1. Catalog of Operating Plans: The Quest for the Best ESD.71 Lecture Michel-Alexandre Cardin, PhD Candidate Engineering Systems Division December 6 2007

2. © 2007 Michel-Alexandre Cardin ESD.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

3. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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.

4. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 4 Research Questions How do we find the best catalog of operating plans? –W hat 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?

5. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 5 Good Ol’ Parking Garage…

6. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 6 Conceptual Application to Garage I The traditional definition of the design problem for “Garage Case” is:

7. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

8. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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?

9. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

10. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 10 Step 1: Build Initial Model Take deterministic demand projection and price Build cash flow model, get initial value of system

11. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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?

12. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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.

13. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 13 Step 2: Representative Scenarios

14. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

15. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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: 3 3 x 2 3 possibilities: 216 combinations!

16. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

17. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 17 Step 4: Build the Catalog of Op. Plans II Adaptive OFAT algorithm:

18. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

19. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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 1 st experiment

20. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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!

21. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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!

22. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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.

23. December 6 2007© 2007 Michel-Alexandre Cardin ESD.71 23 Step 5: Assess Value of Catalog II Each assignment produces one NPV  represent distribution with VARG curve! Results:

24. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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

25. Questions and Comments?

26. December 6 2007© 2007 Michel-Alexandre Cardin ESD.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.