.01 0.08 -0.07 0.04 4 0.01 0.11 -0.09 0.01 0.06 -0.05 0.04 5 0.00 0.05 -0.05 0.00 0.02 -0.02 0.02 Sigmoid & Sigmoid 1 1.26 1.16 0.10 1.39 0.68 0.71 0.61 2 1.59 0.70 0.89 1.53 0.45 1.08 0.19 3 0.30 0.47 -0.17 0.20 0.26 -0.06 0.11 4 0.05 0.33 -0.27 0.03 0.09 -0.07 0.21 5 0.02 0.20 -0.18 0.01 0.06 -0.05 0.13 Average 0.21 0.30 -0.08 0.17 0.15 0.02 0.11 Interaction Response Function Type Combination Budget 200 DMUs 600 DMUs The results indicate that surface smoothing improves the model performance.
  • 17. Analytics Innovation Company ©BrainPad Inc. Strictly Confidential 16 Conclusion • We propose a non-parametric model, which can be applied to a non-convex shape frontier for MMM • From the result of these tests, our proposed model, IFDH, shows the highest performance from the perspective of minimizing the error • The technique that we have introduced to smooth the FDH frontier surface seems to work well because the DEA model performance is better than the original FDH • To confirm this, we increased the number of DMUs to reduce the step level differences in the FDH frontier, and investigate the improvement in performance before and after increasing the DMUs Some future work: Our model has been proven to work on artificial data, it should be tested in a real-world context. Furthermore, the robustness of the model is still subject to be confirmed.