Upload
kirestin-livingston
View
23
Download
3
Tags:
Embed Size (px)
DESCRIPTION
The Network Nation and Beyond A Festschrift in Honor of Starr Roxanne Hiltz and Murray Turoff. Scenario Construction Via Cross Impact. Prof. Victor A. Bañuls Management Department Pablo de Olavide University Seville, Spain Email: [email protected] Web: http://webdee.upo.es/vabansil. - PowerPoint PPT Presentation
Citation preview
Scenario Construction Via Cross Impact
Prof. Victor A. Bañuls
Management Department
Pablo de Olavide University
Seville, Spain
Email: [email protected]
Web: http://webdee.upo.es/vabansil
The Network Nation and BeyondA Festschrift in Honor of Starr Roxanne Hiltz and Murray Turoff
Distinguished Prof. Murray Turoff
Information Systems Department
New Jersey Institute of Technology
Newark NJ, USA
Email: [email protected]
Web: http://web.njit.edu/~turoff/
NJIT – October 2007
Index
• Research motivations• Methodological background• Basics of the CIM-ISM• Generating scenarios• Conclusions
Research motivations
• Why do we need scenarios?– Strategic decision making (policy, business, etc.)
compromise resources in the long term.– We need to think about what will happen tomorrow
before acting today.– A scenario is a tool for managing the uncertainty of
the future.– Our proposal is aimed at contributing to this goal.
Research motivations
• What is the aim of our proposal?– Helping decision makers to manage the uncertainty.
• How?– Structuring and sharing the beliefs and the knowledge
of the people involved in decision making.
• But… how can we do that?– By the structural analysis of the impacts between the
atomic events that are relevant to the decision-making problem.
Methodological background
• Cross-Impact Method– Events cannot be analyzed in a isolated way.– Alternative cross-impact approach (Turoff, 1972):
Inferring impacts between events based on experts’ hypothesis about their occurrence (or not).
Methodological background
1 2 3 4 5 6 7 8 .. n
1
2
3
4
5
6
7
8
..
n
Gi 1 2 3 4 5 6 7 8 .. n
C43
+/-Impacts between events in
the model
Impacts of the events not included in the model
Cross-Impact Matrix
Methodological background
• Interpretive structural modeling– Taking as an input the impacts obtained with the CIM,
this methodology will help us to:• Making hypotheses about the occurrence or not of
the set of events and analyzed them (to generate scenarios).
• Detecting and analyzing the key drivers (critical events).
Methodological background
3
1
5
2
8 10
6
4
7
9Occurring events Non-Occurring events
Key drivers
Scenario
Methodological background
Pi Sij Rij
Cij Gi
CIM
EiEvents
Set of probabilities (isolated and conditional)
Cross-Impact Matrix
Input
Output
Cross-Impact Method
Methodological background
Pi Sij Rij
Cij Gi
CIM
ISM
Scenarios
EiEvents
Set of probabilities (isolated and conditional)
Cross-Impact Matrix
Cross-Impact Method
Interpretive Structural modeling
Input
Output
Basics of the CIM-ISM
• Starting point– Cross-Impact Matrix (Turoff 1972 paper example).
Basics of the CIM-ISM
1 2 3 4 5 6 7 8 9 10
1 OVP -0.29 0.00 -0.81 -0.33 1.57 0.00 -0.25 -0.22 0.00
2 -0.50 OVP -0.23 0.46 0.00 -0.77 0.90 0.29 0.25 0.42
3 -0.41 0.31 OVP 0.43 0.74 -0.58 0.00 0.27 0.24 0.68
4 -0.81 0.58 0.07 OVP 0.33 -1.21 0.33 0.25 0.22 0.33
5 -0.88 0.58 -0.14 0.81 OVP -0.31 0.74 0.00 0.00 0.36
6 0.88 -0.36 0.00 -2.70 -0.42 OVP -0.38 -0.31 -0.28 -0.38
7 -0.41 0.99 0.00 0.88 1.16 -0.29 OVP 0.00 0.00 0.68
8 -1.62 -0.50 0.00 0.58 0.48 -1.16 0.00 OVP 0.60 0.58
9 -1.49 0.00 0.00 0.93 0.00 -1.07 1.25 1.01 OVP 1.25
10 -0.41 0.99 -0.14 0.88 1.16 -0.58 0.68 0.00 0.00 OVP
Gi 0.23 -1.33 -0.30 -0.05 -1.02 0.88 -0.91 -0.97 -3.29 -0.74
Cross-Impact Matrix
Basics of the CIM-ISM
• Starting point– Cross-Impact Matrix (Turoff 1972 paper example).
• Transforming the Cross-Impact matrix – Transition Matrix (square and positive matrix).
