Upload
florence-millicent-todd
View
212
Download
0
Embed Size (px)
Citation preview
Catalog of Operating Plans:
The Quest for the BestESD.71 Lecture
Michel-Alexandre Cardin, PhD Candidate
Engineering Systems DivisionDecember 6
2007
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
2
Context I• Last week, Prof. de Neufville discussed benefits of catalog of operating plans – Enables consideration of major uncertain scenarios•But avoids exhaustive intractable design analysis
– Encourages deeper investigation of situations with greatest impact on performance•Additional operating plans can easily be added to catalog
– Can be tailored to design problem; catalog can be larger or smaller, focused on specific uncertainties, etc
– Using modern computers, expanding analysis effort factor of 100 or so is easy
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
3
Context II• The approach changes expected value in two ways:– Recognizes value added by manager’s ability to adjust to changing uncertain conditions•Value can be large, should not be ignored
– Adds value through explicit consideration of flexibility in design and operations•Several case studies support this•E.g. satellite systems, mining, real estate development, automotive, etc.
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
4
Research Questions• How do we find the best catalog of operating plans?– What is a good way to define members of catalog?
– How should search be expanded to more members of catalog?
– When should search be terminated?– Others?
• What is the role of flexibility in this approach?
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
5
Good Ol’ Parking Garage…
(Source: SARAA, 2007)
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
6
Conceptual Application to Garage I
•The traditional definition of the design problem for “Garage Case” is:Stage for System
Element Possibilities Traditional Design Practice
Structural Design
Number of Floors
Many Many
Periodic Data on Context Factors
Price, Demand, Quantity, etc
Many, over many periods
One Price, Demand Profile
Periodic Management Adjustments
Price Changes; More Floors
Many, over many periods
None Not considered
Performance Metrics
NPV, ROI, Capex, etc
Many One (the focus)
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
7
Conceptual Application to Garage II
•We expanded the analysis, and looked at:–Many different possible demand scenarios
–Expansions with a simple decision rule
–Value At Risk and Gain (VARG) curve
Stage for System
Element Possibilities “Garage Case Design”
Structural Design
Number of Floors
Many Many
Periodic Data on Context Factors
Price, Demand, Quantity, etc
Many, over many periods
One Price, 1000s of Demand Profiles
Periodic Management Adjustments
Price Changes; More Floors
Many, over many periods
Some Simple decision rule
Performance Metrics
NPV, ROI, Capex, etc
Many Several NPV, VARG
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
8
Conceptual Application to Garage III
• As next step, we can now look at range of decision rules beyond those assumed:– How many years to wait before deciding to expand?
– Some periods where one should not expand?
– How many floors to add on each expansion?
Stage for System
Element Possibilities “Garage Case Design”
Structural Design
Number of Floors
Many Many
Periodic Data on Context Factors
Price, Demand, Quantity, etc
Many, over many periods
One Price, 1000s of Demand Profiles
Periodic Management Adjustments
Price Changes; More Floors
Many, over many periods
CATALOGS of decision rules
Performance Metrics
NPV, ROI, Capex, etc
Many Several NPV, VARG
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
9
How to Find the Best Catalog?• Methodology
– Step 1: build initial model according to traditional approach
– Step 2: find representative uncertain scenarios
– Step 3: look explicitly for sources of flexibility in design and operations•How we “add” value to the system
– Step 4: for each scenario, find the best operating plan•This creates the catalog
– Step 5: assess the value added by the catalog approach•How we “recognize” the value of managerial adjustments
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
10
Step 1: Build Initial Model
• Take deterministic demand projection and price
• Build cash flow model, get initial value of system
Year 0 1 2 3Demand 750 893 1015Capacity 0 1200 1200 1200Revenue $0 $7,500,000 $8,930,000 $10,150,000Operating costs $0 $2,400,000 $2,400,000 $2,400,000Land leasing and fixed costs $3,600,000 $3,600,000 $3,600,000 $3,600,000Cashflow $0 $1,500,000 $2,930,000 $4,150,000DCF $1,339,286 $2,335,778 $2,953,888Present value of cashflow $32,574,737Capacity cost for up to two levels $6,400,000Capacity costs for levels above 2 $16,336,320Net present value $6,238,417
0
200
400
600
800
1000
1200
1400
1600
1800
1 3 5 7 9 11 13 15 17 19
Time (years)
Demand (spaces)
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
11
Step 2: Find Representative Scenarios I• Determine sources of uncertainty (e.g. demand, price)
• Incorporate fluctuations around deterministic projection
• Produce a few demand scenarios (10 to 20) and look at representative properties. Any idea?
