Chapter 11– Structured Risk Management Risk Management Expected Value of Perfect Control Sensitivity Analysis – Robustness – easy to do with Precision Tree Value Added structured approach - Individual random events Role of Information Expected Value of Imperfect Information Bayes Rule and EVII Optimal Conditional decision Sequential decisions with information delay Real Options Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 1 Risk Management Theme You cannot manage risk if you do not admit there is uncertainty Managing uncertainty also includes unrealized upside potential and not just downside losses You cannot allocate appropriate resources if you do not quantify the risk or uncertainty Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 2 Figure 11.1: Decision tree Boss Controls automation investment 40.0% 9.9 0 1.9 50% Take 60.0% 16.5 0 8.5 30% Take 40.0% 13.8 0.4 0.8 60.0% 0.6 10 30% Take FALSE Low Automation Investment -8 Take Rate 5.86 How Much 6.32 High TRUE -13 Take Rate 6.32 50% Take Chapter 11 Chelst & Canbolat Value Added Decision Making 23 02/28/12 3 Investment in Automation Question: Robustness of Optimal Solution The “High” investment alternative involves a new technology. Management is concerned that the capital equipment estimate could be off by + 7%. There is even more concern regarding the variable cost estimate that could be off by + 10% The Low investment alternative is well tested and there is hope that continuous improvement could reduce the variable cost by 5%. Because they did not know, they set the take rate probabilities at 0.6 and 0.4 respectively. However, there is a lot of uncertainty regarding this estimated probability. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 4 Investment in Automation Robustness of Optimal Solution Magnitude of Difference between two solutions ($6.32M-$5.86M) = $460,000 Investment(s) How much increase in HIGH Investment fixed cost results in change in best decision? Variable Cost(s) How much would the variable cost for Low Investment have to decline to make it preferred? Probability of 30% take rate: Increases? Decreases? What else and why? Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 5 Activate: Precision Tree & Sensitivity Analysis Output – Separate Worksheets Sensitivity – one parameter at a time One line –Objective function for optimal strategy: A change in optimal decision is usually bend in line Multiple lines – Objective function for each decision. Crossing lines change in optimal decision Chapter 11 Tornado diagram – more variables but less info Spider Plot – more variables, more info, but limited to no more than 3 or 4 variables – too cluttered and confusing Chelst & Canbolat Value Added Decision Making 02/28/12 6 Review: Figure 10.23: Sensitivity analysis automation investment – fixed cost of high investment $13.48 million 7.5 Expected Value 7 6.5 High Low 6 5.5 -$12.0 -$12.2 -$12.4 -$12.6 -$12.8 -$13.0 -$13.2 -$13.4 -$13.6 -$13.8 -$14.0 5 High Investment X axis – Fixed cost input as negative value (-13) Axis would be reversed if cost was stored as (13) Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 7 Review: Figure 10.24: Expected value of the optimal decision for each value of fixed cost of high investment 7.5 Expected Value 7 6.5 6 5.5 -$12.0 -$12.2 -$12.4 -$12.6 -$12.8 -$13.0 -$13.2 -$13.4 -$13.6 -$13.8 -$14.0 5 High Investment Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 8 Review: Figure 10.25: Sensitivity analysis automation investment – low take rate probability 8.5 8 Decision changes when probability approaches 0.6 (a 50% increase) Expected Value 7.5 7 6.5 6 High Low 5.5 5 4.5 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 4 Low Probability Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 9 List of Variable Ranges A. B. C. D. E. F. G. Chapter 11 Fixed investment: High Investment: ±7% of base Price: 0 to –10% of base Variable Cost of Low investment: ± 10 % of base Variable Cost of High investment: 0 to 5 % of base Probability of Low Take rate: ± 0.2 absolute Low take rate (30%): 0 to –10% absolute Volume: 0 to – 15% of base Chelst & Canbolat Value Added Decision Making 02/28/12 10 Precision Tree Sensitivity Analysis Tornado Diagram Many parameters: unlimited Chapter 11 Uses Min & Max values specified in the range and calculates Objective function. Ranks the analysis in order of their range of impact on the objective looks like tornado Does NOT show changed decisions!! Chelst & Canbolat