1. Session 6a

2. Decision Models -- Prof. Juran2 Overview Multiple Objective Optimization Two Dimensions More than Two Dimensions Finance and HR Examples Efficient Frontier Pre-emptive Goal Programming Intro to Decision Analysis

3. Decision Models -- Prof. Juran3 Scenario Approach Revisited Use the scenario approach to determine the minimum- risk portfolio of these stocks that yields an expected return of at least 22%, without shorting.

4. Decision Models -- Prof. Juran4 Using the same notation as in the GMS case, the percent return on the portfolio is represented by the random variable R. In this model, x i is the proportion of the portfolio (i.e. a number between zero and one) allocated to investment i. (In the GMS case, we used thousands of dollars as the units.) Each investment i has a percent return under each scenario j, which we represent with the symbol r ij.

5. Decision Models -- Prof. Juran5

6. 6 The portfolio return under any scenario j is given by:

7. Decision Models -- Prof. Juran7 Let P j represent the probability of scenario j occurring. The expected value of R is given by: The standard deviation of R is given by:

8. Decision Models -- Prof. Juran8 In this model, each scenario is considered to have an equal probability of occurring, so we can simplify the two expressions:

9. Decision Models -- Prof. Juran9 Decision Variables We need to determine the proportion of our portfolio to invest in each of the five stocks. Objective Minimize risk. Constraints All of the money must be invested.(1) The expected return must be at least 22%.(2) No shorting.(3) Managerial Formulation

10. Decision Models -- Prof. Juran10 Mathematical Formulation Decision Variables x 1, x 2, x 3, x 4, and x 5 (corresponding to Ford, Lilly, Kellogg, Merck, and HP). Objective Minimize Z = Constraints (1) (2) For all i, x i ≥ 0(3)

11. Decision Models -- Prof. Juran11

12. Decision Models -- Prof. Juran12 The decision variables are in F2:J2. The objective function is in C3. Cell E2 keeps track of constraint (1). Cells C2 and C5 keep track of constraint (2). Constraint (3) can be handled by checking the “assume non-negative” box in the Solver Options.

13. Decision Models -- Prof. Juran13

14. Decision Models -- Prof. Juran14

15. Decision Models -- Prof. Juran15 Invest 17.3% in Ford, 42.6% in Lilly, 5.4% in Kellogg, 10.5% in Merck, and 24.1% in HP. The expected return will be 22%, and the standard deviation will be 12.8%. Conclusions

16. Decision Models -- Prof. Juran16 2. Show how the optimal portfolio changes as the required return varies.

17. Decision Models -- Prof. Juran17

18. Decision Models -- Prof. Juran18

19. Decision Models -- Prof. Juran19 3. Draw the efficient frontier for portfolios composed of these five stocks.

20. Decision Models -- Prof. Juran20

21. Decision Models -- Prof. Juran21 Repeat Part 2 with shorting allowed.

22. Decision Models -- Prof. Juran22

23. Decision Models -- Prof. Juran23

24. Decision Models -- Prof. Juran24 GMS Case Revisited Assuming that Torelli's goal is to minimize the standard deviation of the portfolio return, what is the optimal portfolio that invests all $10 million?

25. Decision Models -- Prof. Juran25 Formulation

26. Decision Models -- Prof. Juran26 Optimal Solution

27. Decision Models -- Prof. Juran27 Efficient Frontier for GMS

28. Decision Models -- Prof. Juran28 Parametric Approach Revisited (a) Determine the minimum-variance portfolio that attains an expected annual return of at least 0.12, with no shorting of stocks allowed. (b) Draw the efficient frontier for portfolios composed of these three stocks. (c) Determine the minimum-variance portfolio that attains an expected annual return of at least 0.12, with no shorting of stocks allowed. From Session 5a:

29. Decision Models -- Prof. Juran29 Formulation

30. Decision Models -- Prof. Juran30 Optimal Solution

31. Decision Models -- Prof. Juran31 SolverTable

32. Decision Models -- Prof. Juran32

33. Decision Models -- Prof. Juran33 SolverTable Output

34. Decision Models -- Prof. Juran34

35. Decision Models -- Prof. Juran35 Parametric Approach, cont. (c) Determine the minimum-variance portfolio that attains an expected annual return of at least 0.12, with shorting of stocks allowed. All we need to do here is remove the non-negativity constraint and re-run SolverTable.

36. Decision Models -- Prof. Juran36

37. Decision Models -- Prof. Juran37 Preemptive Goal Programming: Consulting Example The Touche Young accounting firm must complete three jobs during the next month. Job 1 will require 500 hours of work, job 2 will require 300 hours, and job 3 will require 100 hours. At present the firm consists of five partners, five senior employees, and five junior employees, each of whom can work up to 40 hours per month.

38. Decision Models -- Prof. Juran38 The dollar amount (per hour) that the company can bill depends on the type of accountant assigned to each job, as shown in the table below. (The "X" indicates that a junior employee does not have enough experience to work on job 1.)

39. Decision Models -- Prof. Juran39 All jobs must be completed. Touche Young has also set the following goals, listed in order of priority: Goal 1: Monthly billings should exceed $74,000. Goal 2: At most one partner should be hired. Goal 3: At most three senior employees should be hired. Goal 4: At most one junior employee should be hired.