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
12
Step 2: Find Representative Scenarios
II• Take demand growth between years 1-5 as criterion– Create five representative scenarios differentiated by early growth level
• How to differentiate the categories?– Use the mid-value between two categories– E.g. simulated scenario with growth above 123% is similar to scenario 1, between 100%-123% scenario 2, etc.
Demand scenario Percentage increase Mid-valuecategory from first to fifth year
1 131% 123%2 115% 100%3 84% 68%4 52% 38%5 24%
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
13
Step 2: Representative Scenarios
600
800
1000
1200
1400
1600
1800
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (years)
Demand (spaces)
Original demand projection Demand scenario 1 Demand scenario 2
Demand scenario 3 Demand scenario 4 Demand scenario 5
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
14
Step 3: Find Sources of Flexibility I
• Demand is uncertain, how can we adapt to it?– Reduce losses: build fewer floors initially, reduce initial CAPX
– Increase profits: expand as demand increases
– Other sources of flexibility?• Every system is different. Not obvious where to find flexibility!– Brainstorm, experts’ opinions, etc.– See work by Bartolomei (2007), Kalligeros (2006), and Wang (2005) for structured methodologies
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
15
Step 3: Find Sources of Flexibility II
• Many ways to exploit the flexibility to expand, in design and operations
– “Levels” correspond to specific choice of design element of management decision rule
– Note: 33 x 23 possibilities: 216 combinations!
Design Elements and DescriptionManagement Decision Rules
A Expansion allowed in years 1-4 No YesB Expansion allowed in years 9-12 No YesC Expansion allowed in years 17-20 No YesD Expansion decision rule (years) 2 3 4E Number of floors expanded by 1 2 3F Number of initial floors 4 5 6
Levels
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
16
Step 4: Build the Catalog of Op. Plans I
• Introducing adaptive One Factor At a Time (OFAT) algorithm (Frey and Wang, 2006)– Used for design of statistical experiments– Applied to design of real engineering systems to effectively search best design combinations
– Provides a shortcut to full factorial analysis– Cost-effective way to explore the space of possibilities
• Our method is inspired from adaptive OFAT…– We do not perform statistical experiments while exploring the space of possible combinations
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
17
Step 4: Build the Catalog of Op. Plans II
• Adaptive OFAT algorithm:
(Source: Frey and Wang, 2006)
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
18
Step 4: Setup the Search I• Pick one representative scenario (e.g. scenario 1)
• Choose one combination of design elements and management decision rules Baseline experiment– This is first “experiment” where you measure value of a particular design, under a particular uncertain scenario, with a particular set of management decision rules
• Choose an OFAT sequence arbitrarily– Determine in what sequence you will explore the design elements and management decision rules
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
19
Step 4: Setup the Search II
• Example:
– Management DR: management decision rules (represented here by letters A to E in OFAT sequence)
– DE: design elements (represented here by letter F in OFAT sequence)
– Baseline experiment: set of design elements and management decision rules chosen for 1st experiment
DEs and Management DRs Description Baseline ExperimentOFAT Sequence
A Expansion allowed in years 1-4 Yes FB Expansion allowed in years 9-12 Yes CC Expansion allowed in years 17-20 Yes ED Expansion decision rule (years) 2 DE Number of floors expanded by 1 BF Number of initial floors 5 A
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
20
Step 4: Explore the Possibilities
• Measure NPV Baseline value• Change one “level” in the combination:
– If NPV is higher, keep change; if lower, go back to previous state
• Explore all levels at least once, keep best combination
• Notice: only 10 combinations explored instead of 216!