Value Added Decision Making 02/28/12 11 Figure 11.2: Tornado diagram Boss Controls automation investment Tornado Graph of Decision Tree 'Automation Investment' Range of parameter Prob of Low Take (0.2 to 0.6) Prob. (D13) Vehicles (850 K to 1 million) Vehicles (Mil.) (C10) Price ($54 to $60) Price (C4) Take Rate (C13) Low take rate (20% to 30%) High_Investment (D6) High Invest. ($13 m + 910K) Low_Invest_VC (C7) 8.5 8 7.5 7 6.5 6 5.5 5 4.5 4 3.5 High_Invest_VC (D7) Expected Value Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 12 Precision Tree Sensitivity Analysis Spider Diagram practical limit of 4 parameters More detailed than Tornado but harder to include many variables. X axis – change input (percent) Y axis – change in expected value Chapter 11 Aggregation of many one-way sensitivity analyses but scaled to a common percentage. Shows the slope of the impact on the objective function and non-linearities. Shows changes in decisions bends in line graph Chelst & Canbolat Value Added Decision Making 02/28/12 13 List of Variable Ranges Fixed investment: High Investment: ±7% of base Price: 0 to –10% of base one sided (lower value) Probability of Low Take rate: ± 0.2 absolute Decision does not change except at the very highest value – slight bend in line at end Volume: 0 to – 15% of base one sided (lower value) Chapter 11 Range of Change in input % from a negative % to 0% Range of Change in input % from a negative % to 0% Chelst & Canbolat Value Added Decision Making 02/28/12 14 Figure 11.3: Spider plot for Boss Controls automation investment Spider Graph of Decision Tree 'Automation Investment' Expected Value of Entire Model 8.5 Decision changes: bend Expected Value 8 7.5 7 6.5 Prob. (D13) Vehicles (Mil.) (C10) Price (C4) High_Investment (D6) 6 5.5 5 4.5 4 3.5 -60% -40% -20% 0% 20% 40% 60% Change in Input (%) Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 15 Manage Risk Impact of Strategies to Change Risk Profile Shift the risk profile to the right Figure 11-4b. add value to all possible outcomes eliminate altogether an operating cost in a project. Cut off the downside risk Figure 11-4c Move outcomes to some guaranteed level. Minimum purchase quantity in a contract increase the mean and remove the most disastrous possibilities. Insurance cuts off the downside risk (costs money) leftward shift in the whole risk profile but reduce the overall expected value Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 16 Figure 11.4: Impact of risk management actions on risk profile Figure a: Baseline Figure b: Shift to right by adding net value (cost elimination) Chapter 11 F Figure c: Chop off left eliminate downside risk Chelst & Canbolat Value Added Decision Making 02/28/12 17 Change Risk Profile – Manage Risk Centrally concentrate uncertainty: Figure 11-4d Risk sharing: sell half of a risky opportunity for a price equal to half of its expected value Reduce but not eliminate extremely negative outcomes: Figure 11-4e Magnitude reduction” consistent with the way managers view risk Probability reduction not as well understood Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 18 Figure 11.4: Impact of risk management actions on risk profile Figure a: Baseline Figure d: Centralize through Figure e: reduce magnitude of risk sharing negative outcome Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 19 Make or Buy Decision: Non-strategic (strictly cost) Decision Context: Manufacture a component yourself or contract with a supplier to manufacture it. There is a design for a component but you are not sure when it comes time to manufacture, that the design will be feasible as is. If not, there will need to be a quick major redesign of the component. If you manufacture it, you expect that with the redesign it will cost 8% more than the original estimate. The decision to make or buy must be made now before you have time to fully check out the design. The demand for the product is also uncertain. If you sign a contract with the supplier for a specific piece price, if the current design turns out to be infeasible, you know the supplier will use the design change as an excuse to increase the price 15%. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 20 Make or Buy Data: Random Events Random Events 1. Design Feasibility Current Design will Work Need a Major Redesign 2. Demand Low Medium High Chapter 11 Volume 1.0 million 1.25 million 1.5 million Chelst & Canbolat Value Added Decision Making Prob. 0.4 0.6 Prob. 0.3 0.5 0.2 02/28/12 21 Make or Buy Data: Cost Data Costs: Make In-House Facility investment fixed Cost $55M Variable Cost/ per part If current design works $100/part If new Design is needed $108/part Costs: Buy from Supplier Facility investment fixed Cost $0 Variable Cost/ per part If current design works $140/part If new Design is needed $161/part Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 22 Design Feasibility Works Does NOT Demand 1 1.25 1.5 Probability 0.4 0.6 Premium Make Costs 100 108 8% Prob. 0.3 0.5 0.2 Complex calculation & NOT sum of values on branches Buy Costs 140 161 15% 30.0% 1 0.12 155 50.0% 1.25 20.0% 1.5 0.2 180 0.08 205 30.0% 1 0.18 163 Demand 187.3 50.0% Medium 1.25 20.0% High 1.5 0.3 190 0.12 217 30.0% 1 0 140 50.0% 1.25 20.0% 1.5 0 175 30.0% 1 0 161 Low Works E(X) 40.0% Demand 177.5 100 Medium Fixed Costs Make 55 Buy 0 High Make TRUE 55 Current Design 183.38 Low Minimize Cost Does NOT work 60.0% 108 Decision Make or Buy 183.38 Low Works 40.0% Demand 171.5 140 Medium High Buy Figure 11.8: Western Co. make or buy decision Chapter 11 E(X) FALSE 0 0 210 Current Design 186.935 60.0% Low Demand 197.225 Does NOT work 161 Medium Chelst & Canbolat Value Added Decision Making High 0 50.0% 201.25 1.25 20.0% 0 1.5 02/28/12241.5 23 Structured Risk Management Step: Summary Within optimal decision Identify random paths with large downside risk Large values that are negative or poor relative to the best paths Probability associated with this sequence is not insignificant Assess impact of Increasing relative value of that path Decreasing the probability of that path Brainstorm strategies for making the above happen Quantify these alternatives Chapter 11 Repeat for 2nd best decision Chelst & Canbolat Value Added Decision Making 02/28/12 24 Summary of Risk Management Alternatives: Table 11.4 Factor Reduce Cost Increase Linked to Redesign Optimal Comments $730,000 If redesign is needed try $3.65 M to contain added cost of manufacturing. Reduce Risk that From 0.6 to 0.5 $980,000 Modify design quickly Design will not Work to reduce need for major redesign later. From 0.6 to 0.3 $4.16 M New Optimal: Use Supplier From 0.6 to 0.0 $11.9 M Value of Perfect Control Manage Uncertainty of Not Does not make sense to Demand appropriate reduce total demand to lower total cost. Chapter 11 Change From $8 to $7 From $8 to $3 Chelst & Canbolat Value Added Decision Making 02/28/12 25 Summary of Risk Management Alternatives Table 11.4 Continued Factor Percentage Price increase by Supplier if design does not work Supplier Price Reduction if Volumes are High EVPI of Design Feasibility Chapter 11 Comments Optimal Change Obtain commitment $0 From 15% to 14% $3.5 M from supplier not to From 15% to 8% take advantage of redesign to raise prices disproportionately. Negotiate major price No Up to $8 Impact reduction for high reduction in price volumes. $2.4 M Test feasibility of current design Chelst & Canbolat Value Added Decision Making 02/28/12 26 New topic: Information Value Perfect Information Imperfect Information Sample Information Expert Information Accuracy of test (medical or engineering) Delay Chapter 11 decision until information unfolds Options Chelst & Canbolat Value Added Decision Making 02/28/12 27 Information Gathering Traditional Approach – Gather information (surveys, tests, pilot plant, prototypes) until time or the budget runs out. Most information is gathered to validate already made decision. New Approach - Gather information if the cost of gathering it is less than the gain in expected value. Process – Restructure the decision tree to determine the expected value with the information Counterintuitive – How can you determine the value of information before you have even gathered the information? Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 28 Expected Value of Perfect Information: EVPI Goal: Determine the expected value of perfect information regarding an Uncertainty or Risk – Hire a Clairvoyant – Prophet Isaiah (Thomas) This provides an upper bound on the value of all information including “imperfect” information. If the information never changes the optimal decision then EVPI = 0. Decision Tree Process: Move the random event in question to the front of the tree “before” the first decision is to be made. Recalculate the overall expected value. The NET Improvement is the EVPI. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 29 Original Decision Tree – Automation Investment Boss Controls – Base Tree 50% Take Rate High TRUE -13 Automation Investment Decision 6.32 (MAX{6.32, 5.86}) 23 Low -8 Chapter 11 0.6 10.0 Take Rate 6.32 (=10*0.6 + 0.8*0.4) 40.0% 0.4 30% Take Rate 0.8 13.8 50% Take Rate FALSE 60.0% 60.0% 16.5 0 8.5 Take Rate 5.86 (=8.5*0.6 + 1.9*0.4) 0 30% Take Rate 40.0% 1.9 9.9 Chelst & Canbolat Value Added Decision Making 02/28/12 30 Figure 11.7: EVPI tree for Boss Controls Investment Take rate event moved to before decision TRUE 1.9 0.4 1.9 How Much 1.9 FALSE High 0.8 0 0.8 Low EVPI = 6.76 - 6.32 = 0.44 Perfect Information 30% Take 40.0% Take Rate 6.76 FALSE 8.5 How Much 10 TRUE High 10 Low 50% Take 60.0% 0 8.5 0.6 10 Optimal decision depends on outcome of random event Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 31 Expected Value of Perfect Information: Boss Base strategy – High Investment & E(X) = $6.32M If information indicates 30% take rate then “shift” to Low Investment with profit = $1.9M If information indicates 50% take rate then stay with High Investment with profit = $10M What is the probability the information will indicate a 30% take rate? Answer 0.4 E(X) with perfect information = 1.9(.4) + 10 (.6) = 6.76 EVPI = 6.76 – 6.32 = $0.44M E(Perfect Control) = 10 – 6.32 = $3.86 M much more valuable to exert control over uncertainty Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 32 Review to contrast EVPC with EVPI Maximum value of risk management Expected Value of Perfect Control: Not about obtaining information but rather exerting control over destiny Goal: Determine the value of eliminating Uncertainty or Risk This provides an upper bound on the value of risk management with regard to that uncertainty. Process: Assign probability of “1” to the best outcome of an uncertain event. Recalculate the overall expected value. The NET Improvement in expected value is the EVPC. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 33 Review: Expected Value of Perfect Control: Automation Investment Assign probability of “1” to best outcome Net Change: $10 – 6.32 = $3.68 million 100% 50% High Automation Investment TRUE -13 Decision 10 (MAX(10, 8.5)) Take Rate 10 =10*1 + 0.8*0) 0% 30% 13.8 50% Low FALSE -8 Chapter 11 Chelst & Canbolat Value Added Decision Making 23 100% 16.5 1.0 10.0 0 0.8 0 8.5 Take Rate 8.5=8.5*1.0 + 1.9*0) 0% 0 30% 1.9 9.9 02/28/12 34 2nd example: EVPI – Western Make or Buy Base strategy Make: E(X) = $183.38M Uncertainties Design works or not Bound on Testing Design (Imperfect) EVPI = $2.41 M Demand Bound on value of surveys (Imperfect) EVPI = $2.16 M Both uncertainties EVPI = $3.16 M Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 35 Prob. Works 0.4 Does NOT 0.6 Premium Design Feasibility Make Costs 100 108 8% Low Buy Costs 140 161 15% 30.0% 1 Works 40.0% Demand 177.5 100 Medium 50.0% High 20.0% 1.25 Demand 1 1.25 1.5 Prob. 0.3 0.5 0.2 Make TRUE 55 1.5 Current Design 183.38 Low 30.0% 1 Fixed Costs Make 55 Buy 0 Does NOT work 60.0% 108 0.2 180 0.08 205 0.18 163 Demand 187.3 Medium 50.0% High 20.0% 1.25 Decision 0.12 155 1.5 0.3 190 0.12 217 Make or Buy 183.38 Low 30.0% 1 Works 40.0% Demand 171.5 140 Medium 50.0% 1.25 High Make or Buy Buy FALSE 0 20.0% 1.5 Current Design 186.935 Low 30.0% 1 Does NOT work 60.0% 161 0 140 0 175 0 210 0 161 Demand 197.225 Medium 50.0% High 20.0% 1.25 1.5 0 201.25 0 241.5 Low Figure 11.9: Make-Buy EVPI: Design Feasibility Net Improvement Make FALSE Works 40.0% 183.38-180.98 = $2.40M High Low Buy TRUE 0 180 0 205 0.12 30.0% 140 140 Demand 171.5 High Current Design 0.2 50.0% 175 20.0% 210 175 0.08 210 180.98 Low Make Design uncertainty resolved before decision TRUE 0 Does NOT work 60.0% 0 0.18 30.0% 163 163 Demand 187.3 Medium Decision High 0.3 50.0% 190 20.0% 217 190 0.12 217 187.3 Low Buy FALSE 0 High Chelst & Canbolat Value Added Decision Making 0 30.0% 161 161 Demand 197.225 Medium Chapter 11 0 50.0% 180 20.0% 205 171.5 Medium Info Design 155 Demand 177.5 Medium Decision 0 30.0% 155 50.0% 201.25 20.0% 241.5 0 201.25 0 241.5 02/28/12 37 40.0% 155 Current Design 159.8 60.0% Does NOT work 163 0 155 40.0% 140 Current Design 152.6 60.0% Does NOT work 161 0.12 140 40.0% 180 Current Design 186 60.0% Does NOT work 190 0.2 180 Works Figure 11.10: Make-Buy EVPI on Demand Net Improvement 183.38-181.22 = $2.16M FALSE Make 0 Low 30.0% 0 Decision 152.6 Works Buy Info Demand TRUE 0 Demand 181.22 Works Make Medium 50.0% 0 TRUE 0 Decision 186 0 163 0.18 161 0.3 190 0 40.0% 175 175 Current Design 190.75 0 60.0% 201.25 Does NOT