Basics of the CIM-ISM
Occurring events Non occurring events
Occurring events
+ cij - cij
Non occurring
events- cij + cij
Transforming the Cross-Impact Matrix
Basics of the CIM-ISM
• Starting point– Cross-Impact Matrix (Turoff 1972 paper example).
• Transforming the Cross-Impact matrix – Transition Matrix (square and positive matrix).
• Transforming the Transition Matrix– Adjacency Matrix (taking an arbitrary Cij value (0.85)).
Basics of the CIM-ISM
• Starting point– Cross-Impact Matrix (Turoff 1972 paper example).
• Transforming the Cross-Impact matrix – Transition Matrix (square and positive matrix).
• Transforming the Transition Matrix– Adjacency Matrix (taking an arbitrary Cij value (0.85)).
– Connection Matrix (adding the Identity Matrix).
Basics of the CIM-ISM
• Starting point– Cross-Impact Matrix (Turoff 1972 paper example).
• Transforming the Cross-Impact matrix – Transition Matrix (square and positive matrix).
• Transforming the Transition Matrix– Adjacency Matrix (taking an arbitrary Cij value (0.85)).
– Connection Matrix (adding the Identity Matrix).– Reachability Matrix (powering until it is stable).
Basics of the CIM-ISM
• Scenario Generation– Determining antecedent and succedent sets– Obtaining the graphical scenario (using graph theory)
Basics of the CIM-ISM
• Scenario Generation– Determining antecedent and succedent sets.– Obtaining the graphical scenario (using graph theory).
• Interpretation of the scenario– Analyzing key drivers.– Analyzing the set of probabilities.
Basics of the CIM-ISM
LEVEL 1
LEVEL 5
LEVEL 4
LEVEL 2
LEVEL 3
1
5
2
8
10
9
6 4
7
ScenarioOccurring events Non-Occurring events
P9=0.1 Key drivers
Why 0.85?
And event 3?
Generating scenarios
• Sensitivity Analysis– Studying the Cij distribution.
0
1
2
3
4
5
6
7
8
9
10
11
12
13
Percentile 90 1.1581
Percentile 80 0.9198
Percentile 70 0.8109
Percentile 60 0.6450
Percentile 50 0.5389
Percentile 40 0.4132
Percentile 30 0.3409
Percentile 20 0.2950
Percentile 10 0.2508
Generating scenarios
Normal distribution with a reliability of 99% (using K-S test)
LEVEL 2
LEVEL 1
LEVEL 3
1
98
6 4 10
Generating scenarios
Percentile 90 1.1581
Percentile 80 0.9198
Percentile 70 0.8109
Percentile 60 0.6450
Percentile 50 0.5389
Percentile 40 0.4132
Percentile 30 0.3409
Percentile 20 0.2950
Percentile 10 0.2508
LEVEL 2
LEVEL 1
LEVEL 3
LEVEL 4
1
5
8
10
9
7
6 4 2
Generating scenarios
Percentile 90 1.1581
Percentile 80 0.9198
Percentile 70 0.8109
Percentile 60 0.6450
Percentile 50 0.5389
Percentile 40 0.4132
Percentile 30 0.3409
Percentile 20 0.2950
Percentile 10 0.2508
LEVEL 1
LEVEL 5
LEVEL 4
LEVEL 2
LEVEL 3
1
5
2
8
10
9
6 4
7
Generating scenarios
Percentile 90 1.1581
Percentile 80 0.9198
Percentile 70 0.8109
Percentile 60 0.6450
Percentile 50 0.5389
Percentile 40 0.4132
Percentile 30 0.3409
Percentile 20 0.2950
Percentile 10 0.2508
LEVEL 2
LEVEL 1
LEVEL 3
58 10
3
7
6 4 1
2
9
Generating scenarios
Percentile 90 1.1581
Percentile 80 0.9198
Percentile 70 0.8109
Percentile 60 0.6450
Percentile 50 0.5389
Percentile 40 0.4132
Percentile 30 0.3409
Percentile 20 0.2950
Percentile 10 0.2508
Generating scenarios
• Sensitivity Analysis– Studying the Cij distribution
• Solving the forecasted scenario– Determining the limit of the forecasted scenario
LEVEL 2
LEVEL 1
5
8
10
3
76 4 1 2
9
Limit = |0.4975|
Forecasted Scenario
Generating scenarios
-1,-6,2,4,5,7,10 3 8,9
-1,-6,2,4,5,7,10 OPV 0 0
3 1,99 OPV 0
8,9 11,53 0 OPV
G' -9,78 -0,70 -9,56
Cross-Impact Matrix for the Forecasted Scenario
Generating scenarios
Generating scenarios
• Sensitivity Analysis– Studying the Cij distribution.