ExperimentDE and Management DR changedLevel changed to:Output = NPVBest output before stepKeep change?1 (baseline) $ 13.42 F 4 $ 10.9 $ 13.4 No3 F 6 $ 15.1 $ 13.4 Yes4 C No $ 15.1 $ 15.1 No5 E 2 $ 15.8 $ 15.1 Yes6 E 3 $ 15.7 $ 15.8 No7 D 3 $ 14.6 $ 15.8 No8 D 4 $ 13.5 $ 15.8 No9 B No $ 15.8 $ 15.8 No10 A No $ 13.5 $ 15.8 No
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
21
Step 4: Get a Catalog of Operating Plans
• Repeat the same procedure for remaining 4 representative demand scenarios
• Get one operating plan best suited for each representative scenario– Now have a catalog of operating plans!
DEs and Management DRs Description Op. Plan 1Op. Plan 2Op. Plan 3Op. Plan 4Op. Plan 5A Expansion allowed in years 1-4Yes Yes Yes Yes YesB Expansion allowed in years 9-12Yes Yes No Yes YesC Expansion allowed in years 17-20Yes Yes Yes Yes YesD Expansion decision rule (years) 2 2 2 2 4E Number of floors expanded by 2 3 3 1 1F Number of initial floors 6 5 5 4 4
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
22
Step 5: Assess Value of Catalog I
• Simulate operator’s ability to choose an operating plan depending on demand scenario (2,000 scenarios, yes…)
• Recall, simulated scenario categorized using mid-value between categories; then assign associated operating plan– E.g. scenario with growth between years 1-5 above 123% is given operating plan 1, between 100%-123% operating plan 2, etc.
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
1 2 3 4 5
Operating Plan Selected
Demand scenario Percentage increase Mid-valuecategory from first to fifth year
1 131% 123%2 115% 100%3 84% 68%4 52% 38%5 24%
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
23
Step 5: Assess Value of Catalog II
• Each assignment produces one NPV represent distribution with VARG curve! Results:
Inflexible Design Flexible Design with Which is Better?Catalog of Operating Plans
Expected initial investment $ 18.1 $ 16.3 Flex. and Catalog BetterExpected NPV $ 2.9 $ 4.9 Flex. and Catalog BetterExpected NPV minus expected cost of flexibility$ 2.9 $ 4.2 Flex. and Catalog BetterMinimum NPV $ -19.5 $ -18.8 Flex. and Catalog BetterMaximum NPV $ 8.3 $ 20.5 Flex. and Catalog BetterValue of catalog of flexible operating plans $ 0.0 $ 1.3
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
-30.0 -20.0 -10.0 0.0 10.0 20.0 30.0Millions
NPV ($)
VARG with Catalog ENPV with Catalog VARG Inflexible ENPV Inflexible
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
-30.0 -20.0 -10.0 0.0 10.0 20.0 30.0
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
24
Summary• Methodology improves current practice significantly, which is simplistic as regards to exogenous factors affecting value
• It is not exhaustive! It does not use a different plan for each simulation. This would:– Take far too long– Be very expensive
• Method uses a “catalog” of operating plans prepared ahead of analysis. These are designed to be “representative”
• It recognizes value from operational adjustments, and adds value through use of flexibility in design and operations
Questions and Comments?
December 6 2007
© 2007 Michel-Alexandre CardinESD.71
26
References• Bartolomei, J. (2007), “Qualitative Knowledge Construction for Engineering Systems:
Extending the Design Structure Matrix Methodology in Scope and Procedure”, Doctoral Dissertation, Engineering Systems Division, Massachusetts Institute of Technology, Cambridge, MA.
• Cardin, M-A. (2007), “Facing Reality: Design and Management of Flexible Engineering Systems”, Master of Science Thesis, Engineering Systems Division, Massachusetts Institute of Technology, Cambridge, MA.
• Frey, D., Wang, H. (2006), “Adaptive One-Factor-at-a-Time Experimentation and Expected Value of Improvement”, Technometrics, 48, 3, pp. 418-431.
• Kalligeros, K. (2006), “Platforms and Real Options in Large-Scale Engineering Systems”, Doctoral Dissertation, Engineering Systems Division, Massachusetts Institute of Technology, Cambridge, MA.
• SARAA (2007), “Harnsburg International Airport – HIA at a Glance”, SARAA, http://www.flyhia.com/images/inset_parking-garage.jpg, accessed on March 2007.
• Wang, T. (2005), “Real Options "in" Projects and Systems Design - Identification of Options and Solutions for Path Dependency”, Doctoral Dissertation, Engineering Systems Division, Massachusetts Institute of Technology, Cambridge, MA.