work 201 40.0% 0.08 Works 205 205 Current Design 212.2 60.0% 0.12 Does NOT work 217 217 Works FALSE Buy 0 Demand uncertainty resolved before decision Make High 20.0% 0 TRUE 0 Decision 212.2 0 40.0% 210 210 Current Design 228.9 0 60.0% Does NOT work 242 241.5 Works FALSE Buy 0 Make Figure 11.11: Make-Buy Decision EVPI on Feasibility & Demand Combined Net Improvement 183.38-180.22 = $3.16M Less than the SUM of $2.16 (Demand EVPI) + $2.40 (Feasibility EVPI) INFO on BOTH events 155 Low 30.0% 140 Buy Works 40.0% 0 Demand 170.5 TRUE 140 Make FALSE 180 Medium 50.0% Decision 0 175 Buy TRUE 175 Make TRUE 205 High 20.0% FALSE Current Design 180.22 30.0% 210 FALSE 163 161 Buy Demand 186.7 TRUE 161 Make TRUE 190 Medium 50.0% 190 Buy FALSE 201.25 Make TRUE 217 20.0% 217 Buy FALSE 241.5 Chelst & Canbolat Value Added Decision Making 0 210 0 163 0.18 161 0.3 190 0 201.25 0.12 217 Decision 0 Chapter 11 0.08 205 Decision 0 High 0.2 175 Decision 0 0 0 180 205 Buy Does NOT work 60.0% 0.12 140 Decision 0 Low 0 155 Decision 0 Make Next slide: Schematic Trees FALSE 0 241.5 02/28/12 39 Original Make or Buy Schematic Trees: EVPI Make or Buy Design EVPI Demand = $2.4 M Demand Make or Buy Design EVPI Design = $2.16M Design Design Chapter 11 Demand Make or Buy Demand Demand Make or Buy Chelst & Canbolat Value Added Decision Making EVPI Combined: Design & Demand =$3.16M 02/28/12 40 Imperfect Information Conditional Decision/ Probabilities P(Positive) Invest P (High | Positive) Downstream values and/or probabilities are affected by an upstream random event Decision made AFTER resolution of random event Optimal decision path differs depending upon the outcome of a random event Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 41 Expected Value of Imperfect Information Imperfect info partial resolution of uncertainty Test or Sample Information Few tests, experiments or surveys are perfect. EVPI is an upper bound on the value of imperfect information. EVII – without well documented test reliability: Conditional probabilities based on judgment EVII with Bayes Rule is used primarily in environments with extensive data on the reliability of tests – both false positives and false negatives. Oil industry – Seismographic data. Test wells Medical Applications Weather forecasts Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 42 EVII – without well documented test reliability Conditional probabilities based on judgment Expert understands the uncertain relationship between test data (performance, throughput, etc.) or market surveys and subsequent outcome. Can the expert provide a probabilistic range of outcomes that have accompanied similar test results? Understand concept of conditional probability – experience with both possible outcomes. Need stable process environment – A priori probabilities are always in a narrow range, for example, of 0.40 to 0.60. Not used to forecast rare events Problem – people have invalid intuition. Cannot factor in “a priori” estimates that are updated with imperfect information. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 43 Boss Controls: Focus Groups & Imperfect Information based on Experience Boss Controls (BC) is gearing up to manufacture an option to be made available on 1 million new cars world-wide. Initial estimates are that the take rate for the option could be as low as 30% or as high as 50%. Assume for simplicity sake, these two take rates are equally likely. Experience with focus groups indicates that for options such as the one BC is considering, the results will either be Enthusiastic (E) or Good (G). In the past if the focus groups were Enthusiastic, the take rate ended up being at the HIGH end 70% time. However, if the focus groups’ reactions were just good, then 80% of the time the take rate was at the LOW end. Focus groups have an optimistic bias and tend to be enthusiastic 80% of the time. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 44 EVII & Decision Trees - Experience Chapter 11 Add an uncertain node at the front of tree to represent uncertain outcome of focus group Insert the probabilities that reflect the likelihood of different responses: Here P(E) = .8 and P(G) = .2 Probability of outcomes (Take rates) are now “Conditional” probabilities based on past experience (or Bayes Rule) Insert the “conditional” probabilities into tree and calculate expected value. Chelst & Canbolat Value Added Decision Making 02/28/12 45 Figure 11.14: Decision tree of EVII for BC automation investment Expert estimates conditional probabilities 30% Take FALSE Low -8 80.0% 6.52 Imperfect Info TRUE High 0 16.5 8.5 30.0% 0.24 13.8 0.8 -13 Conditional Probabilities 70.0% 0.56 Take Rate 7.24 50% Take 23 10 6.436 30% Take TRUE Low -8 Good 20.0% 1.9 20.0% 0.04 16.5 8.5 3.22 3.22 30% Take High FALSE -13 80.0% 13.8 0 0.8 Take Rate 2.64 50% Take Chelst & Canbolat Value Added Decision Making 0.16 9.9 How Much 0 EVII = 6.436 – 6.320 = 0.116 Less than one third of EVPI was $400,000 80.0% Take Rate 50% Take Chapter 11 70.0% 7.24 30% Take Focus Group 0 1.9 How Much 0 EVII = 6.436 - 6.320= 0.116 9.9 Take Rate 50% Take Enthusiastic 30.0% 02/28/12 20.0% 0 23 10 46 Conditional Probabilities Consistent with Original Estimates A Priori Probability that Take Rate is 30% - Use Partition Formula P(A) = P(A|B)P(B) + P(A|B)P(B) Chapter 11 P(T=30%) = P(T=30% | G) P(G) + P(T=30% | E) P(E) P(T=30%) = (3/4)(.4) + (1/3) (.6) = .5 “original estimate” P(T=50%) = P(T=50% | G) P(G) + P(T=50% | E) P(E) P(T=50%) = (1/4)(.4) + (2/3) (.6) = .5 “original estimate” Chelst & Canbolat Value Added Decision Making 02/28/12 47 Boss Control: Conditional Decision If focus group’s reaction is ENTHUSIASTIC then HIGH investment in automation If focus group’s reaction is GOOD then Low investment in automation Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 48 INTUITION? Bayes’ Rule & Reliable Test Rare Disease – How Rare: 1 in 1,000 Probability of positive reading for a person with the disease – test is very reliable P(Pos.| Disease) = P(P|D) = .99 Probability of negative reading for a person without the disease – 4% false positives P(Neg. | No Disease) = P(N|Dc) = .96 Key Question: P(Disease | Pos) = P(D|P) = ?? Let Dc = D complement, or D , or No disease Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 49 Bayes’ Rule & Reliable Test - Results Bayes’ Rule (General Formula): P( B | A) P( A | B) * P( B) P( A | B) * P( B) P( A | B c ) * P( B c ) with Bc = B complement or NOT B Denominator uses partitioning (all ways that A can occur) to determine P(A) Bayes’ Rule (Reliable Test): (Pos = Positive test result) P( D | Pos) Chapter 11 P( Pos | D) * P( D) P( Pos | D) * P( D) P( Pos | D c ) * P( D c ) 0.99 * 1 1000 1 999 0.99 * (1 0.96) * 1000 1000 0.00099 0.024 0.04095 or 1 in 41 Intuitive 1000 tested yields 40 false positives (4% error rate) and 1 true positive Chelst & Canbolat Value Added Decision Making 02/28/12 50 Probability Misunderstanding People do NOT know how to integrate prior knowledge and data accuracy. Especially problematic with Low probability events and highly accurate test Weakly reliable tests Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 51 Bayesian Posterior (after positive result) Probabilities Test Accuracy Assume Positive = Negative Initial Probability of Success .7 .8 .9 .95 .1 0.21 0.31 0.50 0.68 .3 0.50 0.63 0.79 0.89 .4 0.61 0.73 0.86 0.93 .45 0.66 0.77 0.88 0.94 .5 0.70 0.80 0.90 0.95 .6 0.78 0.86 0.93 0.97 .7 0.84 0.90 0.95 0.98 For 0.5, .45, and even 0.40, the final estimates are close to test accuracy. Column heading close to cell value. For initial low probability events, test accuracy and final probability are far apart. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 52 EVII: Make or Buy Decision Decision Context: Manufacture a component yourself or contract with a supplier to manufacture it. Design Reliability is a key concern. Experts initially estimate that the current design will work with probability of only 0.4. However there is a complex test that can be used to ALMOST validate or invalidate the design. This testing procedure is used in a wide range of situations. Looking back at past data over a wide range of initial success estimates If the design worked, how often were Test results GOOD almost validates the test results GOOD? P(Test results Good | Design Works) = 0.98 If the design failed, how often were Test Results BAD almost Invalidates the test results BAD? P(Test results Bad | Design Fails) = 0.94 Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 53 EVII - Bayes Rule & Decision Trees Chapter 11 Add an uncertain node at front of tree to represent uncertain outcome of test Use Bayes rule to calculate conditional probabilities. Use partition rule to calculate the probabilities of the test results. (These appear in the denominator of the Bayes Rule equation.) Green on the next page highlights the test result probabilities Yellow on the next page highlights the conditional probabilities. These vary because they depend upon the results of the tests. Chelst & Canbolat Value Added Decision Making 02/28/12 54 EVII - Bayes Rule & Decision Trees Activity: calculate conditional probabilities Data P(Design Works) = P(W)= 0.4 P(Design Fails) = P(F) = 0.6 P(Test Results Good | Design Works) = P(G|W)= 0.98 P(Test Results Bad | Design Fails)=P(B/F) =0.94 Activity: Use Bayes Rule to calculate P(Design Works | Test Results Good) = P(W|G)= ?? P(G) = ?? P(F/ G)= ?? Precision Tree Calculate Bayesian Probabilities by hand and Insert all of the initial probabilities upfront Insert conditional probabilities downstream. Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 55 Figure 11.12 EVII for Make/Buy: Test Design 183.38 – 181.38=2.0 EVII = $2M and EVPI =$2.4M Works Make FALSE 55 Good Works Buy + means collapsed node 8.4% 161 + + Demand 187.3 Demand 171.5 + Demand 197.23 + Demand 177.5 + Demand 187.3 Red Demand values are expected values Test Design 181.38 Works Make TRUE 55 Bad 57.2% 0 1.4% 100 Design 187.16 Fails 98.6% 108 Decision 187.16 Works Buy FALSE 0 Chelst & Canbolat Value Added Decision Making 1.4% 140 Design 196.87 Fails Chapter 11 91.6% 140 Design 173.66 Fails EVII Make-Buy 8.4% 108 Decision 173.6609 TRUE 0 + Demand 177.5 Design 178.32 Fails 42.8% 0 91.6% 100 98.6% 161 + + Demand 171.5 Demand 197.23 02/28/12 56 Conditional Decision If test results are GOOD then buy from supplier Less fear of 15% price increase If test results are BAD then make it yourself Concerned over supplier’s opportunity for significant price increase Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 57 Activity: Concrete examples of IMPERFECT Information Describe a context in which a decision can be made after gathering imperfect information and there is still related uncertainty. Product Development Example____________________________________________ Imperfect Information ________________________ Decision AFTER ____________________________ Updated future uncertainties _________________________ Can you quantify accuracy? _______________________ Manufacturing Example____________________________________________ Imperfect Information ________________________ Decision AFTER ____________________________ Updated future uncertainties _________________________ Can you quantify accuracy? _____________________________ Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 58 Sequential Decisions with Information in between Inability to predict future accurately Must make decisions under uncertainty A firm unable to determine level of demand in future or predict rival’s reaction’s Understate some perceived risks in order to obtain approval Can management delay high cost PART of decision until more knowledge is available Partial Investment Gain Information Broader scope of subsequent investment Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 59 Investment Decisions Three important characteristics of Investment Decisions Partially or completely irreversible Uncertainty over future rewards from the investment Assess the probabilities of alternative outcomes Leeway about timing of your investment Postpone action to get more information about the future How should a firm decide on an investing on a project or a new facility? Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 60 Real Options An option represents a “Right, but not an Obligation”, to do something under predefined arrangements Buy Option (expand or substitute) Put Option (contract or cut back) Flexibility to adapt in response to new information enhances the investment opportunity’s value by improving its upside potential An approach that offers a positive and radical reassessment of risk and exploration The opportunities to acquire real assetsReal Options Real Options term coined by Stewart Myers (1977) Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 61 Options Analysis Financial options Data Availability Precise models Technical options Data are less accurate One time decisions Estimates of values are approximates within bands described by sensitivity analysis Analytical niceties that might lead to greater precision might be a waste of effort To decide whether to do the R&D that will lead to a real option on a launch of a new product, managers only need to know if the value of option is greater than the cost to acquire it Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 62 Real Options Uncertainty Conventional Approach Minimize Risk React to