40. Decision Models -- Prof. Juran40 Decision Variables There are three types of decisions here. First, we need to decide how many people to hire in each of the three employment categories. Second, we need to assign the available human resources (which depend on the first set of decisions) to the three jobs. Finally, since it is not apparent that we will be able to satisfy all of Touche Young’s goals, we need to decide which goals not to meet and by how much. Managerial Formulation

41. Decision Models -- Prof. Juran41 Objective In the long run we want to minimize any negative difference between actual results and each of the four goals. Of course, our optimization methods require that we only have one objective at a time, so we will use a variation of goal programming to solve the problem four times. The approach here will be to treat each of the goals as an objective until it is shown to be attainable, after which we will treat it as a constraint. For example, we will solve the model with the goal of minimizing any shortfall in the $74,000 revenue target. Once we find a solution that has no shortfall, we will solve the problem again, with an added constraint that the shortfall be zero.

42. Decision Models -- Prof. Juran42 Constraints The numbers of people hired must be integers.(1) Each project must receive its required number of man-hours. (2) Our model must take into account any difference between the actual performance of the plan and the four targets.(3) We can’t assign people to jobs unless we hire them.(4)

43. Decision Models -- Prof. Juran43 Mathematical Formulation

44. Decision Models -- Prof. Juran44 Decision Variables x i (three decisions), A ij (nine decisions), δ k (up to three decisions) Objective Minimize Z = Constraints All x i are integers.(1) A ij = R ij for all i, j.(2) δ Goals ≠ k = v k for all goals < k (3) for all i.(4)

45. Decision Models -- Prof. Juran45

46. Decision Models -- Prof. Juran46 The t k targets are in cells G10:G13. The δ k “negative difference” variables will be in B16:B19. We use the range C10:D13 to track all deviations (both positive and negative), and then refer to the “undesirable” one in B16:B19. For example, it is undesirable to have billings under 74,000, so B16 refers to C10. It is undesirable for the number of new partners to be over 1, so B17 refers to D11. The v k “best achieved” variables will be in D16:D19. The x i are in K7:K9. The A ij assignments are in B2:D4. The R ij requirements are in B7:D7. Cells G2:G4 keep track of constraint (4). We constrain B4 to be zero.

47. Decision Models -- Prof. Juran47 First Iteration At first we’ll ignore all of the goals except the billing target of $74,000. Decision Variables x i (three decisions, cells K7:K9), A ij (nine decisions, cells B2:D4) Objective Minimize Z = 1  (the shortfall, if any, between planned billings and $74,000) Constraints All x i are integers. (1) A ij = R for all i, j. (2) A 11 = 0. (4) i j ij xA 40 3 1    for all i. (5) Note that constraint (3) doesn’t matter in this iteration. Also note the balance equation constraint, forcing E10 = G10.

48. Decision Models -- Prof. Juran48

49. Decision Models -- Prof. Juran49 We have verified that it is feasible to have billings of $74,000.

50. Decision Models -- Prof. Juran50

51. Decision Models -- Prof. Juran51

52. Decision Models -- Prof. Juran52 Now we know that it is feasible to achieve both of the first two goals.

53. Decision Models -- Prof. Juran53 Third Iteration Meeting both the first and second goals will now be constraints, and we’ll focus on the third goal. Decision Variables x i (three decisions), A ij (nine decisions), δ k (two decisions) Objective Minimize Z = 3  (the number of senior employees hired above the goal of 3) Constraints All x i are integers. (1) A ij = R for all i, j. (2) δ Goals  k = v k (for goals 1 and 2) (3) A 11 = 0. (4) i j ij xA 40 3 1    for all i. (5) Note that v 1 = v 2 = 0 from the previous iteration.

54. Decision Models -- Prof. Juran54

55. Decision Models -- Prof. Juran55 All of the first three goals are feasible.

56. Decision Models -- Prof. Juran56 Fourth Iteration Meeting all of the first three goals will now be constraints, and we’ll focus on the fourth goal. Decision Variables x i (three decisions), A ij (nine decisions), δ k (for the first three goals) Objective Minimize Z = 4  (the number of new juniors hired above the goal of 1) Constraints All x i are integers. (1) A ij = R for all i, j. (2) δ Goals  k = v k for goals 1, 2, 3 (3) A 11 = 0. (4) i j ij xA 40 3 1    for all i. (5) Note that v 1 = v 2 = v 3 = 0 from the previous iteration.

57. Decision Models -- Prof. Juran57

58. Decision Models -- Prof. Juran58

59. Decision Models -- Prof. Juran59

60. Decision Models -- Prof. Juran60

61. Decision Models -- Prof. Juran61 There are two constraints that don’t show in the window: $E$2:$E$4<=$G$2:$G$4, and $K$7:$K$9 = integer.

62. Decision Models -- Prof. Juran62 It would appear that it is infeasible to reach all four of the goals. However, the first three can be met while delivering all of the jobs as required. The only goal that can’t be attained is the limit of hiring only one new junior-level employee. The best solution we found requires us to hire four new employees at this level. It is possible to have revenues of $91,000 under this hiring plan. Conclusions

63. Decision Models -- Prof. Juran63

64. Decision Models -- Prof. Juran64

65. Decision Models -- Prof. Juran65

66. Decision Models -- Prof. Juran66

67. Decision Models -- Prof. Juran67

68. Decision Models -- Prof. Juran68

69. Decision Models -- Prof. Juran69

70. Decision Models -- Prof. Juran70 Summary Multiple Objective Optimization Two Dimensions More than Two Dimensions Finance and HR Examples Efficient Frontier Pre-emptive Goal Programming Intro to Decision Analysis