• Solving the forecasted scenario– Determining the limit of the forecasted scenario.
• Solving the alternative scenarios– Determining the limit of the alternative scenarios.
Generating scenarios
• Sensitivity Analysis– Studying the Cij distribution.
• Solving the forecasted scenario– Determining the limit of the forecasted scenario.
• Solving the alternative scenarios– Determining the limit of the alternative scenarios.
• Interpretation of results– Analyzing the information included in each scenario.
Forecasted Scenario
Alternative Scenario I
Alternative Scenario II
Alternative Scenario III
Limits (||, |0.4975|) (|0.4975|, |0.3804|) (|0.3804|, |0.2318|) (|0.2318|, 0)
Interval of reliability
0.5253 0.1019 0.1400 0.2327
cij sum 35.2527 3.9272 6.1414 1.0224
Event Pi Clusters of Events
1 0.5 A A A A
2 0.3 B B B A
3 0.6 B B B B
4 0.5 B B B B
5 0.4 B A B A
6 0.3 A B A
7 0.6 B B B
8 0.2 B A B
9 0.1 B B B
10 0.6 B B B A
Generating scenarios
Forecasted Scenario
Alternative Scenario I
Alternative Scenario II
Alternative Scenario III
Limits (||, |0.4975|) (|0.4975|, |0.3804|) (|0.3804|, |0.2318|) (|0.2318|, 0)
Interval of reliability
0.5253 0.1019 0.1400 0.2327
cij sum 35.2527 3.9272 6.1414 1.0224
Event Pi Clusters of Events
1 0.5 A A A A
2 0.3 B B B A
3 0.6 B B B B
4 0.5 B B B B
5 0.4 B A B A
6 0.3 A B A
7 0.6 B B B
8 0.2 B A B
9 0.1 B B B
10 0.6 B B B A
Generating scenarios
Forecasted Scenario
Alternative Scenario I
Alternative Scenario II
Alternative Scenario III
Limits (||, |0.4975|) (|0.4975|, |0.3804|) (|0.3804|, |0.2318|) (|0.2318|, 0)
Interval of reliability
0.5253 0.1019 0.1400 0.2327
cij sum 35.2527 3.9272 6.1414 1.0224
Event Pi Clusters of Events
1 0.5 A A A A
2 0.3 B B B A
3 0.6 B B B B
4 0.5 B B B B
5 0.4 B A B A
6 0.3 A B A
7 0.6 B B B
8 0.2 B A B
? 0.1 B B B
10 0.6 B B B A
Generating scenarios
Forecasted Scenario
Alternative Scenario I
Alternative Scenario II
Alternative Scenario III
Limits (||, |0.4975|) (|0.4975|, |0.3804|) (|0.3804|, |0.2318|) (|0.2318|, 0)
Interval of reliability
0.5253 0.1019 0.1400 0.2327
cij sum 35.2527 3.9272 6.1414 1.0224
Event Pi Clusters of Events
1 0.5 A A A A
2 0.3 B B B A
3 0.6 B B B B
4 0.5 B B B B
5 0.4 B A B A
? 0.3 A B A
? 0.6 B B B
? 0.2 B A B
9 0.1 B B B
10 0.6 B B B A
Generating scenarios
Conclusions
• Aims of the model– Handle complex systems.– Obtain a set of plausible snapshots of the future.– Analyze interaction between events.– Detect critical events.
• Application areas– Technology Foresight.– Strategic Management.– Policy Analysis.– Emergency Response.– Etc…
Conclusions
• Strong points– A strong theoretical background of the techniques on which
the authors proposal in based.
– The possibility of working with large sets of events.
– Tools for analyzing the key drivers of the scenarios.
– Specific software is not needed for making the calculations.
– A graphic output that gives a clear representation about the forecast.
– It is strongly compatible with other techniques such as the Delphi or multicriteria methods.
Conclusions
• Limitations– We cannot kwon the probability of occurrence of a
specific scenario if it is not an output of the model. – The estimation of the occurrence or non-occurrence
estimation of the scenarios needs the interpretation of the key drivers and sometimes it would be difficult if there is a probability of occurrence close to 0.5.
Scenario Construction Via Cross Impact
Prof. Victor A. Bañuls
Management Department
Pablo de Olavide University
Seville, Spain
Email: [email protected]
Web: http://webdee.upo.es/vabansil
Distinguished Prof. Murray Turoff
Information Systems Department
New Jersey Institute of Technology
Newark NJ, USA
Email: [email protected]
Web: http://web.njit.edu/~turoff/
Thank you for your attention!