uncertainties What is the best choice under the given circumstances? Work with predetermined set of decisions Real Options Proactive towards uncertainties Prepare plans to manage the risks Identify parts of the system that have most uncertainty, and try to see how these situations can be exploited Identify new possible paths: change decision tree by adding flexibility Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 63 Table 11.10 Common Real Options Option Defer Stage Alter Operating Scale Abandon Switch Explore Chapter 11 Description Project that can be postponed allows learning more about project outcomes before making a commitment. A multi-stage project whose construction involves a series of cost outlays could be delayed or killed in a midstream. A project whose operating scale can be expanded or contracted according to market conditions. Project can be abandoned permanently when market conditions are worsen severely and project resources could be sold or put to other more valuable uses. The project permits changing its output mix or producing the same outputs using different inputs in response to changes in the price of inputs and outputs. Start with a pilot or prototype project and follow-up with a full-scale project if the pilot or prototype succeeds. Chelst & Canbolat Value Added Decision Making Relevant Application Industries Real estate development, farming, paper products, offshore oil lease R&D intensive industry such as pharmaceuticals or other long development capital intensive projects Mining, facilities planning, fashion apparel, consumer goods Capital intensive industries (airline, railroad), new product introduction, financial services Any good sought in small batches or subject to volatile demand (e.g., consumer electronics, toys, machine parts) High production cost areas 02/28/12 64 Options Manufacturing and Product Examples Design all vehicles to facilitate pricey add-ons for specific market segments. (Vehicle personalization) Design a truck such that four-wheel steering is a later option that can be designed into it. Production system that can change easily Inputs: Dual fuel burners (oil and gas) Production lines designed to switch equipment so that they can produce different products Flexible machines – rapid tool changeover Modular Design: Option to upgrade a computer system Engines? _______________ Chapter 11 Labor Contract – pay premium for option to reduce workforce or close plants if necessary (Put Option) Chelst & Canbolat Value Added Decision Making 02/28/12 65 Remaining Text Examples and Figures EVII and Oil Drilling Technology Choice Schematic tree Decision tree Risk Profile Contingent Contract Negotiations – 2 perspectives Merck’s Chapter 11 options Chelst & Canbolat Value Added Decision Making 02/28/12 66 20.0% 700 Amount of Oil 64.5 80.0% Low 100 High Figure 11.13: Decision tree for oil drilling case with imperfect information TRUE -5 Drill 10.5% 0 Strong Seismic Test 57.1% -150 Oil TRUE -0.5 Drill Result 34.5 42.9% Dry 0 Decision 34.5 FALSE Don't Drill 0 0.045 -5.5 Test Results 4.23 TRUE -5 Drill 21.0% 0 Inconclusive 14.3% -150 Drill Result 4.5 85.7% Dry 0 Decision 4.5 FALSE Don't Drill 0 Drill Weak 68.5% 0 FALSE -5 Amount of Oil 64.5 + 0.18 -5.5 0 -0.5 1.5% -150 Oil Drill Result -4.5 98.5% Dry 0 Amount of Oil 64.5 + 0 -5.5 Decision -0.5 Don't Drill TRUE 0 0.685 -0.5 Seismic Test 4.225 20.0% 700 Amount of Oil 65.0 80.0% Low 100 High Oil Drill Don't Test Chapter 11 0.048 -55.5 0 -0.5 Oil Oil Drilling 0.012 544.5 FALSE 0 TRUE -5 10.0% -150 Drill Result 2.0 90.0% Dry 0 0 545 0 -55 0 -5 Decision 2.0 Chelst Don't Drill 0 &FALSE Canbolat 0 Value Added Decision Making0 02/28/12 67 Figure 11.15: Schematic tree for technology development example Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 68 Figure 11.16: Decision Tree for Omega case Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 69 Figure 11.17: Cumulative risk profile for technology development case - Omega 1 Cumulative Probability 0.8 0.6 Tech. Y & Prot. Tech. Z 0.4 0.2 0 15 20 25 30 35 40 45 50 Total Cost (Millions Dollars) Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 70 Figure 11.18: Contingent Contract Total sales from perspectives of Biotech and BSG a) BioTech perspective Chapter 11 b) BSG perspective Chelst & Canbolat Value Added Decision Making 02/28/12 71 Figure 11.19: Merck’s options and major uncertainties in Project Gama Chapter 11 Chelst & Canbolat Value Added Decision Making 02/28/12 72